aiLL 7/10/23 - GPT is an Adventure, Not a Threat

Transcript

1. 0:05foreign
2. 0:25this is not showing me how many people are here this is weird I don't know what's going on man tick tock's being weird tonight oh well
3. 0:39I will be clueless as to who's here wait maybe I can do it with this no weird strange
4. 0:54hey simplex I'm here but also in The Matrix I know retro Punk aren't we all how's agency life welcome everyone my name is Kyle Shannon
5. 1:08they call me Hamilton um this is the AI learning lab I have no idea how many people are here
6. 1:23um happy Monday you got any thoughts questions about AI pop them in the comments below I'll do my best to answer them
7. 1:32we'll get oh now they're now now I see some people there there's uh 16 of us hate Tobias which was shaking
8. 1:4021 so far okay it's just it's being really slow to update on my side so feel free double tap the screen get the
9. 1:46hearts going and then if you'd be willing to share the live and maybe maybe we'll get in a pile 42 all right
10. 1:53it's it's climbing I'm only seeing 16 still so whatever whatever I mean you know with these platforms like they don't charge
11. 2:04us anything so what do you want for free if the product's free you're the product [Laughter] oh good lordy welcome to Monday are you
12. 2:18always excited as I am for Monday I had a crappy day at work today not not like nothing bad happened I just got there I
13. 2:24like stared into the middle distance for most of the day just like nothing it's just like [ __ ] flatlined all day
14. 2:3683 oh thanks thanks for keeping me posted on how many folks are in here uh this is the AI learning lab I'll show
15. 2:42you these URLs in a second here um my name is Kyle Shannon I'm an entrepreneur of many many years many decades actually
16. 2:50at this point I'm an old [ __ ] and uh I swear occasionally so if you don't like that I wouldn't stay uh I can
17. 2:57be a little smarmy I try to answer any questions I can about uh about AI so pop them in the questions below I am not an
18. 3:06engineer or a machine learning guy I'm more like an entrepreneur that deals with innovates with technology once it
19. 3:13gets ready for prime time and so I kind of Mark the launch of chat EPT is a new new uh era a new inflection point it's it's
20. 3:22actually very similar I was uh I was an entrepreneur back in the mid 90s when the World Wide Web first came out and
21. 3:28it's there's a lot of parallels between the early days of the web and the early days of this generative AI stuff so
22. 3:34that's why I'm so excited about it that's why I started the channel so if you have questions about it pop them in
23. 3:39the in the comments below I'll do my best to answer them oh let me let me show you so why do I have these on screen I'm not
24. 3:48affiliated with any of these URLs or these companies in any way but if you have not played with chat GPT go
25. 3:56to that top URL that's the official chat GPT website chat.openai.com go there and start playing if you if you
26. 4:03have zero idea what to do if you don't know what it is it kind of looks like a Google search bar once you sign up and get an account
27. 4:09but it's so much it's it's so profoundly deeper than a Google search um so if you want to know how to talk to it go to that URL right there
28. 4:19prompts.chat that's a little primer that we'll talk to you about what prompting is what large language models like chat GPT are
29. 4:29um or the underlying model is is a large language model chat GPT is the app sitting on top of it um but go start playing uh and and you owe
30. 4:38it to yourself if you haven't played with chat gbt is play with it long enough until you have your Kevin
31. 4:42McAllister moment which is this moment go see if you can find that moment for yourself then bing.com is also chat GPT
32. 4:52if you click on the chat button there Bing chat is gpt4 connected to the internet you can also tell it to make
33. 4:58you images and it uses the Dali model to make you images and then poe.com gives you access to six different large language models
34. 5:07um that I think are at least worth worth playing with you can compare them there which is interesting there's one from
35. 5:13quora from Poe there's two from open Ai and there's three from anthropic the quad models are there so that's an
36. 5:20interesting site and then futurepedia.io that's a big directory of um what are they called uh AI apps so go there and do that
37. 5:33all right let's get to some questions so what will we be learning about wait what were we learning how about
38. 5:44ethics oh what will we be learning how about ethics um well I I guess what do you want to know about ethics like like in what context
39. 5:55copyright um alignment um politics I'm not going to talk about that ethics and politics doesn't exist anymore so we're good there we're
40. 6:06covered it's non-existent oh does AI have ethics or or do they need a a comedy I I assume you mean the community
41. 6:21um does AI have ethics well if you you should go listen to Sam Altman talk about you know the dangers
42. 6:34of AI and the benefits of AI and things like that he's a very thoughtful guy um the open AI model called human reinforcement learning um
43. 6:47is involves humans so there are humans looking at things going that's good that's bad that's how they train it
44. 6:55um there are flaws with that so even though there may be ethics there's also bias because humans are involved in the
45. 7:02data that these things are trained on have biases in them so so there's a lot of issues with that one
46. 7:09of the reasons I like anthropic which is you can play with the anthropic models at poe.com is that they use something
47. 7:15called the Constitutional learning model where there they create a set of rules like you know increased Prosperity don't
48. 7:23harm people things like that and then the the uh training is done against that Constitution so it's not
49. 7:31it's not a human Constitution I think that approach is ultimately really interesting when it comes to alignment
50. 7:37or aligning the goals of the robot with the goals of the people because we want those two goals aligned
51. 7:46and if they're not that's bad because they could just decide they don't need us anymore if they get smart enough
52. 7:51they're not there yet but holy crap is it coming Microsoft just put out a paper yesterday or the day before about a
53. 7:59billion a billion token context window which is just insanity insanity uh so so these things are going to get smart very
54. 8:14very fast like like way smarter than they are now do you prefer a key do you prefer a keyboard plus ai.com domain oh oh keyword Plus ai.com
55. 8:30or keyword dot AI I am incredibly frustrated right now with the domains of every company in web 3 and AI right now because
56. 8:45the the web 3 crypto Bros started it with the dot XYZ and the dot this and the dot dot I mean we've had alternative
57. 8:53domains you know I started one of the first agencies in the mid 90s there have been alternative domains for a while but
58. 8:58they've never caught on and all of a sudden in the past I don't know three or four years like I can't I
59. 9:06can't ever remember any sites is it.com is it dot AI is it dot IO is it dot XYZ is it dot bubble butt like uh it's just [ __ ] exhausting
60. 9:19so like like whatever anyone asks me you know something I can always remember remember the keyword the name of it I
61. 9:27can never remember the uh the domain so I'm an old guy so you know I'm like get off my lawn with you Daddy eyes you know
62. 9:36I got my company's called storybind I did get storyvine dot AI because we're doing some stuff in the space and I
63. 9:43thought it might be important we might need to distinguish it from the.com um but I suppose I'm a traditionalist I
64. 9:51just can't remember all the alternative domains I mean it's fine if you're clicking on it right if you have it as a
65. 9:57hyperlink somewhere it doesn't really [ __ ] matter what it's called but if you expect people to remember your your url
66. 10:08dot com's the way to go which company would you recommend applying to if you want an entry level if you want to do entry level AI
67. 10:19language model do you want to work on the on the engineering side or do you want to work on the applying large language
68. 10:28models to a business side that's there's a big distinction there because if I think if it's the latter um
69. 10:38I I'd say it's it's not so much the type of company but more like the type of attitude you want to find someone that's
70. 10:44embracing AI you don't want to go in and be a salmon swimming up the river if you're looking for a new place you might
71. 10:50as well go to some place committed to it if you're on the other side I don't know open AI seems like they're hiring
72. 10:58um inflection just raised 1.3 billion dollars they've got that new pie Pi agent they look really interesting that's an X Deep Mind founder
73. 11:12um that created inflection and then um anthropic I think is really interesting um stability AI I think is really
74. 11:20interesting if you want to work on the more on the research and Science and you know building the products rather than using the products
75. 11:57oh do they need a committee for ethics so oh the question about ethics do they need a committee um yeah there's some interesting things in
76. 12:07here um yes they they need an Ethics Committee they need a Safety Committee they need a human alignment committee um
77. 12:15yes they do they need a regulatory committee so so and I mean listen the good news the good news is this the
78. 12:26companies that are big enough to have like the big models open AI anthropic inflection Google meta they're all big enough to have
79. 12:39you know ethics boards and things like that you might be cynical about whether or not they're going to be ethical
80. 12:44that's a whole different conversation what's what's a bit scarier right now is because meta released their
81. 12:52foundational models out into the wild there's now all these open source projects that don't have any safety guard rails at all um so
82. 13:05the the good news right now is that those models are the open source models are not as powerful as gpt4 but they're
83. 13:13getting better all the time like you know on a on a every three day basis they're improving dramatically so yeah pretty interesting stuff um
84. 13:25Sarah Silverman is suing open Ai and meta for copyright infringement like I said the next 20 years are going to be a
85. 13:32never-ending stream of lawsuits at some point someone's gonna make precedent one way or the other and you know
86. 13:40Sarah will win that lawsuit and after 38 appeals they'll be like okay you have to shut down the AI they won't at least there's already open
87. 13:49source like like at this point it's I I think what these lawsuits ultimately will do is sort of figure out where the
88. 13:55boundaries are and you know similar to how streaming when streaming services came out everyone freaked out
89. 14:04and said no one's going to get paid No One's Gonna Get Paid there are all sorts of lawsuits and then they came to some
90. 14:09relatively shitty compensation model for musicians um I think in the world of AI that compensated compensation model gets even
91. 14:19shittier because we're entering a world of infinite content creation right and almost all of the um copyright monetization stuff is
92. 14:33happening on the usage side on the output side and if you're entering a world where fan art may sound as good if not better
93. 14:42as the original artists work but it's going to be derived from them we might need to find compensation and
94. 14:49attribution models that happen on the generation side not on the usage side but we're not anywhere close to that
95. 14:56right now right so so we've we're gonna have to go through four or five or ten years of just everyone's suing everyone
96. 15:04and then at some point realizing well we're not going to uninvent this [ __ ] and uh and and we've got to figure out a way to
97. 15:13to work forward so that artists can get compensated and businesses can stay in business so I don't have any kind of a clue what uh
98. 15:24how it's gonna play out other than just expect more and more lawsuits I'm a big fan of Sarah Silverman and it'll be
99. 15:32interesting to see what happens I think high profile lawsuits like that are going to make a difference and they're
100. 15:36going to establish some sort of precedent I think we'll see I'll we'll see some on the copyright side as well
101. 15:43um at some point some famous person is gonna try to copyright something where AI was used to generate a lot of it in the
102. 15:52copyright office is going to shut that [ __ ] down and that's gonna I don't know we're gonna see some interesting stuff
103. 15:59hello Albert how are you I wonder how long it'll take for these lawsuits to settle litigate like the the here's the
104. 16:06thing that I've learned about lawyers I was I was uh I was part of a public company my the agency I started in the
105. 16:13mid 90s we went public and when the.com Bubble Burst and our stock price Tanked I was part of four class action lawsuits that named me
106. 16:25personally for hundreds of millions of dollars and every single one of those class action lawsuits settled for wait for it
107. 16:40the exact amount of the lawyer bill four out of four class action lawsuits settled for the amount of the lawyer bill
108. 16:51so lawyers don't give a [ __ ] who they sue or what they do or if they win it's nice if they win but even if they lose
109. 16:58they win right because those those settle those they were frivolous lawsuits and the lawyers made a [ __ ] fortune
110. 17:07and that's that's what's the this is just this is you know the lawyers are out pre-ordering their Yachts right now because of AI
111. 17:18so I think it's going to take a while they have no incentive to make it to make the lawsuits any less or any
112. 17:25shorter because the longer they are the more more money they make someone said.com is king.com gen xers.com is the only way baby
113. 17:39main names if you can't spell.com get off my goddamn lawn how learn AI have to go to college no go there right now
114. 17:54chat.openai.com go to prompts.chat start learning so okay two things there's an inflection point before November 30th 2022 if you want to
115. 18:08get into machine learning and AI probably good to go to college you need to be a scientist a mathematician a researcher something like that
116. 18:16post chat GPT there's now I don't know I'm guessing 200 300 million people who do not have engineering degrees that
117. 18:26are using these tools in their business in their personal lives to start businesses you can learn all this stuff without
118. 18:34knowing anything about mathematics does having some concept of technology help absolutely because we're very early
119. 18:41these tools are very very sketchy and janky and if you want to do any of the advanced stuff you might have to do some
120. 18:49geeky stuff like fire up a Google collab create a virtual machine and go make yourself some Lang chain or whatever
121. 18:58so it doesn't hurt you to have some technical background but you don't need it go to chat.openai.com go to
122. 19:04prompts.chat and that's why I have these on screen for my entire lives is go start playing right now start
123. 19:12getting AI literate start understanding what these tools make possible because they are insane what does that mean billion context
124. 19:20window what it means is you could put 750 million words in a prompt to put that in context um that's more words than an average human
125. 19:37will read in their lifetime and it basically means you can put in you could put in massive amounts of data essentially into the um
126. 19:52into your chat session right so right now you've got a 3 000 word context window so you know how when you're
127. 19:57chatting with chat CBT if you do it for a long time sometimes it just forgets what the hell you were talking about
128. 20:03you're like but I gave you my whole life story up there and now you don't seem to know who I am that's because the life
129. 20:10story slid out of that context window it's only three thousand words 750 million words you can start you
130. 20:17could put in massive amounts of data and still have massive amounts of data you could you know do back and forth and it
131. 20:24would have sort of this incredibly long memory um they're saying with the approach that they came up with this is a bunch of
132. 20:33Microsoft researchers they could get it into the trillions which basically means like what what I saw a video on it today
133. 20:40and basically what it means is if they get it into the trillions you can put the entire internet into essentially into active memory so
134. 20:51you could be chatting with the entire internet as you're sort of known knowledge part of the memory it's it's insane like
135. 21:01so here's what it basically means it don't worry about the numbers the numbers don't mean anything what it basically means is
136. 21:09if we're jumping from like right now at poe.com you can play with the Claude 100K token model if we're jumping from a hundred thousand
137. 21:19tokens to a billion tokens the speed with which we get to AGI starts to look more like 2024 than 2029.
138. 21:30AGI is artificial general intelligence the point at which these machines can reason become smarter than the smartest human
139. 21:40starts to look like a year year and a half rather than three to five so that that's Insanity it's it's like
140. 21:51and then and then once that happens once these machines get that smart then they start to accelerate themselves and you
141. 22:00know and the the improvements keep accelerating and this is why folks are talking so much about alignment right
142. 22:06now and openai mentioned yesterday or the day before I know it's the day before um that they're they're committing 20
143. 22:13percent of their compute resources to Super alignment is what they're calling it which basically just means let's make
144. 22:19sure these things have the same goals as humans so we're not at war with them because they're going to get way smarter than us
145. 22:27I think a lot faster than any of us are prepared for yeah so crazy crazy so ethics covers a broad spectrum of stuff exactly like I
146. 22:37don't know quite what you're talking about there but for me it's about it's about alignment the ethics thing I mean
147. 22:43ethics isn't you know some of the like bias at least is in the eye of the beholder different people have different
148. 22:50biases and they're gonna I think what we'll have is we'll have different models but the ethics and integrity and
149. 22:56Alignment thing is super important depends on who's doing the AI training I kind of think of it similar to dog
150. 23:02training that's probably not a bad metaphor actually so I was on Discord read the reason for Bing was the paint wall issue
151. 23:10think they will bring it back yeah I do think they'll bring it back so so here's here's how I understand it
152. 23:17if you're even remotely geeky when when you land on a a site like New York Times or Fortune Magazine it was Fortune
153. 23:26Magazine was was the issue apparently and it pops up that thing that says you can't see this site there's kind of two
154. 23:34flavors of paywalls one of them if you view the source code of the site if they have a shitty paywall you can
155. 23:42actually go read the article you just go look at the you just go look at the source code of the site you know inspect
156. 23:47and then you can go read it and go copy it um then there's other paywalls that you can't you've actually gotta have a a
157. 23:54token to to see the article um I I think it was that the former kind of paywall where you could just essentially inspect
158. 24:03the uh the code of a site and that's that's what it was pulling out why it's been gone for it's been almost
159. 24:11a week now that that that feature hasn't been there let's see if they brought it back I kind of check it once a day just to see if if they
160. 24:19brought it back now wait let me reload let me reload a chat GPT you got my money
161. 24:33nope it's not there I don't know why it's taking so long I mean what's what's weird is um plug-ins still work and they're
162. 24:41connected to the internet and bing.com is still connected to the internet like that's still functional the chat there
163. 24:47is still functional so this was clearly something about open ai's implementation of browse with Bing was doing something that obviously
164. 24:58broke something and it's more complicated to fix than I thought it would be but I don't know because you'd think
165. 25:09I I don't know maybe maybe they got to build a whole white listing kind of application that if if you know companies are on that list
166. 25:19then it it won't scrape it or something like that I like I honestly have no idea I don't know why it's taken so long
167. 25:27um.com is King yes applying large language models side oh applying large language model side okay so then what I would do Special K
168. 25:36Special K 353 you're special um so what I would do is um I just start I
169. 25:51I would get as AI literate as you can as soon as you can so you know go to these tools um understand the different image
170. 26:00generation models mid-journey stable diffusion Adobe Firefly generative fill within Photoshop you don't have to be
171. 26:07Masters at those but just at least understand that world understand you know as deeply as you can what generative AI
172. 26:19tools like chat EBT make possible um you know understand that they can write code and they can write copy and that
173. 26:26they can make tables and that they can help you do research and that they hallucinate and you know all that sort
174. 26:32of stuff so if you're if you're already decently AI literate like that great the the next thing I would do is start
175. 26:41putting it out there right so you don't need to start a tick tock Channel like I did because that's [ __ ] insanity
176. 26:48but but you know post on LinkedIn hey I just did this experiment with chat apt um join meetups or join you know AI
177. 26:58communities put that on your LinkedIn um if you can I work for this group and we did this cool content generation tool
178. 27:06and the guy that runs the group said hey do you mind if I give you the title of Chief generative officer and I'm like
179. 27:13sure why not I don't care and he said I'm pretty sure it's the first Chief generative officer in the world I'm like
180. 27:18awesome I get to claim I had that first um so whatever what does it mean it doesn't [ __ ] mean anything but
181. 27:26when I put that there people started reaching out to me so put yourself out there um talk about this stuff with your
182. 27:34friends and family even if they think you're weird again just because we're we're very quickly and it's I'm starting
183. 27:40to see it happening we're very quickly moving into a mode where companies are gonna say we gotta figure this out this
184. 27:50is urgent like someone's gonna get religion in one of these companies and it won't be religion like they
185. 27:55understand this stuff it's going to be religion like holy [ __ ] we got to do something who knows anything about this
186. 28:01and if they're surfing LinkedIn and they see that you know about it they might click on your link um and then the other thing I would do
187. 28:07is as you start to um get better and better at that you can put yourself out there and do education you could do workshops for companies
188. 28:16um and then from that maybe that turns into a job um I would just be very very reticent of any company that's like well we want you
189. 28:26to do AI but we're we're you know we're taking a wait and see attitude with it anyone who's not actively going
190. 28:34we're trying to disrupt ourselves here this is this stuff is so profound it puts everything we do at risk and we're
191. 28:41going to figure it out before we get you know bowled over by it um I would I would actively seek out companies that that are on the more
192. 28:49aggressive side of dealing with this stuff will AI make humans obsolete if you go listen to the uh Lex Friedman interview with George hotz
193. 28:59George hot certainly thinks so he he talks about the incremental value of humans is is getting increasingly niche as he put it um
194. 29:10I don't know I I think that um where I'm hoping it lands is that these machines can do a lot of the
195. 29:24low-level repetitive um non like the you know the soul-crushing part of your day like depending on your job
196. 29:35it might be like a lot of the day but the part of your job that's just like oh I gotta go write those 10 emails
197. 29:43or I gotta go deal with the insurance company or you know I just write you know iterations of the same copy that someone
198. 29:52else came up with and I'm like just a cog in a machine I think that these AI tools do that [ __ ] really well so my hope is that all of the
199. 30:03complicated repetitive low-level soul-crushing work of the world will be done by these things and then that frees us as humans to either
200. 30:14operate at a higher level or work Less hours or do less for the same amount of productivity output so so we see it we see a gap where
201. 30:27what's required to do the work is lower than than doing the work what fills that Gap I don't [ __ ] know
202. 30:36AI is going to generate a bunch of new jobs so there might be new jobs that fill that Gap I Now understand why Andrew Yang was
203. 30:44talking about Universal basic income because that Gap could look like there's a bunch of unemployed people and you
204. 30:50might need to fill that Gap with some Universal basic income until the new jobs come but that Gap could also look like
205. 30:58now there's a bunch of people that were doing kind of low level Cog in the machine work that are now going to say
206. 31:06you know what I've always wanted to do write the Great American novel start my own company start a flower shop you know
207. 31:12I think I might like to be a plumber that doesn't [ __ ] matter so I think we're going to see a lot of personal reinvention and a lot of
208. 31:21innovation but I think it's also going to be painful I think as as the AI drives up productivity and capability
209. 31:32and we don't need as many people I think that Gap is going to look like a lot of displaced people for the next two years three years maybe
210. 31:42there's a new show on Netflix called unknown Killer Robots what happens when machines make life or the documentary will explore the dangers
211. 31:52of artificial intelligence in military application yep yeah like like what Boston Dynamics has said is you know if if you build robots just
212. 32:02rule number one don't put guns on them you know the military's [ __ ] putting guns on them you know that right you know that right so
213. 32:19this gets back to the alignment thing right super alignment you know are the goals of these things aligned with human rights
214. 32:30and if they are good but how do you do that and when you have a world where you've got open source large language models
215. 32:38which we do now and they don't have any alignment and they don't have any corporate overlords and they don't have any regulation
216. 32:45what happens then do you make any AI illegal unless it's through one of the regulated companies I don't think that's going to fly
217. 32:54because that's not going to happen globally but it could happen state by state it could happen even Nation by Nation but
218. 33:01then your neighbors just gonna do stuff and they're going to innovate I don't know it's wild what's that new it's called unknown
219. 33:11Killer Robots see that's the other thing right now is the fear-mongering with this stuff is is kind of stupid we've had 50 years of
220. 33:21Hollywood telling us the robots are going to kill us so I think there's a predisposition for people to um
221. 33:29fear them because change is hard anyway and when you've got things that are this profoundly powerful that the answer to if I'm a copywriter
222. 33:42like is my job the same as it was before it's not and it never will be when you have things that that produce kind of
223. 33:48instant existential questions for people's livelihood that sort of compounds the fear of Hollywood telling us the robots are
224. 33:57going to kill us so I like I hate the fact that they're calling it unknown Killer Robots but whatever
225. 34:05that's why that's why I do this channel because I think it's just too easy to give into that fear it's really
226. 34:10sensationalistic I think it is it is as likely that these things profoundly improve life as as much as they profoundly destroy
227. 34:18life but we'll see it will certainly outlive URI well it might so will our children and grandchildren um
228. 34:33it won't if um the internet and electricity doesn't exist anymore if the infrastructure goes bye-bye then AI doesn't survive it although
229. 34:47to that point well no I mean you need functional computers for this stuff to work so so if they drop some uh
230. 34:55electromagnetic pulse weapons then then you know that's a that's that's the off button for this thing is computer seats
231. 35:03working but you know then then we're then we're into the dystopian future that we all saw in the movies
232. 35:10and I hope you're good at gardening and defending yourself [Laughter] um I want we learn we learned for my real estate company
233. 35:23you mean AI learning for your real estate company what do you mean when they say chat GPT hallucinates Okay so I also like to describe it as
234. 35:34mansplaining as a service so when you ask chat GPT um to give you an answer it will give you an answer
235. 35:46and then if you're like I'm pretty sure that answer is wrong it's like oh I'm sorry here's the right answer and then
236. 35:52that answer might be wrong and then you're like no that's still wrong it's like that's my apologies this is
237. 35:57embarrassing here's the real answer and then that answer is wrong that's hallucinating the way the technology works
238. 36:07is you know when you're in Google and you're typing a sentence and like the the last three words of the sentence the autocomplete shows up
239. 36:16that's all these tools are except it's like that on steroids right so it writes you A big one dissertation of you know
240. 36:25some research that you asked it to do and then it provides citations of here's the links to the articles that I used to
241. 36:32come to this opinion and then you click on those links and they're all dead links when it's writing all of these words
242. 36:39it's doing them one word at a time just as a probability engine it doesn't actually go back and look at what it
243. 36:46wrote and go is that correct now there's a lot of companies working on that piece of it but it's not there right now so
244. 36:55just the nature of how it works it is just this it's it's got the knowledge of what it does as much as a calculator has knowledge of
245. 37:06what it does right we just trust that calculators have the right formulas in them when we put in two plus two it
246. 37:12seems to come out consistently of four if calculators when you put in two plus two came out with some random number we
247. 37:20wouldn't trust them as much right so that's where these things are right now is it's like we've built the fancy word
248. 37:26calculator that two plus two is most of the times four but occasionally not and it's up to you to deal with that now
249. 37:38there are times when that's completely irrelevant right if I'm working on a screenplay or some creative writing or
250. 37:45writing lyrics for a song or social media copy whatever it is I might not need that but if I'm doing a blog post that makes a claim and that
251. 37:56claim has got to be backed up by some research and that research is in there well that's then on you to make sure
252. 38:02that that research is accurate there was about two weeks ago there was a lawyer that had chat EPT write a legal opinion
253. 38:09for him and he handed it to a judge and all of this the the case law that was cited in his opinion didn't exist
254. 38:22and the judge called him out on it and said what the [ __ ] is this and the lawyer was like I wrote it in chat GPT
255. 38:30that's on the lawyer in my opinion you know these things do this right now like you know a year or two from now we won't
256. 38:37have to worry about this but right now these things are they hallucinate um so if you're a lawyer check your work [Laughter]
257. 38:49all right is there an easy way to avoid plagiarism of chat CPT writes an article for me besides rewriting it all so it's
258. 38:55not plagiarism like I said the way it works is that when you write your prompt it generates essentially a mathematical probability
259. 39:07that it then looks at all of the words it knows and all of the words it knows is a lot of them it's so so the the core
260. 39:15large language model sitting underneath um Chachi BT it was trained on all of the public internet and so what it's not doing is it's not
261. 39:26copying and pasting from specific pieces of work and pasting that into your document what it is doing is saying based on the
262. 39:34prompt that you wrote the mathematical probability that the next word is the is 99.999 so it'll type the word the and
263. 39:44then it takes what you wrote and the word the and it creates the probability for the next word and it just does that
264. 39:49over and over and over again so it's looking at all of the stuff it knows and whatever has been kind of filtered out
265. 39:55depending on your prompt right so that's why if you tell chat EBT I want you to act like a lawyer that went to Harvard
266. 40:02Law School you get better answers than if you just say give me a legal opinion because they give me a legal opinion
267. 40:09looks at all of the internet when you say I want you to ask like a Harvard law professor it's going to look at you know
268. 40:16the the words that are kind of in that semantic cloud um so so it'll get it'll get better um so it's not plagiarizing
269. 40:27what you might be talking about is can it be detected that it was written by AI that's where you might want to have a
270. 40:36rewriter rewrite it if you're if you're at a university where they're using AI detectors you go find a Ai rewriter and
271. 40:46it will rewrite it in a way and apparently gpt4 uh doesn't get caught by the detectors a lot um I think this whole idea of detecting
272. 40:55whether something is a AI or not is going to be very short-lived because these tools are going to be incorporated
273. 41:01in into like Microsoft Word Google Docs note taking apps it's going to be in everything and and the the way that when you start working
274. 41:10with this stuff on a regular basis it's very rare that I'll take something out of this machine and use that wholesale
275. 41:17what I'll often do is I'll say um I kind of have an idea for a piece or something I need to do and I'll use that
276. 41:24I'm like give me a first draft get me to 80 and then I kind of take it from there and then if I get stuck in there I'll
277. 41:29say oh let me take this paragraph and have it rewrite it or give me an example and it'll do that and I'll put it back
278. 41:34in so almost all of what I write is some sort of hybrid of Ai and me and I think increasingly that's how it's going to be so I don't
279. 41:44think so so a I don't think you need to worry about the plagiarism piece if you're talking about AI detectors that
280. 41:50that you'll need to do some research on depending on the organization you work for how does AI learn to use it for my
281. 41:57real estate comp you mean compensation or computer if it's compensation so I would so go to prompts.chat go to that thing
282. 42:09read about prompting and how prompting works and then look at so what that document is is it's a whole list of act
283. 42:16like a so you could say act like a real estate you know specialist that understands real estate compensation or
284. 42:24I I don't know what comp means I assume it means compensation but if you mean computer basically tell it to act like a
285. 42:31real estate expert and then start asking it questions for what you need there's also if you pay for chat GPT it's 20 bucks a month for the
286. 42:42plus version you get this new thing called code interpreter which holy crap code interpreters insane you can upload up to 100 megabyte file
287. 42:57to code interpreter and then start interacting with it so you could put up a hundred megabytes worth of real estate data
288. 43:06upload it to this thing and start interacting with it I don't know what you're looking for but like find me oh
289. 43:11con I assume you maybe you mean comps comparisons yeah so you could say take data of here's all the houses on
290. 43:19the market find me any comps in neighborhoods that are similar to this one and it will go find them um so check out code interpreter I can
291. 43:27show it to you here in a bit remember VCRs no one was going to be allowed to record television shows lawyers yeah baby yeah exactly exactly
292. 43:36it's this is the thing that's amazing about every time these new technologies hit the market it's identical every time
293. 43:48the established professions are like this is awful it's going to replace everyone everyone's gonna die and then
294. 43:57and then the lawyers are like we'll sue them we'll block it we'll stop innovation well uninvent it and it never happens it's the same every
295. 44:08single time now the speed with which this is coming and the depth of impact of this is is is going to eclipse probably anything we've
296. 44:20ever seen shy of things like fire printing press um steam engine it's big like that um so we ain't stopping it
297. 44:34but it's going to be messy for a while because just fear and disruption and and things like that so why this channel
298. 44:41exists I want everyone here to get curious about this stuff if you haven't played with it yet if you're sitting on
299. 44:46the sidelines like well I heard it's not all that accurate or yeah you know the robots are gonna kill us I'll just give
300. 44:51it three years and see what happens don't do that don't be that person if you go learn this stuff and then
301. 44:59decide I hate it and I'm gonna disengage from it more power to you but don't just sit on the sidelines and
302. 45:05assume that you've got time because I don't think I don't think we do we don't have time it's coming very very fast very very fast
303. 45:15um and I don't know that we're prepared for it remember microwaves I could have been rich I know I've got I got a bunch of
304. 45:23those if you're on the planet long enough you know you've got one of those Oh I thought of that three years before they did yeah here we are
305. 45:32I mean that's the other scary thing about this technology is it's going to be trivial to um replicate a piece of software trivial
306. 45:43so someone comes up with something novel and interesting anyone's gonna be able to go give me an app that does that oh and by the way
307. 45:50make it look like that too and come up with an intro Ninja for it and change change everything so the words are different
308. 45:57but it does exactly the same thing so that's going to be happening as well are we going to get sued for writing books
309. 46:06with the help of AI I don't think you're gonna get sued for it I think if you want to copyright it you're going to
310. 46:11need to prove that your contribution to it was significant enough that it's copyrightable um but that's even up in the air um
311. 46:22Sarah Silverman just brought a multi-million dollar lawsuit to open AI saying they trained um chat gbt on
312. 46:31um her written material and right so if she wins that then everyone who's ever written anything is going to do the same
313. 46:38kind of lawsuit so I don't think it's gonna it this is gonna be this is gonna be not dissimilar to what happened in the 90s
314. 46:49with sampling music and Hip-Hop but the difference here is in the 90s when you're sampling music you were you were literally copying and
315. 46:58pasting the Beats from someone else's song and putting that in your song this generative AI stuff is not doing that it's it's
316. 47:08the the metaphor is closer to this machine went to college and it learned all the stuff so if you went to college and you learned all about
317. 47:18a particular kind of writer right and then you went out into the world and you started writing and you were so
318. 47:25inspired by that writer you kind of wrote like that writer well that's still your writing it's just you're taking
319. 47:30your knowledge of all the stuff you learned and the filter that happens to sound right to you is writing that sounds like
320. 47:37they're writing that's not their writing um so so it's more like that um but copyright holders are still going to sue
321. 47:50they're going to say Hey you trained on my stuff you're making money on that but then I would go to what percentage
322. 47:57of the words on all of the public internet are Sarah Silverman's it's probably a very small percentage and
323. 48:04then what percentage of that ended up in the thing that you wrote where you said write me a joke about something
324. 48:12it's not copying and pasting her joke and putting it in there if it were doing that then I'd say there's a much bigger issue
325. 48:18but she's just part of the contributing Corpus of knowledge that when you say write me a joke it's probably pulling as much from Jim
326. 48:27Gaffigan as it is from Sarah Silverman she would probably be offended by that which is cool Taylor Swift is going to
327. 48:35have a lot of fodder for songs AI plagiarism yeah well and there's so so there's really interesting stuff there's uh Elon musk's
328. 48:46ex-wife or girlfriend um um she said she said sample my voice all you want just give me 50 percent so I think some people are gonna gonna
329. 48:59go the route of I'm gonna sue your ass and other people are gonna go the route of we're not stopping this thing so just
330. 49:06use it and compensate me fairly and I think in the end that's where we'll end up is people create models to say okay
331. 49:14here's the official Taylor Swift model where if you want to create songs that sound like Taylor Swift you click on this link and then every
332. 49:24song that you generate Taylor gets a nickel and if your song You Know starts to make money on streaming services then
333. 49:31Taylor gets 50 or some [ __ ] like that so those those systems will be created and by the way one of the things I've been
334. 49:39doing on these lives is talking about okay here's a business idea so there's a business idea create create a thing that allows artists to
335. 49:46opt in to create their officially licensed model to let other people generate on that and you know build that system make it
336. 49:56really easy for artists to do that I'm sure there's competition for that but that's that's going to be a massive [ __ ] business
337. 50:02because because there's so so the current copyright thing is after something is published right depending
338. 50:10on the usage that's where we're going to charge you money right after so everyone's out here arguing about the
339. 50:17after you took my copywritten material you trained your model and more shit's getting out in the world and they're
340. 50:23they're all worried about usage and and after the fact I think there's money to be made on the front end of this
341. 50:30let me take all those songs put them into an official model so if Disney let's Disney does this right they're not doing it now but they they
342. 50:39could do this and and I would be surprised if they didn't you want something that looks like Mickey Mouse in your
343. 50:47whatever the hell you want to do wouldn't it be amazing to be able to click on the officially licensed Disney model
344. 50:55that says you can generate from this and if you use it here's here's what it costs you so you can do original work that's
345. 51:03officially Disney licensed maybe they have to approve it whatever left you brain I don't know what that means left you brain
346. 51:13you guys are weird with the English tonight it's it's like I see words I recognize but when I put them together they make no sense uh
347. 51:25uh welcome everybody if you're new it is here urls why are they here because if you haven't played with chat gbt that first URL is
348. 51:34chat GPT go there go there right now start playing while you hear me ramble or just leave uh and then bing.com is also chat gbt if
349. 51:44you click on the chat button at bing.com that's chatgpt4 connected to the internet and then poe.com's got six
350. 51:51different chat gpt-like things you can play with plus a lot of bots post kind of fun there's a lot to play with there um
351. 51:59just go start playing go start understanding what these things are all about and then prompts.chat if if you're intimidated by like the
352. 52:07just the blank box like what do I do here prompts.chat will teach you about prompting and interacting with these things
353. 52:13um and getting some interesting results out of them so that's what this channel is all about and if you've got
354. 52:24um questions about AI pop them in the comments I do my best to answer all of them um why does Poe why does it saypo.com
355. 52:32when it's an app because it's also a website you can go to poe.com but you can also get the PO app but they're
356. 52:39they're basically the same thing so if we go to poe.com so there's poe.com and so you've got you've got your
357. 52:54different models you've got your different models here and then you have a bunch of bots and you can explore Bots you can create a bot
358. 53:03come on focus focus focus all right anyway so that's why
359. 53:21do you listen to AI oh to the all-in podcast with Jason kalkanus yes I do I know Jason from the Silicon alley days
360. 53:30the early days of the internet when he did Silicon Valley Reporter I was featured in that a couple of times Bing is terrible
361. 53:41um Bing is actually getting better but it is not anywhere near as good as uh as GPT as as gpt4 at chat.openai.com
362. 53:52threads threads hit 100 million hit 100 million users in five days so so chat GPT is no longer the fastest
363. 54:00adoption of technology in history but I will argue that it's cheating to start with a billion person user base of Instagram and just say
364. 54:10click over here to our new Twitter competitor um it's okay like like it's missing a lot of the features like I want to be able to
365. 54:20see things that that people put like I want to be able to have followers and all that sort of [ __ ] and I I don't know
366. 54:26it's just it seems a little basic right now people seem to be enjoying it I'm I'm exhausted like I you know I'm
367. 54:33I got a [ __ ] Tick Tock channel so it's a lot it's a lot just to have a whole nother competing place to be clever
368. 54:47so I don't know I'll use threads when when uh if if musk flushes Twitter down the toilet and threads takes over I'll
369. 54:55move there I've got an account there I did one post I put up a mid-journey image you can go see me I'm at Kyle Shannon
370. 55:05over at uh at the threads Special K of school cereal so why pay for po okay here's why pay for Poe this is why I pay for Poe you
371. 55:16can choose to get that right here so so when you go to Poe there's six different models that you can play with sage
372. 55:25Claude instant and chat EPT you can use for free and then there's these three that say subscription access gpt4 Cloud plus and
373. 55:34Claude instant 100K gpt4 you know quad plus is on anthropics version of gpt4 it's not as good but it's close
374. 55:44but this Claude instant 100K that means it's a hundred thousand token context window which basically means you can you
375. 55:52can have a 70 you've got a 75 000 word context window you could put in a sixty thousand word novel and and have it
376. 56:02write you an epilogue you can start interacting with um high level or like like high quantity um kind of data where chat GPT is
377. 56:12limited to four or eight thousand tokens which is about three quarters of that is words um so that's why I pay for it is for
378. 56:22that but now that chat GPT has this code interpreter I can now upload a hundred megabyte file and start interacting with it so Poe all
379. 56:39of a sudden is a lot less interesting to me because code interpreter is a big big deal so if I were gonna pay 20 bucks a
380. 56:48month I'd pay it to open AI ah [Laughter]
381. 57:05best AI tool to assist with short form content for tick tock go learn how to prompt it now listen you
382. 57:16can go to futurepedia.io there are tools like Jasper and copy Ai and any Myriad of tools that are going to do tick tock
383. 57:26scripts for you easily you can go pay them they're gonna they're gonna charge you whatever they're gonna charge you 20 30 40 bucks a month
384. 57:35um what what almost all of those services are is someone has written The Prompt for you and I would say we're early enough and
385. 57:47these tools are raw enough and simple enough that it it would it would serve you to go learn how to do
386. 57:54you know have chat GPT write short form content for you gpt4 is staggeringly powerful for [ __ ] like that
387. 58:05like staggeringly powerful even 3.5 is really good but four four is very very good and so so go to prompts.chat
388. 58:14you know read the sort of primers in there and some of the examples and you'll start to get a sense of it and
389. 58:18then also just start to hang out like go to go to the chat GPT subreddits um start following people on Twitter that
390. 58:29you know are talking about prompting and prompt crafting there's another um Tick Tock maker here named Rachel Woods
391. 58:37she's got uh a newsletter called the AI exchange she's really good at talking about this stuff start following her
392. 58:44um and I would say just get good at it because it in a very short amount of time you'll do a prompt and say make me a
393. 58:53tick tock script and you'll give it enough context and it'll do something and you're going to be you're going to do one of these
394. 59:00I didn't I could do that I made it to that yeah so go go do that Nicholas go teach yourself yeah open AI Chief
395. 59:11generative officer that's me man first one I think not that it matters which certificate would you recommend for generative AI none
396. 59:22I you know the the certificate of Hard Knocks like go learn this [ __ ] learn the concepts below it um I don't think certificates matter
397. 59:30right now um if you want to if you wanna play more on the um on the backhand side then then they're more important now
398. 59:41um Google just came out with one and was it Facebook or LinkedIn a couple of people just came out with generative AI I don't know if they're
399. 59:50certificates but they're they're courses um if you just want to get your head around what these things are how they
400. 59:55work things like that you you can go do those um you can also go watch a bunch of the videos on my channel like a lot of what
401. 1:00:02my early content is is here's what these things are but there's a lot of content out there but I think what's most
402. 1:00:09important right now is start using them go there and start playing get curious get exploratory think of it as an
403. 1:00:15adventure start learning it start learning it start learning it all right um I identified another secret cyborg at
404. 1:00:25work today that's freaking awesome so so what she's referring to there is there's a Wharton Professor named Ethan
405. 1:00:33Malik and he's like my dog is whining Hannah what do you want
406. 1:01:05so so Ethan Malik is this Wharton Professor that um he wrote this wrote this piece a week or two ago called detecting the secret
407. 1:01:16cyborgs and the secret cyborgs are employees and organizations who are getting AI literate but not telling the
408. 1:01:22organization about it because a lot of the organizations some of them have you you can't use chat GPT at all policies
409. 1:01:31some of them are like you know we'll fire you if you use that some of them are like wow we're taking a wait wait
410. 1:01:37and see attitude and and then also some of the employees are like I have dramatically increased my personal productivity here and I don't
411. 1:01:45necessarily want my boss to know about it because they're just gonna expect me to do more so so you kind of have this
412. 1:01:51this this invisible class of AI literate employees in companies that aren't dealing with this um and and he's talking about they need
413. 1:02:00to find a way to do that so it sounds like cabruno found a secret cyborg that's awesome and that's you know that's again why I
414. 1:02:09think that I think that it's valuable for you in an organization if you are a secret cyborg if you are AI literate talk about it
415. 1:02:17talk about it if you're working for a place that's gonna fire you for being AI literate and doesn't want to have
416. 1:02:24an interesting dialogue about it and say tell us what you know and you know and what you think about this stuff then
417. 1:02:33they're going to get blindsided at some point you probably don't want to be there anyway which accredited institution would you
418. 1:02:38recommend or learn to get a certification um again right now for the generative stuff like chat GPT I wouldn't
419. 1:02:47um if if you want to get into the machine learning side of things and like go work for open AI or anthropic or one of these companies
420. 1:02:54then I don't know I have a degree in acting so you're asking the wrong guy Stanford MIT you know start at the top and then work down
421. 1:03:06I mean not for nothing Stanford and MIT they have free courses like all of the universities have free courses to teach
422. 1:03:12you all of that [ __ ] Google's got courses um I'm sure meta's got them Stanford MIT it's all it's all out there for free so
423. 1:03:26just go do that or if you don't have the discipline like I don't like I I can't I couldn't self-educate like that because I just
424. 1:03:34I've got add and I'm like I'd rather just learn it myself um yeah I don't know I don't know right now for for me it's more important
425. 1:03:48that this stuff is moving the reason that I'm saying any sort of certificate in chat GPT or prompting or anything like that
426. 1:03:58well it's going to give you a leg up right now by definition anything that they're teaching you is going to be obsolete by
427. 1:04:06the time they're teaching it because this stuff is moving fast like it's moving like like code interpreter this thing that just came out
428. 1:04:15it's a fundamentally different way of interacting with chat gbt like I don't have my head around it yet at all like I know what it does
429. 1:04:24I know that you can upload a file and then you ask it to do [ __ ] and it starts writing python code and executing it to
430. 1:04:32do the thing that you wanted including like making you charts and graphs of the analysis of the data that's in the file that you uploaded
431. 1:04:40okay that's what it does but like what do you actually use it for and how would you actually use it in business
432. 1:04:47and how would I as someone who's interested in marketing and copying this and that like how can it be valuable to
433. 1:04:53me I don't know yet I'm still trying to get my head around it so so that's the the problem if you go learn prompting
434. 1:05:00you're going to be be essentially learning three to six month old stuff because it takes a while to get a course
435. 1:05:06together and like people have even asked me on here to do a workshop and I'm a little reticent because
436. 1:05:13I I don't know what I'm gonna teach that isn't going to be obsolete so I mean there's there's some Basics and Basics are fine
437. 1:05:22but but I don't think it I don't think it's I ain't the amount of time it would take you to find a course take the course you
438. 1:05:29could learn more on your own buy a lot Chief disruption officer just applied for the new position love it
439. 1:05:37I love using AI for emails yeah it's so good for [ __ ] like that all that low level crap what about Microsoft so so
440. 1:05:42listen well that's what bing.com is Microsoft um Microsoft's invested um 10 billion dollars they've committed
441. 1:05:5210 billion dollars to open AI so they're pretty invested they've already created GitHub co-pilot they're rolling co-pilot into all of
442. 1:06:02Office 365 they're rolling it into windows they're rolling it into all of their Azure Services they're rolling it
443. 1:06:08into Dynamics and it is the primary um Search tool at bing.com is now chat GPT so they're all in and and I mean what's
444. 1:06:22I've been a Microsoft Basher for decades I hated them and what they've done in the past six months has been remarkable like the how they're
445. 1:06:32incorporating this stuff and how they are [ __ ] with Google is is just entertaining they're like gleefully [ __ ] with
446. 1:06:42Google because they were just they were like gum on Google shoe for you know what a decade and a half in the search game they just couldn't
447. 1:06:51make inroads and like you know they've got it built into the [ __ ] operating system and they couldn't compete with
448. 1:06:57Google and Along Comes Lilo chat GPT and it changed everything and so they're all in so I'm I am I'm impressed with what they've been doing
449. 1:07:09I'm impressed with Microsoft um might they [ __ ] it up sure they're big company they can drive it off a cliff like
450. 1:07:18anyone but but I would say right now they're looking they're looking in a stronger position than Google currently
451. 1:07:25wouldn't count Google out but they're definitely on their heels they so here's the thing about Google they [ __ ]
452. 1:07:31invented the T in GPT is Transformer Google invented it and then sat on it for seven years because I think they knew that if they
453. 1:07:42released something like Chach EBT it would compete with their core product and it is and now they're trying to catch up
454. 1:07:52what comments is he reading I'm very far behind new Eureka Queen okay awesome name I'm the next New Yorker I'm living
455. 1:08:00in Denver now but I miss New York I miss my New York my new eurekans um I'm just way more far behind like
456. 1:08:08here's the deal this is Elder talk I'm an old guy using Tick Tock so there will be no twerking and um I actually try to answer all the
457. 1:08:17comments so I am very far behind on comments so that's why it seems like I'm not actually answering answering them
458. 1:08:23but if you're still here new eurekin know that I am it sounds like AI is going to put the marketplace through a brutal but
459. 1:08:31beautiful process that is a beautiful way to put it brutal but beautiful process yeah the
460. 1:08:47every company and every person every day has the opportunity to reinvent what they do but they don't why
461. 1:09:00because change is hard and reinvention sucks and it means you've got to give something up and it means you've got to
462. 1:09:06flush some things down the toilet and so I think that generative AI is going to brutally force reinvention of both people and
463. 1:09:18companies and it's going to do it quickly and the tools are so powerful that the ability to reinvent yourself is
464. 1:09:29going to be Amplified and accelerated and so I think we're going to see once we get through the brutal phase
465. 1:09:37which my guess is it's it's two to three years right starting now like we're starting to see it you're starting to
466. 1:09:44hear announcements of oh this company's laying off 4 500 people and or they're they're not gonna you know they're doing
467. 1:09:52a hiring freeze they were going to hire 7 500 but now they're not they're going to focus on AI um so I think it's going to be a two to
468. 1:10:01three year brutal brutal patch but I think the the Renaissance of creativity and Innovation that comes out the other side
469. 1:10:10of it is is going to be beautiful brutal but beautiful process I love the way you say that Chief super alignment officer yeah oh by
470. 1:10:21the way open AI for the for the alignment division for that super alignment thing that they're talking
471. 1:10:27about they're hiring the base salary for anyone in that group is four hundred thousand dollars that's starting starting salary
472. 1:10:37you give me money so let me get this straight I just got to get the robot not to kill the people yep and you give me 400 grand for that to
473. 1:10:49start yep go learn about alignment start talking about it on LinkedIn have a theory go watch David Shapiro on YouTube
474. 1:11:05you want to feel intellectually sketchy go watch David he's one smart [ __ ] he's written five books on this stuff
475. 1:11:16and he's got his last three or four videos are all about alignment and about open AI super alignment and what's good
476. 1:11:25about it what's bad about it he's brilliant yeah you want to get into this business you want to get into this business save
477. 1:11:33Humanity it pays well [Laughter] also if you're geeky I'd get involved with David Shapiro's uh open source projects
478. 1:11:44I think I think that is probably time well spent if for no other reason you're going to learn a [ __ ] ton um FRFR what and where are people
479. 1:11:53learning um well the what they're learning is what these things are and what they can do they're also learning about image
480. 1:12:07generation so mid-journey stable diffusion Adobe Firefly Dolly those tools are the picture equivalent of these words
481. 1:12:15then if you go to futurepedia.io there's a whole other Suite of um Technologies around video generation audio generation text to audio
482. 1:12:28music synthesis music generation things like that um he's on some other stuff I'm on your stuff now New York and queen I'm on you
483. 1:12:43I'm here for you ask a question I'll get to it eventually but then you gotta hang out maybe that's maybe you know what I I just it just hit
484. 1:12:53me this is a it's a [ __ ] brilliant strategy if you know that eventually I'll get to your question you'll hang
485. 1:12:59out longer huh like a steel [ __ ] trap all right people are moving more motivated by fear than desire of game yeah no I agree
486. 1:13:09people will do way more to avoid pain than than to gain pleasure um and so what this channel is about is well the Pain's [ __ ] coming
487. 1:13:26so why don't you start now and disrupt yourself before you get disrupted that I guess that's that's probably what where Where My Philosophy
488. 1:13:36is right now I want to learn as much as I can about this one because it excites me like the the the potential of what's coming in the future
489. 1:13:45is really [ __ ] exciting for me it looks really bright but I also know this shit's gonna [ __ ] up my like my
490. 1:13:54current companies an 11 year old automated video storytelling platform called storyvine this stuff's gonna [ __ ]
491. 1:14:00it up and I've got to reinvent it before it gets reinvented for me so that's that's where I want to be in these
492. 1:14:07conversations I want to be on the front end of the reinvention not being forced into it and that's that's where I think
493. 1:14:13the pain is going to be the highest is people that are like I've got three years I'll get to it um so yeah I totally agree
494. 1:14:23is one language any better than another for creating prompts French more descriptive I honestly don't know that's
495. 1:14:29a really good question I think that just by the nature of the public internet you know a lot of it being English
496. 1:14:38um I think it I think it probably indexes more strongly toward English but to your point certain languages are
497. 1:14:46going to be better at certain things I do know that you can prompt in one language and you know get results in
498. 1:14:52another like it's it's fantastic at translation so but that that would be an interesting experiment to to play with
499. 1:14:58if you've got that capacity let's see not me having assuming it had a meta awareness oh what kind of tripod or phone stand do
500. 1:15:11you use for your lives it's what's called a chicken foot monopod so so let's see wait can I um it doesn't matter it's basically a stick
501. 1:15:24with three feet on the bottom and then a and then a camera tilting pan head on the top and just the phone thing and then this
502. 1:15:34little window this is just a tick tock feature called dual where this the back camera is on and the front camera is on at the same time
503. 1:15:46um what will it do to the music industry in Grammy's will it make movies yes it will make movies if you want to go make some
504. 1:15:53movies for yourself they are laughably bad right now but so were images a year ago and now they are terrifyingly good so that's going to be
505. 1:16:01the same trajectory with video so go to runwayml I think it's runwayml.com let me get that for you yeah runwayml.com they've got two things
506. 1:16:17one called gen one one called gen 2. gen 1 you can upload a movie and then describe a new style like claymation
507. 1:16:26animation and it'll take your clip and turn it into a claymation animation but you know with all the composition of
508. 1:16:35your thing and then Gen 2 is text to video so you just describe a video and I think right now it'll make a three or four second clip and
509. 1:16:44they're disturbingly bad but so were images a year ago and so they're going to get really really good in fact if you
510. 1:16:51listen to the CEO of stability AI he said the the the models that they're working with right now inside stability AI stable diffusion um
511. 1:17:02the the things that we get to play with right now to create an image it takes about I don't know five seconds ten seconds
512. 1:17:10something like that depending on how busy it is the models they're working with internally are generating 60 images a second
513. 1:17:19why is that number significant that's that's the frame rate of video games 24 is the frame rate of movies 30 frames
514. 1:17:34per second is the frame rate of video 60 is the frame rate of games they're 120 but 60 60 frames a second
515. 1:17:42so so we are very very quickly going to be entering a world where you can say you know write me these five scenes or
516. 1:17:51or tell chat gbt hey go write me you know 16 different scenes of a video send that to the stable diffusion API
517. 1:18:03and and then I want to watch that video put this kind of music behind it put this kind of dialogue behind it I want
518. 1:18:09Tom Cruise doing the voiceover I want Morgan Freeman doing the voiceover and go make me that movie and it'll just
519. 1:18:17start playing we're not that far from that um what will it do the music industry it I it it transform it it transforms it in
520. 1:18:30ways we can't comprehend so if you if you if you want to see kind of the first you know iterations of this just go to
521. 1:18:40Twitter and look up Drake the weekend AI and you'll see a song there that someone created that was like a fan generated
522. 1:18:48song where they took the weekend and Drake's voice and put it in this AI generated song and it got all these it
523. 1:18:56got like 20 million hits on Spotify and then they took it down and that's when Grimes Elon musk's ex said hey you can
524. 1:19:04sample me all you want go make as much music as you want just give me 50 of the proceeds which I think is smart but but she's
525. 1:19:12going to be the exception most artists are going to be like I'll sue your ass but but we're quickly getting to a point
526. 1:19:19where fan generated music is going to be as good or better than the original what do you do with that do you sue your fans
527. 1:19:33I worked for a guy named Gordon Patterson who worked at new line Cinema and and when the when the first Lord of the
528. 1:19:41Rings came out when it when it was you know in production um all the geeks on the internets started making fan sites
529. 1:19:52and the lawyers at new line Cinema were like we're gonna sue their ass and Gordon Patterson was the guy that
530. 1:20:01said don't sue them invite them to Comic-Con he was the guy that invented social marketing for movies I think it has to go in that direction
531. 1:20:15right you're going to have fans creating content that sounds like you the songs might be better than yours
532. 1:20:24and they're leveraging your likeness and technically they should be paying you for that but they just did it out of love
533. 1:20:32what do you do with them do you sue them or do you embrace it and and what does that do to the to the if if a fan generated song
534. 1:20:41becomes a hit what happens who gets paid I don't know that shit's gonna have to be figured out and and what's going to happen is the
535. 1:20:55recording industry and the movie Industries it's happening right now with the Writers Guild with movies
536. 1:21:00part of the reason they're on strike is you can't use AI to write screenplays without us so you've got the most powerful writing
537. 1:21:12tool in the history of mankind and the union for the writers is saying don't use that I've written seven screenplays seven
538. 1:21:25feature-length screenplays like I know that world well why the [ __ ] wouldn't you use the most powerful writing tool Humanity has ever seen
539. 1:21:36well because it's gonna [ __ ] change everything it's going to change the economics it's going to change the power centers
540. 1:21:44it's gonna happen in music it's gonna happen in film it's going to happen in writing it's going to happen in the
541. 1:21:49agency world it's going to happen in the Consulting world all knowledge work all creative work the models are going to be decimated
542. 1:22:01brutal but beautiful was said just before brutal but beautiful is coming what's the beautiful I don't know we come out the other side
543. 1:22:13I mean imagine a world where we each get to have our own radio station that is music we create that sounds like our favorite
544. 1:22:28music whatever how do you comprehend that I mean there's a Spotify playlist sure but what if that Spotify playlist is all
545. 1:22:40stuff that you created they're singing about Taylor Swift is singing about your personal issues why is that a bad thing that's [ __ ]
546. 1:22:50staggeringly beautiful what does the business side of that look like I don't have a [ __ ] clue but it's a pretty remarkable possibility
547. 1:23:03did the lawyer just straight up say that I don't I don't know what I said but I'm sure he said something stupid all right
548. 1:23:09is chat you PT hallucination drag legal or illegal I want to try it yeah that's good it's just oh drug legal yeah hallucination drug
549. 1:23:23it's you know it basically Hallucination is a fancy word for it it's just it just it's just ignorant of what it said
550. 1:23:35I'll explain I'll explain how the hallucination works it's actually fascinating so Andre carpathy who's the I guess he's
551. 1:23:42the the CTO of open AI he was talking to the Microsoft developer conference and he was talking about um why these large language models
552. 1:23:52hallucinate like what's going on so the the metaphor I like to think is so imagine the Grand Canyon is full of
553. 1:24:00all the words that were on the public internet so everything that open ai's been trained on right the large language models
554. 1:24:06and if you're in a helicopter and you're and you're high enough you can see down all the Canyons in the Grand Canyon you
555. 1:24:12can see all the words so if you say to chat GPT solve this mathematical equation and you're really high in the helicopter it's gonna see
556. 1:24:23all of the places where that that mathematical equation was solved and and in that data set you're going to have
557. 1:24:30the Stanford mathematics professor that got it elegantly right and you're going to have all of the students from all of the universities
558. 1:24:38who got it devastatingly wrong it's gonna all of that's going to be in the data set So when you say solve this
559. 1:24:45mathematical equation your helicopter is really high up it sort of takes all those answers amalgamates them and gives
560. 1:24:51you an answer it says here's your answer and you're like that answer is comically bad and it's like I'm sorry sorry I got
561. 1:24:57that wrong here's the answer it's like well that's better but you didn't add these two numbers right oh I'm sorry I
562. 1:25:03got that wrong because it's it's sort of amalgamating right and wrong answers now if you say to chat gbt I want you to act
563. 1:25:10like a Stanford mathematics professor that specializes in this branch of mathematics that really understands this
564. 1:25:17particular corner of the universe of this kind of equation and I want you to solve that equation what you've done is
565. 1:25:23you've kind of flown your helicopter down in one of the little side Canyons where the information in that side
566. 1:25:29Canyon is only for professors of mathematics and so it will hallucinate less because it's got less information to draw from
567. 1:25:37which is just wild and and again all all it's doing all it's doing it's just [ __ ] remarkable but but what it's
568. 1:25:45doing is predicting one word after the other to give you the answer so your trick in prompting these tools is to
569. 1:25:52give it enough context to limit the semantic universe so that you have a higher likelihood of getting what you want that's the art
570. 1:26:02is lower your helicopter into one of the side Canyons when will it start thinking for itself probably within three years
571. 1:26:17and that paper from Microsoft where they've got a billion token context window 750 million words um
572. 1:26:30if that [ __ ] starts showing up it maybe happens sooner than that sometimes it really pauses for a long
573. 1:26:39time when I ask it complex questions generally Angela That's not that you ask that a complex question generally it's
574. 1:26:47that the servers are just slammed because hundreds of millions of people are using it um that that's generally just a a server load issue
575. 1:27:04yeah good evening Emilio's wife hope you're well I think comp equals comparables I got that eventually in real estate terminology
576. 1:27:15it took me a while it was like com compensation and then oh yeah real estate comps but it can do comps really well
577. 1:27:22our prompts copyrightable there's a there's a site called prompt what the hell is it called [Music]
578. 1:27:43prompt base if you go to promptbase.com you can sell prompts why you would buy a prompt there I mean
579. 1:27:53so here's the deal about prompting right now prompting is really important it isn't going to be for long if if you
580. 1:28:04want to build tools you're going to need to be good at prompting if you just want to use these tools a lot of this prompt crafting [ __ ] is
581. 1:28:12going to drift to the background because the tools are going to start prompting themselves if you go look at what what um what
582. 1:28:22um code interpreter does you you upload a file so let's say you you upload an Excel file and then you say tell me what's in this
583. 1:28:33it writes itself its own set of instructions it then writes itself python applications to execute those instructions executes that
584. 1:28:43python application and if it gets it wrong it says you know what I didn't I didn't do that right there's
585. 1:28:50another way we could attempt this let me try that and it'll write another python script and execute it so these things
586. 1:28:56are increasingly going to be self-executing um very sophisticated prompts from very minimal input they're not there right
587. 1:29:08now so that's why learning The Prompt crafting right now is valuable but I don't know that you would need to um
588. 1:29:17copyright a prompt because they're so easy to to you know just copy but go to prompt base and maybe short term you can sell some
589. 1:29:29if you're really good at it and you got something slick that no one else has figured out why not I doubt it Captain lit
590. 1:29:41is one language any better than the other for creating prompts oh we talked about that one you keep talking about
591. 1:29:46that code interpreter I just started installing it and using plugins today um all right let me go show you we'll go
592. 1:29:55do a code interpreter thing here if you pay for chat GPT plus you get code interpreter and plugins um I'll show you how to turn them on in
593. 1:30:09a second here there's a fourth mode that used to be there that's no longer there which is browse with Bing it's it's gpt4
594. 1:30:17connected to the internet they they [ __ ] up and so they had to kill it so they'll bring it back when they fix it
595. 1:30:25but I'll show you code interpreter if you don't see code interpreter if you do pay for Chad gbd plus and you don't see
596. 1:30:30code interpreter you have to turn it on so you go to your settings and you click on beta features and then you
597. 1:30:41have to turn it on so you have to explicitly turn on plugins and code interpreter um and then when you put code interpreter
598. 1:30:51here you can see there's there's a plus button here and so that plus button is you can upload a file so I am going to go grab
599. 1:31:12let's grab this one so it's IMDb tv.zip so this is a ZIP file so I'm going to say what is in this question mark
600. 1:31:30so it says let's unzip the file and inspect its contents okay cool that's cool well how's it doing that all right
601. 1:31:38let me turn down my brightness so you can see a little bit better now you can barely see it okay um if you open these things and show the
602. 1:31:47work in order for it to open the zip file it actually wrote a python application it it imported libraries and wrote python to unzip the file
603. 1:31:58so it says it seems there was an error the unzipped folder's name might not be IMDb and then but then it did it so so it unzipped it
604. 1:32:10and then it says okay what's in here are the following CSV files so mystery sport biography so this is IMDb
605. 1:32:18categories data about IMDb TV shows these files are likely data sets related to TV series in their respective genres
606. 1:32:26would you like to inspect the content of any specific file can you um analyze the data across these files and
607. 1:32:43visualize the trends you think are interesting in them question mark
608. 1:32:59sure to start the analysis so now it's writing instructions load the CSV files into Data frames inspect the structure
609. 1:33:08of the data to understand its columns and their types identify interesting columns for analysis so let's start by loading the first file
610. 1:33:16so the csvs contain title ID release year genre cast synopsis rating runtime certificate the age certificate of the
611. 1:33:26series oh that's interesting number of votes and then gross revenue given this structure we could potentially analyze and visualize Trends
612. 1:33:34related to ratings genres release years certificates and number of votes across all the files let's proceed by loading
613. 1:33:42all the CSV files and combining them into a single data frame so so then it writes code to do that and look here's
614. 1:33:53this is a it made a little thing here certificate PG-13 number of votes Revenue not available genre mystery
615. 1:34:06so we're going to clean the release year to only include the start year of the series uh distribution okay there you go look so
616. 1:34:19so now it said let's start by looking at the distribution of ratings across the different genres we can do this by creating the box plot
617. 1:34:26for each genre so it figured out what chart it wanted to build it wrote The Python code to take all the data
618. 1:34:35it colored all this it created the Box chart it labeled it all so this is Mystery sport biography so here's all the
619. 1:34:44different genres down here so here's some sort of scatter chart that didn't doesn't look like it had good data so that was a bad piece of
620. 1:34:53data it looks like this is number of series released per year per genre and you know 1920 there was not much TV released
621. 1:35:08and then 60s 80s 2000 so here's so right after the 80s is when cable comes in and you start to see it increasing and then now we're
622. 1:35:20in streaming and you know all this stuff and and this thing's going to continue to go exponential because
623. 1:35:28um of all the content that's going to be created by AI but anyway like what what was our input here our input here was
624. 1:35:36two or three requests and it just went off and is doing full-on data analysis across all of the CSV files within this one zip file so
625. 1:35:57that's what code interpreter is now what can you use that for I don't know yet you can upload images and pull out
626. 1:36:06color palettes out of it you can swap images you can turn a static image into an animation with a zoom and a pen um you could upload
627. 1:36:21tax information you could upload insurance information start interacting with it I think something interesting I might do is
628. 1:36:34throw all the screenplays I've written plus all of the fragments of screenplays I've started which there's probably 20 or 30 of those
629. 1:36:46put them all in a zip file and upload that and then be able to interact with be able to ask it about anything that I've written
630. 1:36:54and have it write me new [ __ ] that follows my style what would that do I don't know I mean I could have it compare plot
631. 1:37:10lines across screenplays and say what are things that that my screenplay structures have in common would that be valuable I don't know maybe
632. 1:37:24it might it might inspire me to tell a new story I might be able to take a fragment of a screenplay and say hey
633. 1:37:32you've got seven complete ones here you've got four fragments which of the fragments seems most um thank you very much I appreciate that
634. 1:37:43which of the fragments seems the most interesting to you and let's go ahead and complete that so give me an outline complete the
635. 1:37:53outline so that it matches what I like to do apparently and then you've got two scenes there let's let's write the third scene I don't know
636. 1:38:04it's like I I was just starting to get my head around chat GPT and feeling like I had it together and then code interpreter came out and
637. 1:38:15now I'm like right back to Ground Zero feeling like an idiot what do I do with this I don't know I'm gonna figure it out
638. 1:38:21I want a grand Adventure you should join the adventure go there go there now if you have not played with
639. 1:38:31chat EBT go there until you have this moment and then start your adventure you forgot the light bulb I don't know
640. 1:38:41which light bulb you're talking about the ticket for the train car first seat is a dollar but there's another train coming for two uh um
641. 1:38:54and another train but the ticket is ten I think you're building to a point there or maybe that was something I was
642. 1:39:00talking about and you're adding to it it seems right get on the train now okay wait so code interpreter can take
643. 1:39:10in 100 megabytes of excel software data I want to give it a word doc give it a word doc give it a PDF you can give it a PDF
644. 1:39:22you know how there's PDFs that have data in them and there's PDFs that are just scans that are useless you can give it a scanned PDF and it
645. 1:39:32will OCR the PDF pull the data out of it and throw it into a spreadsheet for you it's [ __ ] insane [Applause]
646. 1:39:47and by the way code interpreter is the worst possible [ __ ] name for it we talked about it last night we're going
647. 1:39:51to come up with a better name jumped into AI just for fun but eventually found myself using it daily for work and
648. 1:39:57personal life Yes Albert that is the journey I jumped into it just for fun and then I realized holy [ __ ] [ __ ] did you know
649. 1:40:06what this thing can do that's the perfect Evolution Albert it's the perfect evolution now I'm using it daily and and I think
650. 1:40:16the evolution after that is I got an idea for something I'm going to change how I do something I'm going to reinvent something I'm
651. 1:40:26going to reinvent myself I'm going to reinvent my job I think this I think this is just going to be this rolling
652. 1:40:34everyone who who starts playing with this stuff early it's going to be a rolling set of Revelations about what it makes possible
653. 1:40:46and it's going to be both invigorating and [ __ ] terrifying and I think that's the right like I keep going back to this idea of we're
654. 1:40:57on an adventure whether we want to be on the adventure or not it's like at night there seems to be a dragon
655. 1:41:05coming into the village and killing people and they're like well who's gonna go [ __ ] deal with the dragon and
656. 1:41:13everyone's like I don't want to deal with the dragon and then there's like five people that like are having drinks
657. 1:41:21one night they're like you know what I I think it's just gonna keep killing people maybe we should you want to go find this dragon
658. 1:41:27well that's really scary yeah but you're strong and you're good with a bow and arrow and you can do those magic spells
659. 1:41:33maybe we should you wanna all right you know and then little cousin Lucy's eaten by the dragon the next night and
660. 1:41:42they're like [ __ ] it we're going I feel like that's where we are with this stuff so so those kind of Adventures are
661. 1:41:52terrifying and they're inspired and you can't do them alone and you figure out
662. 1:42:06who's the secret hero and you figure out who's not that's where we are so yes Albert that feels very right to
663. 1:42:20me how are they making money from it you got my money not from those [ __ ] um so here's the deal about Ai and these
664. 1:42:32tools is they're just tools now they happen to be the most powerful tools in the history of humanity so
665. 1:42:41you know I'm not diminishing the value of them but um they're just tools so how you make money with tools is you solve problems and so
666. 1:42:51what I say a lot is so how people are making money with them right now is they're learning these tools well enough
667. 1:42:57that they can teach other people how to use them so they're they're doing consulting they're doing workshops they're um
668. 1:43:03building automations for people they're saying hey I can I can make you've got a social media agency I can build you a
669. 1:43:09tool that will automatically generate tweets and posts and this and that all you have to do is put in the topic and
670. 1:43:15answer a question or two and it'll generate all that for you you want me to make one of those for you uh sure okay
671. 1:43:22what do you want me to deliver that by Friday you can you can do that yeah here you go here's your tool that's
672. 1:43:29[ __ ] remarkable yeah I know do you want to learn how to do that yeah okay pay me this money to you know so that's
673. 1:43:35that's how people are making money right now um and then and then it it comes down to what are you good at
674. 1:43:49so no name coming up with clever names is Apparently one of them um but you know so you've got writing
675. 1:43:58skills or you've got project management skills or you're good at coding or you're good at illustrating doesn't matter like
676. 1:44:06what is the thing you're good at and then I would say take what you're good at what you're passionate about
677. 1:44:10ideally those things are decently aligned if they're like this well that sucks sometimes sometimes you got to pay the bills right
678. 1:44:18but take the things where you've got competence and ideally where you've got some passion and then start applying these tools to that stuff
679. 1:44:27start learn how AI can make what you do in project management better faster stronger cheaper and then start putting it out in the
680. 1:44:37world and say hey I figured out a better way to do project management you want to pay me for that and that could look like a software
681. 1:44:45application that could look like an agency that could look like Consulting so you solve problems and then these are just the tools
682. 1:44:54so the trick is you got to get competent enough with the tools to understand what the [ __ ] they are in the first place and
683. 1:44:59that's almost impossible because it changes every three months but start now like if you're feeling if you're feeling like holy [ __ ] am I
684. 1:45:08behind that's do you you don't think I feel like one thousand percent inadequate to deal with something as powerful as code interpreter
685. 1:45:19I don't know anything about data analysis I have a degree in [ __ ] acting and even though I've been kind of
686. 1:45:24tech you know in the tech periphery I've innovated on the on the sort of boundaries of of technology for years
687. 1:45:32that's significantly beyond my current capabilities but it's going to change the [ __ ] world so why would I not play with that so I'm
688. 1:45:41gonna go through the pain on the adventure of walking through the swamp with all the snakes and the leeches
689. 1:45:51thinking that I might drown or die or go in quicksand to get on the other side of it because I know on the other side if I can get my
690. 1:46:00head around what code interpreter is I can do some [ __ ] remarkable [ __ ] we all can we all will I just want to be there first
691. 1:46:10well no not first I want to be there I want to be in the party that slays the Dragon and I think that that you know that party is is
692. 1:46:22entering the Forest right now I think this [ __ ] gets boring like five years from now when all this shit's figured out
693. 1:46:31and we're back to like grinding out the social media post for whatever the [ __ ] SEO looks like five years from now it
694. 1:46:39ain't gonna look like what it looks like today I'll tell you that I'd like to see AI do mashups it can do
695. 1:46:49all sorts of mashups figure out which ones you want them to do and go find the tools and start doing it how do they control bias and AI
696. 1:46:58um they do it through a combination of um improving the training data that they put in the base model
697. 1:47:07they improve it doing the fine tuning of how they get it from that base model to something usable um right now with open AI it's done with
698. 1:47:17humans human reinforcement learning I think increasingly it'll move to machine mediated so if you're interested
699. 1:47:27in this stuff go look up a company called anthropic um that's got a model called Claude there are a bunch of EX open AI people
700. 1:47:35and they got this thing called the Constitutional learning model where they basically write a set of rules like
701. 1:47:40don't hurt people increase Prosperity make happy whatever the heck their rules are and then when their system generates
702. 1:47:49things it Compares it against those rules I think that's a really interesting model what's up hang on
703. 1:48:16love you all right I love that um I have a beer truck side business and I've been using AI program some things
704. 1:48:28on our website oh that's super cool yeah and I mean you can also use uh AI Albert to do uh you know have it come up with marketing ideas
705. 1:48:37um tell it you want to do you know community events and you want you want ideas for that um you know describe your community
706. 1:48:45describe your business you know give it give it as much context as you can and just yeah keep kind of expanding that
707. 1:48:52out that's that's really awesome I mean and Albert it's so I love the comments that you're making so far because it's
708. 1:49:00I think I think it's very easy to over complicate what you do with these tools like I gotta learn about machine
709. 1:49:08learning and data science and this and no no no no thank you um you already know what you know you just need to learn enough of these
710. 1:49:21things to understand oh those could help me do the [ __ ] I already do better like these are amplifiers of humanity of
711. 1:49:32what you want to do so you're still you still get to drive the train this is just your tool like it's it the best way
712. 1:49:41I can describe what what Chachi BT is is it's like an idiot savant meaning it's really clueless as as a human as a
713. 1:49:51you know you got to tell it everything and it will lie and it won't you won't even know that it lied it will
714. 1:49:57confidently lie it will mansplain to you but it's also a savant in that it knows everything it knows all the stuff
715. 1:50:07so if you say Okay idiot savant I want you to act like a master marketer that specializes in food trucks and I've got
716. 1:50:16a beer truck and I got a this and I gotta that and this is my community can you come up with some ideas for me
717. 1:50:23like what what are some ideas that aren't going to cost me a lot but are going to drive lots of Engagement and
718. 1:50:29drive lots of awareness oh sure here you go okay I like two and three but the rest are horrible can you give me more like
719. 1:50:39two and three oh sure sorry about that so so while they're remarkably powerful they're at your service you're the pilot that's the co-pilot
720. 1:50:57so yeah love that love it love it love it love it love it love it love it love it now Tick Tock just jumped me to the bottom and I lost
721. 1:51:08my place in the questions and I'm a dum dum dum dum dum dum what'd you do tonight watch some old guy scroll through tick tock
722. 1:51:17that sounds awful yeah pretty much so as if I got a haircut yes I did thank you Mike looks good thank you oh wow
723. 1:51:26it's I hadn't even really noticed it it was up in the letters above that's uh that that's almost like I tried it says
724. 1:51:37thank you hair growth geek oh good lord good Lord the hair Saga at the AI learning lab it's it's always something
725. 1:51:55Chief fear officer that was a good one sorry I'm still trying to find where he was oh well it didn't look it's not let
726. 1:52:01me scroll all the way back Google should have bought chat GPT yeah Google should have used the technology they invented I said this the
727. 1:52:10other night this happened at Kodak hey bosses I just invented the digital camera I think this is going to change everything the bosses
728. 1:52:21how's that going to sell us more film uh it doesn't use it's not it doesn't use it's a new way of making pictures yeah
729. 1:52:28that's not for us we make film let me let me get let me give you a little some lessons here little kitty this is about
730. 1:52:36the silver halide business not about pixels that's what Google did when they invented the Transformer apparently so yeah
731. 1:52:48they could have bought open AI at some point I suppose but they were hiding from the technology they invented because it threatened
732. 1:52:56their business model so now Microsoft is like [ __ ] it we're in here's 10 billion dollars you want to go do some damage together Sam
733. 1:53:05all right let's go and off they go I use code interpreter to create a vegetarian menu and ask it to make Excel sheet with
734. 1:53:14ingredients that's awesome and then with plugins you can take that the ingredients and just have it order it
735. 1:53:19although you're in a restaurant business so you'll get them from your wholesalers that's awesome all right okay so I renewed my chat CBT
736. 1:53:26and installed new IOS app but I can't find how to upload code generator Okay so the official open AI chat GPT app
737. 1:53:39doesn't have all the features so what I do on my phone is just go to Safari or Chrome if depending on the device you're
738. 1:53:47on and go to chat.openai.com and log in there now in your settings you're gonna have to turn on the beta features to make sure it's
739. 1:53:56available so either on a on a laptop go to that site or on your phone go to that site and then you'll be able to see code code interpreter
740. 1:54:04code generator is so new that Google doesn't know what what I'm talking about when I ask for help I know it I know I
741. 1:54:11know this is we're on the we're on the that's that's where we are right now I think how steady and consistent wait I
742. 1:54:20think how steady and consistent you but I feel bad because I don't know you would get to all the comments oh oh yeah well
743. 1:54:31you don't need to feel bad I just that's just the nature of how I do this is I just read all the things you don't need
744. 1:54:37to feel bad at all I I wish I could read faster and talk less but I just this is how I talk [Laughter]
745. 1:54:51I'm like Chachi pite you can make money much of a tie just like life beautiful and brutal the brutal and beautiful
746. 1:55:03the the yeah brutal but beautiful that's that's what we're at the beginning of and what's what's fascinating about it is
747. 1:55:16like there's a couple of things going on that I find really fascinating I think because blockchain and nfts
748. 1:55:23were over hyped and then crashed and burned dramatically I think there's a reticence to oh this is just another
749. 1:55:35hype train so okay so let me get this straight last year it was nfts we're all gonna get rich on nfts and and this year
750. 1:55:42it's AI is it that's how it's going all right yeah that's good so so you got that and then you've got 50 years of
751. 1:55:49Hollywood telling us the robots are going to kill us and then you've got a tool that got to 100 million users in six weeks
752. 1:55:59that is profoundly powerful where enough people are using it and discovering its power like Albert I just started playing with
753. 1:56:07it and all of a sudden now I'm using it for real [ __ ] that we're in this kind of weird confusing place right now um but man is it coming
754. 1:56:23brutal but beautiful any recommendations for books not on this [ __ ] well there's probably lots of books if you want to understand machine
755. 1:56:32learning and neural networks and Transformers and weights and embeddings and Vector databases and all that sort
756. 1:56:41of [ __ ] if you want to be on the science side of this there's probably lots of books I don't know that world at all I'm I'm on the
757. 1:56:48other side of the inflection point let's just call it post chat gbt where it's about applying these remarkable tools to
758. 1:56:55what we do day to day to day um if there are books on that [ __ ] right now I wouldn't buy them because because they're they're by the
759. 1:57:06by the time you could even do a print-on-demand book right now for this generative AI [ __ ] it's going to be obsolete
760. 1:57:14I was I was pertin near an expert at um content generation engines using zapier and GPT 3.5 and now they're deprecating 3.5
761. 1:57:28you've got to go to four um or 3.5 turbo which acts different and you've got to prompt it different and
762. 1:57:36then Along Comes code interpreter which is a completely different Paradigm and is probably closer to what this thing
763. 1:57:44looks like in the future anyway um and I don't even think there's API access to it yet so it's like just everything's changing everything's
764. 1:57:52changing all the time and and like like we're on it's funny in in when you listen to VC's talk they're looking for
765. 1:58:01what they call the hockey stick where revenue for a company is like zero or going along like this and then all of a
766. 1:58:07sudden it goes like that or or you know adoption goes like that what's happening with these tools right
767. 1:58:12now is that's happening with Innovation where the Innovations have have hit this exponential curve where these things are
768. 1:58:19getting more and more and more powerful exponentially and and we as humans don't have a good way of understanding exponential growth so it
769. 1:58:32it took me four months four months so I've got a company that I've been building for 11 years I know that thing
770. 1:58:41[ __ ] cold I know what it's good at I know what it's bad at I know how to position it I know generally how to sell
771. 1:58:49it although I wish I were better at that but I know it called it took me four months as I started learning this AI
772. 1:58:56stuff to even get them like this like not even like this like it took me four months to just have it not be a completely foreign concept
773. 1:59:06and then over those four months as I learned it more and more I'm like oh okay so wait this piece of what that
774. 1:59:11does that applies to this piece of what we do and then and then like I'm now at the point where I can kind of see the future of my
775. 1:59:19company with this AI stuff based on what I learned in February now I take something like code interpreter and it's like it's like
776. 1:59:34right over here so now I've got this vision of my company that's based on February technology and here's code interpreter I'm like I
777. 1:59:42don't even [ __ ] understand what it is much less how it relates to my business or to what I do so there's no books
778. 1:59:53I mean don't get me wrong there's probably lots of books right because this thing can write books really quick
779. 1:59:59in fact I talked to a guy at an event he's like hey I published 12 books in the past three days and I'm like oh good God really
780. 2:00:10he was so proud of it they had covers and titles and self-publishing you had them printed out and you could look at them and hold them
781. 2:00:18he's like yeah you just tell it to write a book and it writes a book I'm just publishing them I'm [ __ ] exhausted
782. 2:00:30so so yeah there's lots of books there's no books I would recommend 400 000 we're starting starting at open AI
783. 2:00:40so if you go to the openai.com blog look at their article for super alignment that's a new group that Sam Altman is
784. 2:00:52dedicating 20 of their compute resources to making sure that their AI models are aligned with human goals or or rights
785. 2:01:04and that group the starting salary if you get a job in that group is four hundred thousand dollars this 56 year old guy had fun blowing the
786. 2:01:13mind of a 20 year old with introducing Chachi BT to the yes [ __ ] gen xers I think I'm telling you man this is this AI [ __ ] is Revenge of
787. 2:01:24Gen X we're gonna come out of the Shadows because we know how to use computers like we know how to [ __ ]
788. 2:01:30use computers we started some [ __ ] and we're gonna get good at this and with 30 years of life experience plus these
789. 2:01:38tools we're gonna be able to [ __ ] some [ __ ] up and blow some 20 year olds Minds I like it I like it that's solid that's great
790. 2:01:47oh wait wait you're going so fast go where for image video music oh sorry the very bottom URL is called futurepedia.io
791. 2:01:57that's a directory of like 4 000 AI generated tools so here's here's where I am philosophically with all these tools the only things I'm
792. 2:02:11really paying attention to right now are chat GPT and variations of it and and some of the image generating tools
793. 2:02:18because I like doing the images so mid Journey stable diffusion Adobe Firefly um all of the other [ __ ] text to video the music stuff
794. 2:02:32I'm doing a little bit of voice synthesis stuff I'm doing a little bit of video stuff for my business but just kind of on the periphery
795. 2:02:39um there's so many tools out there there's tools that you describe an app and it will design an app for you and
796. 2:02:46put the files in figma there's ones that'll do that for websites there's ones that'll take a website and extract
797. 2:02:51the design out of it and build you a whole figma design Library there's things that write songs there's things
798. 2:02:57that like I can't keep up with the tools so that's futurepedia.io is like 4 000 tools most of which are
799. 2:03:06[ __ ] right but there's probably 400 in there that are like earth-shatteringly powerful um so that's why I have that there because I
800. 2:03:18can't because I don't I don't [ __ ] know I can't keep my head around chat GPT see fear officer
801. 2:03:36does an acorn grow after being cracked into an oak tree yes it does that's why I'm here every night just waiting for Mr Shannon to answer my
802. 2:03:47question see I answered your question emilios that's right it's it's all a master it's a dastardly strategy to get
803. 2:03:55you all to hang out here longer I like it all right what are the best image programs mid-journey I've tried
804. 2:04:02repeatedly to generate specific images but it never works okay Angela so um mid journey is the one in my opinion
805. 2:04:10that is the most sophisticated and best stable diffusion is the one that is the most flexible and has the most
806. 2:04:18developers developing things for it but you need to be geeky to use it there's a tool called leonardo.ai that's built on stop and on top of
807. 2:04:30stable diffusion oh this is a good this is a good note I got for my tick tock um leonardo.ai you have to get on a wait
808. 2:04:42list but I heard last night from someone in my comments if you get on Leonardo's wait list and go join their Facebook group they'll
809. 2:04:51give you instant access to Leonardo so Leonardo's an incredibly well done website where you that sits on top
810. 2:05:01of stable diffusion so you don't need to do all the geeky stuff so go do that I would also play around with Adobe Firefly
811. 2:05:09their generative tools if you've got access to the Adobe suite play with Adobe Firefly and then play with generative fill within Photoshop
812. 2:05:19because that's really good as well so those three are the three that I would play with you can go to bing.com and just tell it
813. 2:05:27to create an image but that's built on top of Dolly which is the weakest of all the image models so if you want to see something
814. 2:05:34generated poorly go there if you want to see something good mid-journey or Leonardo and then it's worth
815. 2:05:45understanding the Adobe products because they're commercially safe John Henry drilled a 14-foot hole into a
816. 2:05:53rock but the steam engine was only able to drill nine that seems like something I should look up up on chat GPT
817. 2:06:04you too can learn the history of the steam engine which had gbt [Laughter] that's awesome is Chachi PT knowledge standardized
818. 2:06:20across languages that means other cultures can influence the West yes it knows all the languages it knows all
819. 2:06:29the computer languages it knows all the language languages I translated I did something from English what did I do English to Dutch to
820. 2:06:39Japanese to Portuguese back to English and I could still read it um it it knows languages and and yeah I
821. 2:06:46agree and that's that's the other thing Google just announced two weeks ago or something like that one of their new
822. 2:06:53real-time translation models um is is is gonna do Eleven Hundred languages I didn't know there were 1100 languages to translate um
823. 2:07:05again we're probably within a year maybe two of being able to speak in English with someone else who's a native French speaker
824. 2:07:15I'll speak in English they'll hear me in French in fluent French sync to my lips they'll respond in French I'll hear them
825. 2:07:25in English that's coming it'll start out for like the top eight languages and then it'll quickly go to a
826. 2:07:35hundred and then it'll quickly go to a thousand and then it'll quickly go to you can speak with people in made-up languages I don't know
827. 2:07:43so yes I think and and I think your point Angela that what what the what this tool actually kind of does it's kind of a mirror that
828. 2:07:56reflects Humanity so I think this is going to build Bridges I think it's going to build empathetic Bridges and I think it's it's
829. 2:08:04going to build cultural Bridges where we learn we're all kind of the same and this will become the Babel Fish kind of cultural language
830. 2:08:16Humanity translator and amplifier um I think it's got It's it's very much got that possibility in its in its uh in its future
831. 2:08:28you just got up to the first comment from like an hour ago I know it made no sense without context yeah I know most
832. 2:08:35of my words make no sense and have no context so that's part of the joy of the of the channel wait I sat here for two hours for that uh-huh
833. 2:08:48hahaha you did all right do you think you'll be able to take a screenplay and produce a film all on your own at some point yeah and I
834. 2:09:00don't even think you'll need the screenplay quite frankly I did an experiment what was it with it was just with was it chat gbt
835. 2:09:10yeah I did an experiment where I said I said take the take the structural plot elements of Star Wars
836. 2:09:27and write me a screenplay outline of an animated movie of animals on the Galapagos Island at the time of Charles Darwin
837. 2:09:41um and basically follow the plot of Star Wars and it was [ __ ] brilliant so so I think you're not even going to need to to
838. 2:09:51start with the screenplay I think you can say I want to see something that has the dramatic Arc of you know Indiana Jones
839. 2:10:02and the Temple of Doom the comedic stylings of Caddyshack starring The Simpsons with a soundtrack by David Bowie and
840. 2:10:15it's going to generate that now I you know that's that's not going to be for some time but maybe within five to ten years
841. 2:10:23so I think the Stepping Stones to that are going to be um you know take this movie and change the ending or take take these two movies and
842. 2:10:34and combine you know I want these actors in that movie and I think I think that kind of stuff is going to be coming soon I'm scared so
843. 2:10:43scared is good as long as it goes with curiosity and and uh a sense of adventure um don't let the don't let the fear just
844. 2:10:54have you sit back and not engage because it's really easy to dismiss this stuff as either evil and bad and the
845. 2:11:02robots are gonna kill us or as well it's not perfect it makes mistakes it didn't I asked it to give me something and it sucked
846. 2:11:12don't dismiss it because of that this stuff is as bad as it's ever going to be and it's [ __ ] remarkable but it's as
847. 2:11:21bad as it's ever going to be it only gets dramatically better from here in terms of its capability um so so being scared's okay we're we're
848. 2:11:30we're we're entering a brutal but beautiful transition um and and being scared I feel like is an appropriate response to this looks
849. 2:11:39like we have a bunch of new people here my name is Kyle Shannon this the AI learning lab welcome everyone uh I
850. 2:11:44ramble for hours and uh that's what this channel is about apparently um the URLs on screen right now these are
851. 2:11:53the top one is the official chat GPT website if you haven't gone there please go there just go there um and if you don't know what to do
852. 2:12:00there prompts.chat this thing down here um is a little primer that will tell you about what prompting is and how these
853. 2:12:07things actually work and it'll give you a whole bunch of ideas of things that you can tell that to do like act like a
854. 2:12:14screenwriter act like an accountant act like a therapist act like a coach and it will just go off and do that and and and
855. 2:12:23you owe it to yourself to use that long enough for you to have this moment and once you have that moment then you
856. 2:12:32should be hooked and and like Albert in here you can start out and say oh it's just [ __ ] around with it and now I'm
857. 2:12:39using it every day to run my business but yeah now I'm going to reinvent my business now I'm going to reinvent myself
858. 2:12:53all right Deluxe random AI only creates from the past I disagree I mean you could argue that humans only create from the past
859. 2:13:09we create from what we have learned and they can say well we've got imagination we create original new stuff if you use these things with
860. 2:13:21you know any level of depth these things create now it might be assimilating what it does from different places
861. 2:13:32but by me prompting it to give me something new I want you to combine this concept with that concept I want it to be structured like this
862. 2:13:44go out it comes that's a new construct that's never existed before so did it create it did I create it I
863. 2:13:55prompted it it's my intention to combine this with that with that it just happened to do it like that
864. 2:14:03and it happened to have the knowledge of all the past to be able to do that but it actually created something new so I
865. 2:14:09actually disagree with that I think all of what Humanity does to some degree is based on the past right
866. 2:14:15it's all language that we created in the past all meaning all of what we do all words were invented by someone else that's dead now
867. 2:14:32says this and then fans are going to want artists to sing songs of them that they didn't make and the fans are going to be able
868. 2:14:42to make that song that's that that was my point on that thing that's right fans are going to want those songs and
869. 2:14:49fans are going to make those songs and some of those songs are going to be better than the artist's songs
870. 2:14:55and I guarantee you there's going to be a hit song created by a fan that sounds like an artist that's going to be bigger than the
871. 2:15:05artist's biggest hit mark my words I think we'll see that within two years and then what do you do do you sue the fan
872. 2:15:13because the fan has a bigger hit that sounds like you than you do or do you invite the fan out on tour and give them backstage passes
873. 2:15:24and encourage other fans to make more I don't know brutal but beautiful brutal but beautiful that's I love that
874. 2:15:39hey Kyle what's the first policy a company should have about chat CPT um that's a good one I mean it's a good question I don't have
875. 2:15:48a good answer well it's a really good question
876. 2:16:06I think the first policy is they should aggressively encourage
877. 2:16:22everyone in the company to understand what this stuff makes possible combined with understanding the risks
878. 2:16:36that it the the risks of using it right so okay we want you to use chat gbt but while we're all learning it you can't
879. 2:16:50put any client data in it and you can't put any of our trade secrets in it but you can put recipes in it you can put
880. 2:16:56little project management stuff in it but right so I think you need to find some boundaries like until we know until we
881. 2:17:03have a security and privacy policy around this stuff and until maybe we've created our own Standalone server that's
882. 2:17:11got its own firewall so that we're confident that our data is safe you can't do anything but in the meantime
883. 2:17:18let's learn about this let's learn about it together let's have weekly lunch and learns where someone
884. 2:17:26who learns something new we did this at agency.com in the early days we had these things called geek and tells like
885. 2:17:32show and tells and what a geek and tell was was someone would geek out on something they'd learn something new
886. 2:17:37with the early you know web technology and then they go oh I'm having a geek and tell and people would gather around
887. 2:17:43their computer and they'd show them how they did it and then everyone would go oh that's great they'd run back to their
888. 2:17:48desks and they'd go do that and they'd go play with it um so so I think I think having both management and staff in an organization
889. 2:18:00get AI literate um aggressively um and and do it in a way that encourages individuals to share ideas and have
890. 2:18:16management and and the executive team be open to those ideas because some of those ideas might look like a challenge to the current business
891. 2:18:25model and if an individual says oh I've got an idea and they put something out there that's a challenge to the business and
892. 2:18:32one of the executives chops their head off for it no one will ever speak up ever again right so that's a cultural thing
893. 2:18:42so I think getting aggressive about learning but then also learn enough to understand data and privacy risks and and make sure to
894. 2:18:51mitigate those so I don't know if that's a good answer or not that's that's where I'd be Sarah Silverman's suing open AI she is
895. 2:19:01that's correct social marketing is evolving thanks to AI a new monitor wait social marketing is evolving for sure a
896. 2:19:10new modernization strategy is needed to stay ahead absolutely brutal but beautiful yes imagine an unlimited supply of music
897. 2:19:20from a deceased artist oh that's coming that's absolutely coming in fact here's a business I keep giving business
898. 2:19:27ideas away take them run with them start a label that is nothing but um new music from artists who've passed or
899. 2:19:38no longer making music at different eras of their life young Elvis Vegas Elvis bloated Elvis yeah exactly
900. 2:19:52the lawsuit claims the three authors did not consent to use the use of their copyrighted books yep again lawsuits are gonna be everywhere I
901. 2:20:02don't whether or not these it's going to be really interesting to see where the sort of fair use and all of the laws
902. 2:20:10that that emerged out of the hip-hop fair use stuff it's going to be interesting to see how they get leveraged and and how they evolve
903. 2:20:19but none of this stuff's getting uninvented and the the open source llms are gonna anything that happens to open AI it's
904. 2:20:28just going to squeeze through the cracks and and migrate over to the open source llm so none of this stuff's stopping
905. 2:20:36GPT isn't good enough to write scripts yet um you can I've gotten it I've gotten it I've gotten it to absolutely um
906. 2:20:46give me good outlines I've gotten it to give me dialogue that's probably 60 there but it's at least a first draft
907. 2:20:58um non-screens screenplay stuff I can usually get it to about 80 percent of finished work so 60 Cent ain't bad I mean you have to
908. 2:21:09do a lot of back and forth with it a lot of collaborating a lot of you can give it examples here's an example of a scene
909. 2:21:13you know that I wrote so follow that tone follow these characters I reckon business will shift to a more
910. 2:21:22intellectual property based than content um possibly yeah I think I think businesses shift to what becomes
911. 2:21:33more and more critical is critical thinking strategy um you know the why why are we doing this what are we trying to accomplish and
912. 2:21:44then the tools will do all the Tactical execution mixed Tiesto with Elvis songs that would be cool first Industrial Revolution destroyed
913. 2:21:55physical labor now this destroys creative labor and knowledge work yes correct but it also will generate entire
914. 2:22:03new categories of jobs we can't even imagine today so again the brutal beautiful part the brutal is the next two the
915. 2:22:11disruption in the next two to three years is going to be brutal the re-emergence of whatever the new
916. 2:22:19jobs are and whatever the new Prosperity is when our ability to produce accelerates dramatically without an increased need and labor
917. 2:22:32it's brutal initially because that looks like people that are displaced but ultimate ultimately this starts to look
918. 2:22:39like increased prosperity and and new ways of doing work and new ways of doing all sorts of things all right my butt's starting to get numb
919. 2:22:54that means it's been at least two hours [Laughter] so if it's 10 30 here in Denver 10 24. Do You Know Where Your Children
920. 2:23:05Are listen listen you and me we gotta talk what do you do on these Tick-Tock lives I don't know I just I just talked to my phone
921. 2:23:15why do you do that I know this there's people there do they talk to you sorta why why do they watch you I don't know
922. 2:23:25it seems dumb yeah oh it's dumb it's dumb dumb it's a dum dum all right carry on all right let's see wait
923. 2:23:40Alibaba Cloud unveils tanji wanzing an AI image generation model oh cool that's interesting yeah it's I mean it's
924. 2:23:50interesting we haven't seen kind of a major open AI type player come out of China yet so maybe we'll see like a mid-journey kind of thing
925. 2:23:59Alibaba Cloud okay I'll look that up that's interesting haircut yes thank you damn we gone can I write a plug-in and deploy it
926. 2:24:11internally for private use well sort of so so I don't know about the plugins themselves I think those have to go into the plug-in store but
927. 2:24:25the the GPT 3.5 turbo API now has function calling so basically you can describe a set of functions within your code you can
928. 2:24:35describe 20 functions and then the chat bot will dynamically know which of the functions to call based on your prompt which is
929. 2:24:43essentially plugins so yeah look up look look up function calling and the answer is yes Elton Elton and Harlem that's a cool name
930. 2:24:56does AI learn and retain my prompts to assist other people's future prompts uh sort of you can tell it not to but you
931. 2:25:04lose your prompt history so make sure you download your data um I don't think there I I don't think it's
932. 2:25:12so much the prompts it's the combination of the prompt and response and like if you give it thumbs up they're using what
933. 2:25:17you're doing to train the model but it's not like they're taking your prompt and saying hey someone else you
934. 2:25:23were talking about this want to prompt someone else wrote it's not like that it's just more like they're they're
935. 2:25:28improving the response accuracy based on the work everyone's doing is there a way to use an AI tool to find or remove malware
936. 2:25:38probably but I don't know I would go to futurepedia.io and look up security security and things like that um
937. 2:25:46there will be ways to use AI to do that a lot of the malware detection tools are probably already using AI don't forget
938. 2:25:52ai's been around for two or three decades um and it's been you know powerfully integrated into almost everything we do
939. 2:26:00for the past decade so um so this stuff is just kind of a a a layer sitting on top of a lot of those
940. 2:26:08core technologies that are already there what plug-in is this I think I was showing code interpreter which is the
941. 2:26:17new mode it is a plug-in but it's it's way more than a plug-in it's a so what what code interpreter is actually doing
942. 2:26:23this is freaking wild and I I I'm wondering if if you're if your chat GPT chat session is already this but I just didn't get it before
943. 2:26:33what code interpreter is doing is when you launch a code interpreter chat session it's basically spinning up a virtual computer
944. 2:26:42and then when you ask it to do stuff it writes code and then executes the code within your virtual private server executes the code
945. 2:26:53generate stuff saves it to the virtual memory or hard drive and gives you results and so you're basically interacting with a with a
946. 2:27:05dynamically generated computer that dynamically writes and executes code on your behalf that's [ __ ] nuts
947. 2:27:16what in zip file that's crazy so fast too yeah I think you can have anything in the zip file you can have hundreds of five like
948. 2:27:24that's what I'm going to do I'm going to put I'm going to put PDFs of my screenplays into a zip file
949. 2:27:31plus I might like I've probably got like word outlines I probably got all sorts of things I'm going to put in a zip file and then
950. 2:27:41just have it go read what's in the zip file and then start interacting with it it's I don't know so many subscriptions to pay for I've
951. 2:27:48limited mine too right now I have openai po mid Journey Google collab zapier and I'm probably going to get rid of Poe
952. 2:27:59because code interpreter I can now do long form long form stuff with code interpreter so I'm going to probably get rid of Poe
953. 2:28:08those are those are my biggies right now I don't I don't think you need more than that what a what I a I am I using that I just
954. 2:28:18mentioned them I have it checked for plot holes that's a great idea or or plot cliches or dialogue cliches
955. 2:28:27I worked with a screenwriter once and whenever he didn't know what to write he he used the word look
956. 2:28:34so it'd be like two people to be talking and there'd be an awkward pause and then the one character would go look I want
957. 2:28:40to tell you this and whenever you use the word look I knew it was shitty dialogue because that was his like crutch
958. 2:28:46and ever since working with him I now notice in TV shows and movies all the time you can see a character go look and I'm like ah
959. 2:28:57that writer didn't know what to write right there use code interpreter to do part of the number crunching that I don't like
960. 2:29:04leaving more than more time for what I do yeah yeah exactly you're talking about screenplays and code interpreter what do you think about
961. 2:29:14personal.ai I don't know personal.ai do I um
962. 2:29:31everyday messaging supercharge your memory and relationships with your own AI messenger I'll check it out remember everything
963. 2:29:41and connect every oh okay so this is something that's gonna gonna basically add memory of who you are and what you talk about I think
964. 2:29:53there's gonna be yeah so so I think um Sam Walton has said that that that's going to be something that's
965. 2:30:00added to open AI is is personal profiles and things like that I think that's that's absolutely where we're going I'll
966. 2:30:08check that out I haven't played with it but but that's absolutely where we're going the ability to have it have memory of our interactions
967. 2:30:18and have it trained on all of our data is the a big place how big of a CPU do you need you don't need one if you're using
968. 2:30:30chat.openai.com they're using Microsoft servers like like it costs them a lot of money every time you do a search it's
969. 2:30:37like 75 cents it's some ridiculous amount of money um if you want to run things locally then you need you know
970. 2:30:46a big GPU or you need at least an M1 Mac ideally an M2 Mac um but if you're not if like wait a year and probably even before a year
971. 2:31:01someone is going to create an open source large language model that's the equivalent of gpt4 that you can just
972. 2:31:07install on your laptop or maybe even on your phone so if so right now like to use chat.openai.com you just need you can do
973. 2:31:18it on your phone it's all the processing is being done in the cloud so you don't need your own if you want to train your
974. 2:31:24own models that's a whole different thing then you've got to rent like a Google collab with virtual servers and virtual
975. 2:31:31gpus or you got to have your own iron or you got to buy one of George hotz's new tiny boxes which has six gpus in it and costs 15 grand
976. 2:31:42so if you want to spend money you can Sam Altman said he thinks it's going to take a hundred billion dollars to get to AGI for openai so
977. 2:31:54how can it make money you can make money commenting on the time delays like learning to structure chat GPT prompts
978. 2:32:03character limit in context wow mind blown honestly is code generator another subscription code interpreter it's not it's just part
979. 2:32:11of GPT Plus how do you upload you click the plus button um is chat CPT a good name and what would you name it
980. 2:32:24if you were marketing and branding it well at this point Chachi BT because it got to 100 million users so quickly um
981. 2:32:36I think it effectively established its own brand like I think it's I think it's there there's real problems
982. 2:32:45with it it's not really a chat bot chat Bots have a really bad Legacy of just being bad and then GPT everyone calls it GTP GPT PPT
983. 2:32:58so it's horrible in that sense um what would I call it I think the the inflection one pi pi
984. 2:33:13is is really interesting something like mac right like the reason Steve Jobs called the Macintosh um Macintosh was because the nickname
985. 2:33:21was Mac it made it informal it made it memorable it was three letters so I think something like that
986. 2:33:29um something like mac something like pie something that implies it's your assistant it's your friend best friend Eva Eva something like that
987. 2:33:42um but I I see I don't think chat GPT
988. 2:33:58becomes like the World Wide Web where it's just this sort of like Xerox kind of thing I I think what's going to end up happening is
989. 2:34:08the underlying technology that right now we have to go to chat GPT that all goes away what it gets replaced with is Agents like like pie
990. 2:34:19or Apple's gonna have one you know they call them memojis right now which is [ __ ] horrible but but imagine
991. 2:34:28the equivalent of a memoji that's just like your agent and it looks like you it sounds like you it understands you it it
992. 2:34:37it's it knows your entire history and then you can just tell it to go do stuff um I think the underlying technology
993. 2:34:44people aren't going to talk about it for long I think what they're going to talk about are the agents and what the agents are doing
994. 2:34:50and then those agents are going to start virtual and then you know with Vision Pro you'll start to have your agent in the world
995. 2:34:58and then with physical robots you'll start to have those agents manifested in some sort of physical form so I think
996. 2:35:05the underlying technology will pretty quickly drift to the background um so chat gbt is okay not great code interpreters [ __ ] atrocious
997. 2:35:15foreign Magnum CI yeah exactly we were calling it Magnum CI last night I didn't come up with that but it's a really good one
998. 2:35:24because it you know it plays into the Gen X thing we're leaning heavy into Gen X all right a few more few more questions
999. 2:35:32then I gotta go go go go go go go go go go go go go go go go go go go go go go whoever finds the dragon may discover
1000. 2:35:38how to tame the dragon and use it for good exactly right I just imagine that there's all the flame was coming out it goes and the
1001. 2:35:45little smoke goes tamed him or we've killed him yeah yeah and then back to the Village he can be our friend he can help us Farm [Laughter]
1002. 2:35:57all right you won't be replaced by AI you'll be replaced by someone who uses AI exactly that's from uh from Paul rateser of the marketing AI
1003. 2:36:07Institute said that that is correct so be one of the people that learns ai go there play learn that [ __ ]
1004. 2:36:15today we celebrate the birthday of Nikola Tesla I forgot that just imagine if he had AI holy crap you know what would be really cool oh
1005. 2:36:27this is something to do with code interpreter take everything you can from Tesla patents any notes you've got any books that right
1006. 2:36:39throw that all into a zip file and upload it to code interpreter and let's go discover some of the [ __ ] in
1007. 2:36:50Tesla's patents that haven't been discovered yet that would be a fun one that would be a fun one because one of the things that these GPT
1008. 2:37:04models are seem to be astounding at like the speed with which it can parse a massive spreadsheet of data is just mind-numbing
1009. 2:37:18so go look at all his data that would be fascinating hmm I wonder if you could even because there
1010. 2:37:34were a bunch of patents of his weren't there a bunch of patents in his apartment that the government came in and took and scurried away
1011. 2:37:41I Wonder If You Could reconstruct some of his work through notes and you know application documents and [ __ ]
1012. 2:37:51like that I I bet there's something there do you think we can figure out how to pick strawberries faster have you seen the new [ __ ]
1013. 2:38:06laser tractor this seems [ __ ] insane they've got this tractor now this this this big sort of I don't know 15 or 20 foot wide box
1014. 2:38:17that sits underneath the tractor and it rolls along at like three miles an hour and lasers it's got cameras that
1015. 2:38:26are identifying the crop from the weeds and lasers are [ __ ] zapping the weeds as it rolls along so yes and then I saw I saw yesterday
1016. 2:38:40um someone invented this thing it's it's like a it's like a rack a rolling rack um that goes in between like down in
1017. 2:38:50apple orchard and then it's got drones that each of the drones has like a long pipe with a suction cup on the end and
1018. 2:38:58so the drones sort of fly off this platform identify the ripe apples fly up to the Apple suction cup it pick it and
1019. 2:39:07drop it into the bin in the in the cart so there were like 10 drones picking apples and dropping the apples into the
1020. 2:39:14cart all coordinated so yes I think they're going to be picking strawberries faster it's it's just gonna I'm telling you man we
1021. 2:39:26it's this is going to start accelerating like this kind of remarkable mind-blowing stuff that that in any
1022. 2:39:33other year of any other decade in any other century you would you would [ __ ] stop the presses and go holy [ __ ] everyone's got
1023. 2:39:43to know about this and now it's happening like every three days and it's gonna start happening like every day and
1024. 2:39:49then it's gonna happen like three times a day so yes I think Matthew we're picking more strawberries faster
1025. 2:39:59oh it's flipping amazing it's just amazing what is a prompt Library so prompt libraries just if you've got a
1026. 2:40:06prompt base or or if you go to that document prompts.chat right there that's a prompt Library um it's just it's just a collection so
1027. 2:40:14here's the deal unlike Google what what Google wants is for you to be as concise as possible find me the competitors of this company
1028. 2:40:24right they want you to just put in one short thing that doesn't kind of pollute the search and then they're going to
1029. 2:40:29give you as relevant results as possible this generative AI stuff works kind of the opposite it needs a lot of context
1030. 2:40:36right so you say I want you to act as a magician and you're a magician that specializes in stage magic and your job
1031. 2:40:46today is going to be to teach me you know how to saw a lady in half in a way that seems relevant to Modern audiences
1032. 2:40:53do you understand yes I understand so all of the crafting that you do to sort of wind chat GPT up to deliver good results
1033. 2:41:05it takes a lot of troubleshooting like you'll you'll get it you'll think it's great and then you'll ask it something
1034. 2:41:10and it's like oh no that's not great and you're like okay I need to know what I need to do and you'll modify your prompt
1035. 2:41:15and and and the prompts can get quite complicated and Byzantine and right and so a prompt library is just to say okay
1036. 2:41:22for the for the you know saw a woman and half magician dude here's the prompt for that you throw that in a library and then someone else can
1037. 2:41:31if they want to do magic stuff they would go search your library if it were public and go see it so that's all it is
1038. 2:41:42you're great thank you Mike appreciate that I use chat EBT to create a quiz to tell me a per wait tell me a person's
1039. 2:41:51invest and risk tolerance yep oh that's great so it created a quiz so so the next step for that Anthony
1040. 2:42:01take that quiz make that a Google form then do a zapier automation where when someone fills out that form
1041. 2:42:14you send their answers to chatgpt with a prompt that then responds to them with you know here's your risk tolerance
1042. 2:42:21report and then you could end it with by the way if you want to talk more about this reach out to me and we can talk I
1043. 2:42:27can help you with this or you know here's a here's a here's the kind of stuff you might want to invest in that would be a fun one to build
1044. 2:42:41you talking about Hunter maybe I don't know damn RT I'm not sure what you're talking about um why do all the AI Channel
1045. 2:42:51why do all the AI channels talk like their profits because this is the most profound technology in the history of
1046. 2:43:03humanity and so the minute you get your head around what it's actually doing and what it makes possible
1047. 2:43:12it's really hard not to go to a place of this is going to be brutal and beautiful which was talked about earlier in this live
1048. 2:43:22um so I don't know that there's a way to talk about this stuff without talking about the future because we don't know what it's going to
1049. 2:43:28be and it could be bad it could be good it's definitely going to be profound it's definitely going to be disruptive
1050. 2:43:36so probably because they're paying attention I mean what I can promise you is I don't have any [ __ ] answers I have a lot of thoughts
1051. 2:43:48based on my experience of having gone through this with the World Wide Web going from the early days of that through the.com Bubble Burst um
1052. 2:43:59like I've seen these markers before and what I can tell you is with the early days of the web I could look at what that did
1053. 2:44:10and I had a really clear sense of what the future looked like 20 years out and it pretty much looked like that
1054. 2:44:16this I don't have a [ __ ] clue because this is so staggeringly more powerful than what the World Wide Web was all the World Wide
1055. 2:44:27Web really did was allow you to connect this piece of information to this piece of information instantly that's pretty straightforward this
1056. 2:44:39takes the collective output of humanity or at least what was on the world wide web and and you know ultimately it gets much broader
1057. 2:44:48and it makes it available as this sort of simmering pot of knowledge you can do anything with what does that mean for the future I
1058. 2:45:03don't [ __ ] know but I know it's going to be big and profound so if this Channel and other channels like
1059. 2:45:10this sound prophetic probably because they're paying attention don't let Lucy die in vain my brother's
1060. 2:45:18writing a book on high Crown sombreros it helped him research using Chachi BT Yeehaw that's awesome AI these nuts nice
1061. 2:45:26that that's a that's a solid solid retort there Dust in the Wind love it if you're gonna troll here be creative
1062. 2:45:35with it you know ideally your troll is somewhere between a compliment and a put down but AIDS nuts is is it's it's it's
1063. 2:45:43class it's a classic it's a classic you can bring it up almost anywhere it still works still good solid
1064. 2:45:53I gotta go to bed people what what do you want with me I saw that Salesforce is getting AI yeah they have like 10 tools or something
1065. 2:46:01like that it will build workflows which sounds awesome Salesforce needs an AI front end so [ __ ] bad it is the most painful
1066. 2:46:11torturous piece of [ __ ] that you can't get rid of it's horrible and if you could just go met this guy stick it in there
1067. 2:46:24and it just automatically listens to your conversation with him writes the notes summarizes them pops it in there
1068. 2:46:29does all whatever crap sales force requires you to do because it you customized it for your stupid [ __ ] sales workflow like we have
1069. 2:46:39God damn does it need AI like I don't want to have to deal with Salesforce is such the perfect example of where these
1070. 2:46:48tools need to disrupt and destroy the complexity of the world it's the perfect example ugh Salesforce it needs AI so bad like it can't come
1071. 2:47:01soon enough hey lightning the the new interface lightning it's the slowest most horrible piece of [ __ ]
1072. 2:47:10oh my God does it need AI hey I just put LegalZoom out of business yeah and a lot of lawyers AI is an amazing tool I use
1073. 2:47:18for brainstorming projects creating outline saves hours it does hours days I'm autistic and honestly getting chat
1074. 2:47:26gbt was so exciting it fact dumps and it doesn't mind pointed questions it doesn't Angela and it's it's um
1075. 2:47:36to the communication thing the thing one of the things that really surprises me about it but it shouldn't but it's just
1076. 2:47:42because of the nature of our society right now because we're so polarized there there are certain words you literally can't
1077. 2:47:50say right like right now you can't say trans right because trans just means right so there's certain words you can't
1078. 2:47:57say because we're so polarized but you can with chat GPT you can tell you can say anything and you can ask it anything
1079. 2:48:03and it will respond with Nuance it'll say well here's part of the situation here's part of the situation here's why people might
1080. 2:48:14act like that here's how white people might feel that way oh okay I can actually have a conversation with that
1081. 2:48:20to understand like sometimes you just want to understand an issue right but we're so polarized right now you can't even bring up the
1082. 2:48:29issue if you wanted to understand it and if you Google it all you find is like the SEO injected misinformation [ __ ]
1083. 2:48:36like you know whoever spent the most is is you you get their point of view but with this thing you can just say could
1084. 2:48:42you educate me about this completely polarizing issue it's like sure here's a piece of it here's a piece of
1085. 2:48:49it would you like to know more yeah tell me about this ah well here's the history of that oh great it's amazing for that amazing
1086. 2:49:00all right let's see I'm gonna do one more and I'm gonna go Bard Bart sucks that's why Bard's not on this list yet
1087. 2:49:11Bard will get on this list when it stops sucking ass all right how can we slow down globalists from taking over with AI okay
1088. 2:49:26we'll do this as the last one laughs let's see um I think meta I think meta has actually you know it's so funny it's like
1089. 2:49:41Zuckerberg was just like the like the evil billionaire forever and all of a sudden in the past two or three months like he's gonna end
1090. 2:49:50up being the voice of reason um I think the the fact that meta released their llama models to the developer community
1091. 2:50:00and the fact that there's a vibrant open source large language model Community right now um means that the globalists or just you know huge
1092. 2:50:15centers of power whatever you want to call them huge centers of power aren't going to be able to monopolize this thing um as easily like if
1093. 2:50:29if chat gbt hadn't been popular if if Microsoft said okay we're going to make a billion dollar investment in open Ai and they
1094. 2:50:36launched chat GPT and no one really gave a [ __ ] about it because not for nothing gpt3 was out for like a year and a half before
1095. 2:50:44you could you could ostensibly do with gpt3 what Chad gbt was right so it had been out for a year and no one really
1096. 2:50:51gave a [ __ ] so I mean you know people that knew about this stuff did but Chachi BT getting to 100 million users forced Google
1097. 2:51:03to reveal what they were doing with AI forced meta to do it they launched released llama Into the Wild um
1098. 2:51:12AWS is in the game you you've got all all of the major players with the exception of Apple who's just they're
1099. 2:51:20keeping a little low-key right now because I think they're going to come out with something big um like they do um
1100. 2:51:29if it had just been kind of open Ai and Microsoft and then Google and meta and they all still had their
1101. 2:51:37their AI platforms kind of behind the firewall and they're like well we're going to control all these then I think
1102. 2:51:43it would have been much easier for the the centers of power to monopolize that and and control it I think the the fact
1103. 2:51:51that meta released their their weights their large language model the secret sauce actually changes the game and and the
1104. 2:51:59fact that chat GPT was so popular so quickly it means that massive amounts of money and massive amounts of engineering
1105. 2:52:09Talent are all aimed at all this AI stuff right now so I think the speed of innovation and the speed of innovation outside of
1106. 2:52:22those controllable companies is is actually beyond the ability to control it at least for the short term maybe
1107. 2:52:29long term over 20 years so so the early days of the internet was very much like this where there were some big players
1108. 2:52:35but the promise of it was this is open and it's Democratic and everybody gets to do anything they want over time it
1109. 2:52:42kind of normalized toward commercial entities right Facebook Google right um I think with this AI stuff it's going
1110. 2:52:52to be similar and and maybe over time you know it centralizes toward the bigger companies if indeed those bigger companies have an advantage
1111. 2:53:04by needing massive amount of resources but what the open source Community is kind of showing us is that we might not
1112. 2:53:13need massive amounts of resources now that we've got the core models that we can play with so I don't know
1113. 2:53:19it's going to be fascinating I got no flipping clue I don't think we can slow it down at all though but I don't think
1114. 2:53:25it's the globalists that are pushing the agenda I think this is it's it's a remarkable technology that time that's
1115. 2:53:31time has come and chat GPT was the thing that broke the dam open so with that that's probably something good to end it
1116. 2:53:39on I really appreciate y'all hanging out with me this is remarkable that you're all here and we'll end a list of this old fart
1117. 2:53:46talk prophetically about the brutal beautiful destruction that is coming transformation that is coming it's not just destruction
1118. 2:54:01old things die new things are born that's what's coming but um I really appreciate all your time peace out
1119. 2:54:08I got him to ten thousand likes awesome man thank you until next time see you Silver Fox I appreciate everybody when
1120. 2:54:14are you back tomorrow night I've been doing these nightly I do this instead of sim racing which is which is my my love but I have uh
1121. 2:54:21uh Sim racing heel right now so and this is hopefully going to make a difference so peace everyone