aiLL 6/1/23 - Learn About ChatGPT, AI Ethics, Job Loss, & More!

Transcript

1. 0:03foreign [Music] learning lab I'm Kyle Shannon and my Wi-Fi sucks so I was live and then I was
2. 0:21not live and then I was sort of Live And then everyone disappeared and then they started to come back and then it went
3. 0:26away but uh this is the AI learning lab if you got questions about AI pop them in the comments uh if you're new to AI
4. 0:33That's the official chat gbt website right there and a couple other resources I am not affiliated with them in any way
5. 0:40other than I'm trying to get people to get curious about AI um it doesn't want you to talk about Boston apparently not uh yeah so if
6. 0:51you're asking about Boston before um this document the city of Boston it is an incredibly well thought out well organized thoughtful
7. 1:05concise useful um set of guidelines from a city from Municipal municipality I think my computer's dying
8. 1:21um so I think that's absolutely fascinating and someone was asking me about schools and I thought it was the school system
9. 1:29in Boston but it wasn't it was the city of Boston and then it's the New York City school district is reversing their ban
10. 1:39of chat GPT sorry I'm changing my I'm plugging in my laptop because otherwise she is gonna die so um yeah welcome back uh there's 18 of
11. 1:55you here now welcome all 18 of you hopefully that'll go up a bit but nice to see everybody uh this is the AI
12. 2:03learning lab if you've got questions about AI pop them in the comments below I'll do it again about them um now it looks like some folks are
13. 2:12coming in uh feel free to share this uh chat if you um think people might enjoy learning about AI chatting about it um I'm happy to do that
14. 2:23so greetings from coyote California greetings greetings indeed I appreciate it welcome welcome I'm from Denver Colorado here I'm uh I'm an
15. 2:36ex-new Yorker I'm a New Yorker at heart I miss New York real bad but we make our choices so I chose to come
16. 2:45out to Denver Denver's nice but it ain't New York I like the sauciness of New York well now we're up to 120 folks hey everybody
17. 2:55this is the AI learning lab Lab my name is Kyle Shannon um I've got I'm an entrepreneur multiple time entrepreneur I'm a CEO of a company
18. 3:06called storyvine which is a an automated video production platform a video storytelling platform and uh and but
19. 3:16it's it's not I'm adding AI to it but it's not an AI driven company yet it's coming uh we're doing AI generated
20. 3:24transcriptions so we're we're starting to incorporate some things into it um so that's pretty exciting but this
21. 3:30channel is all about getting people up to speed on chat GPT and generative Ai and answering questions so if you have
22. 3:36questions about AI ask them in the comments below I will try to keep up with them um I either get yelled at for skipping
23. 3:43questions or for taking for not keeping up with questions so I will screw up somehow I've also have been booted off
24. 3:51live twice now for Community guidelines because I've never read them and I don't really give a [ __ ] about them but but uh
25. 3:58I don't know why I've gotten kicked off but I just try to I just try to talk the truth about AI here all right what do
26. 4:04you think about the AI drone that took out its operator did it actually take out the operator did it actually shoot
27. 4:11them down like you know or shoot shoot down the thing or was it a test I don't know I I don't know enough about that but you know
28. 4:20these tools are like the generative AI tools the large language model tools they've been out for a while they've
29. 4:26been out for for a bunch of years the the really sophisticated ones like gpt4 are relatively new but you gotta think
30. 4:33that the militaries had their hands on some fancy s AI ship for a while um I would assume the Air Force thing was
31. 4:43testing but if they're doing live fire exercises with AI driven drones and the drones take on a mind of their
32. 4:50own well that's just [ __ ] stupid so they should not do that until I got this [ __ ] worked out uh I think they need to
33. 4:58uh put some safety measures in place okay uh your live is how I learned about chat gbt and now I'm obsessed thank you
34. 5:07you're welcome wait what's your name Holly Holly you're welcome Holly I appreciate the uh I appreciate the kind
35. 5:12words yeah that's why I'm here the whole the whole point here is just you know um just talk about this stuff get it out
36. 5:18there it's a little late tonight I was out at a CEO Soiree hanging out with a bunch of CEOs here in Denver uh which
37. 5:26was kind of fun but uh this is more fun I like this better than hanging out with boring CEO types um what's the longest length prompt that
38. 5:36you have used is it worth a super long prompt that's actually a really good question um so if if you want to have chat EPT analyze a
39. 5:50big long chunk of text um so so so one thing I've done one thing I did that would live in the neighborhood of a long prompt
40. 6:00was based on this text and then I literally copied and pasted my entire LinkedIn profile and then I pasted in
41. 6:08three different bios that I've written about me over the years and then I pasted in like an article I wrote so I
42. 6:15pasted in three really long things but the prompt was only based on all this crap and then I said write me a whatever
43. 6:23three year four paragraph bio um about me uh for speaking at an AI conference basically so add AI to my bio
44. 6:35right and then so so the prompt itself was actually short but what I pasted in was quite long um the the current limit to chat EPT is
45. 6:45four thousand tokens or three thousand words gpt4 will ultimately go to thirty two thousand and probably pretty quickly exceed that
46. 6:53there's a model out there right now called clot Claude 100K by a company called the anthropic and if you go to
47. 7:01poe.com which I'll go to right now it's I think I actually have it up here somewhere come on come on internet is my internet
48. 7:13sucking okay so if you go to poe.com what Poe is is it's it's a Chachi gpt4 is gpt4 Claude Claude instant 100K and
49. 7:24Claude instant are from anthropic which is a company I'm a big fan of and there's Chachi BT but Claude instant
50. 7:31100K means a hundred thousand tokens you can paste in a 75 000 Word document into Claude 100K and it will parse it I I
51. 7:43took a transcript from one of the AI salons that I run every every other week we record them and and transcribe them
52. 7:51and I pasted a transcription in it was like 56 pages and then just started started asking it to summarize things
53. 7:58about the salon so you can do that I assume what you're asking here though I know I answered a lot of uh things there
54. 8:06um but what I assume you're asking is does it make sense to do long complicated prompts like all in one shot
55. 8:16and then try to get the perfect answer out of it my my experience is that it's not good to do that what's good to do is actually
56. 8:25break it up into discreet little chunks so if you want to write the Great American novel for example
57. 8:31maybe what you start with is prompted to say teach me about different kinds of story Frameworks and it will tell you
58. 8:37about save The Cat and hero's journey and whatever different Frameworks there are and and then you can say Okay based
59. 8:45on you know why don't why don't you combine the save the cat story framework with Aristotle's theory of story I don't know sorry not
60. 8:54educated enough there why don't you think these two-story Frameworks and give me a hybrid of those because I kind
61. 8:59of like both of them and then um ask me a series of questions about my story until you have enough information to
62. 9:08write me an outline and then go back and forth with it and then say okay write me that outline and then if you like the
63. 9:15outline great if you don't like the outline say hey around the second act you know it took a dark turn you know I
64. 9:22was thinking of this more like a comedy and so rewrite the outline is that and then you know go in and have do a little
65. 9:28prompt about the you know first scene or a middle scene or whatever it is so I've found that it's better to kind of break your prompts up
66. 9:37think of chat GPT less like a Google search where you're trying to put in the perfect prompt to get back the perfect
67. 9:43answer and more like you've got a collaborative partner that's smarter than anyone you've ever [ __ ] met because that's what it's like
68. 9:5220 years of AI don't quite know what that means but yes it's been around for a while um but November 30th the day November
69. 10:0030th 2022 six months ago five months ago is the day that everything changed that's the day that chat GPT came out here's my contention
70. 10:09um back in the mid 90s when Tim berners-lee launched the World Wide Web the internet had been around for decades
71. 10:16but it was largely the territory of scientists and researchers and things like that and when the World Wide Web
72. 10:22came out it kind of made this thing available to the rest of us and that was when I kind of pounced on it and said
73. 10:29I'm going to start a magazine and I'm going to start an agency and I'm going to do everything I can to get my head
74. 10:34around this thing because I think it's going to change everything and it did um and I think this is another one so uh
75. 10:41ai's been around for you know a couple of decades now three four five decades depending on where you define where it began
76. 10:48and but chat GPT is the thing that said hey everyone who's not a data scientist how would you like to play with
77. 10:56something that's [ __ ] Magic and and 100 million people did in six weeks in six weeks chat GPT went from
78. 11:05zero users to 100 million users that's staggering uh context is everything context is absolutely everything that's in in
79. 11:13regard to prompting I assume and I agree with you [Laughter] yeah oh it was a simulated test but AI determined that the operator is
80. 11:25preventing its objective yeah I went back and shot him down that's okay good I'm glad it was a simulated test I can't
81. 11:31imagine that the Air Force was was doing a lot live fire with large language models especially if they're you know
82. 11:37basing it on you know gpt4 that came out five months ago from a private company yeah they got some they had some uh they
83. 11:47got some what do they call it some testing to do they got some testing to do there okay um played with vicuna uh oh have I
84. 11:55played with fukuna or other local llms I haven't uh well that's not true I played with what did I play with there's one called
85. 12:03GPT for all is that right that you don't need to do any sort of technical installation let me see if that's right GPT for all
86. 12:16yeah a free to use locally running privately aware chat bot no GPT or GPU or Internet required a little bit
87. 12:26um I think open source large language models are going to be huge um there was an interesting um document that was leaked from Google a
88. 12:39few weeks back about that was titled uh we have no moat meaning we have no intellectual property defensive defenses
89. 12:49that that open source models for a hundred dollars with an open source model that can fit on a thumb drive
90. 12:58um are essentially almost at the same performance level as what Google spent 10 million dollars training in open eye
91. 13:04AI spent five million dollars training um so I I think they're going to be quite a big deal also the open source
92. 13:11large language models don't have the the safety guard rails that these other things do which may or may not be a good thing all right
93. 13:20um context is everything okay let's see can you talk about how chat GPT wait can you talk about how to I guess
94. 13:31put chat gbt into learning mode well so so every time we use chat GPT they're they're using it to train and refine the
95. 13:42core model or not the core model they're they're using it to re to to fine-tune the the base large language model that
96. 13:50the training for the base model stopped at September 2021 all of the stuff that we've been doing since is
97. 13:57kind of adding to a layer on top of it for lack of a better term and um so so it's constantly learning if what
98. 14:07you're referring to is how do you train it on your data there's a couple of couple of issues there if you want to
99. 14:13train a large language model on a lot of your data then you need to use something like Lang chain l-a-n-g chain
100. 14:21which is a series of tools that allows you to embed your own data put it into what's called a vector database good
101. 14:29geeky terms here and then be able to search on those and be able to connect it to Google and things like that and
102. 14:34then you can have your own sort of version of chat GPT trained up on your own data that's called Lang chain it's
103. 14:40relatively straightforward to install but it's geeky enough that like I haven't done it yet because it's just
104. 14:45like you need to know some command line stuff and it's a bit of a pain in the ass um you can also do what's called I think
105. 14:54it's called few shot learning which is basically this let's say let's say you're a writer and you've got
106. 15:01a particular style when you blog you blog with I don't know wit and charm rather than saying chat EBT write me a
107. 15:10Blog with wit and charm what you can say is based on this blog post and then you paste your blog post into chat gbt
108. 15:19write me a Blog about topic A B C or D with the following facts and figures in it and use that earlier example for tone
109. 15:28match the tone of that earlier piece so you can train in a single session you can train chat GPT to to mimic you
110. 15:38um so it's pretty good at that again it's got this limited what's called context window you have kind of this
111. 15:43think of it like a 3000 word sliding window where anything within 3000 words as as you generate more and more stuff
112. 15:53the stuff at the top is going to get forgotten so sometimes you're working with chat GPT and it's working great and
113. 15:58then all of a sudden it's like it forgot what you were talking about that's because it literally forgot what you
114. 16:04were talking about so that's some of the limitations right now all right let's see can I explain sorry I'm trying to figure
115. 16:14out how to read this can I explain how to add wait how add-ons work what the add-ons uh do certain things can't
116. 16:24the can't chat can't do without okay so let me show you plugins plugins are interesting I tried to show
117. 16:32them last night and they they were it was not behaving GPT has been I think it's been overloaded for the past couple of days because
118. 16:42um my zapier automations are so if you're at chat GPT in order to access plugins you have to be a chat GPT plus member so that the
119. 16:53little plus sign here um means that I this is my little badge that says I pay you 20 bucks a month but
120. 17:01what you get is you get access to gpt4 then to turn on these fancy new settings you go to your settings
121. 17:11and then in your settings you go to Beta features and you turn on browse with Bing which is chat GPT connected to the
122. 17:22internet and you turn on plugins so that's how you turn the stuff on now when you go to chat GPT you have three
123. 17:29options hang on let me see if I can get this to focus that's a little better um you have three options you've got
124. 17:35default which is basically the old gpt4 not connected to the internet so that's the one that says I don't know anything
125. 17:40past September 2021 you've got browse with Bing beta um so let me let me con you know do a chat with with the one connected to the
126. 17:52internet so now I can say um what's the weather for the next three days in Denver question mark so obviously you couldn't do that if it
127. 18:06weren't connected to the internet but now it says browsing the web it says Denver weather forecast for the
128. 18:12next three days clicked on a link you can open this up and see what it's doing so it clicked on weather.com It's thinking it's thinking
129. 18:22it's reading content right so so it is off performing functions and now it's saying the weather forecast is Friday
130. 18:31the temperature is expected to be between 52 and 77 Saturday 50 and 63 with rainstorms and thunderstorms it's
131. 18:39been raining and thundering so much here in in Denver it's bizarre it's like East Coast kind of kind of thunderstorms for
132. 18:46the past two or three weeks been very very strange um so anyway that's how you use that one okay so now if we want to use plugins
133. 18:59um you click on the little plugins Tab and then there's a there's this pop-up menu at the top um where you've got these are your
134. 19:09installed plugins so I went to the plug-in store and I found things that looked interesting and I installed a
135. 19:14bunch of them and then you can activate up to three of them at a time so the two that I've got activated right
136. 19:21now are ask your PDF and Link reader and so um so what it will do is I can just pop in a URL and it will go to that URL and
137. 19:32read it so let me go find is this a PDF yeah this is a PDF so this is the Boston um guidelines for using generative AI so I
138. 19:44copied that URL so I'm going to come back over here to chat EVT and I'm going to pop in um based on this colon and then I'm
139. 19:54holding down the shift key to make some line breaks without submitting it and then I popped in my URL there and then I'm going to say
140. 20:03what are the top three recommendations spelled very wrong um from this article boom
141. 20:18and so now it should figure out that it needs to go use one of its plugins so it says using ask your PDF
142. 20:28and if I open this up here's a little chunk of code that it gave it generated a summary of the piece
143. 20:38and I said based on the document the top three recommendations are be specific about prompt summarizing text coding and programming
144. 20:55oh I see so it's got It's kind of categorized them from page four drafting documents or letters okay so so the the
145. 21:02top three recommendations are you can use it to draft documents or letters you can use it to summarize text and you can
146. 21:07use it for coding and programming and then it gives me some some bullets from it so what the plugins do is they give
147. 21:13you additional little superpowers that you you can have these combinations of them installed like for example you might have
148. 21:23some food related things like OpenTable for for making reservations and whatever the grocery one is for delivering
149. 21:32groceries and you could get calories from wool from alpha so you can make your own little applications by
150. 21:37combining the uh the plug-in so I hope that answers how those work and what they are it's taken me a while to get my head around
151. 21:44them it's it's a bizarre kind of concept that you've got this thing that you can sort of ask anything in any order
152. 21:50and then it kind of figures out which of those plugins it needs to use based on what you're asking and then it kind of
153. 21:56puts them together sometimes they work sometimes they don't all this stuff is very very new but here's the other thing
154. 22:02about plugins Microsoft just announced that Bing and Office 365 and windows 11 and everything that's going to have gpt4 built into it
155. 22:15is all going to use the open plug-in architecture so any plugins that you use in chat GPT are also going to work in
156. 22:22all the Microsoft products so this is going to be this is like the App Store on crack because all of these things are going to
157. 22:29be intermingled and intermixed and dynamically called it's going to be absolutely wild so if you want to start
158. 22:35getting your heads around it I think it's worth playing with okay this is the AI learning lab if you are new here hello welcome I'm Kyle
159. 22:43Shannon I like talking about AI crap and I'm a pain in the ass and I'm a Gen X curmudgeon and I swear um that URL chat.openai.com that's the
160. 22:55official chat GPT website there's a lot of scam sites and scam apps out there so that's the official one they have an official app now too
161. 23:04um but basically I'd say start there um start playing with this stuff start using it it's it's incredibly powerful
162. 23:10and so what I talk about here is just sort of answering questions if you have any questions about AI pop them into the
163. 23:16comments I try to keep up with them I'm old my eyes are going you know all that sort of stuff um Jim Carrey I don't know what that means
164. 23:27I don't think I look like Jim Carrey but maybe I'm making faces like Jim Carrey I don't think that was Jim Carrey at all
165. 23:34but it was something it was some bad cartoon of a character wasn't it oh wow I'm really behind on my comments so
166. 23:40here's what I'm doing now I'm scrolling back up through the comments so if you think I'm ignoring you or I'm not paying
167. 23:45attention to the latest things that's because I'm way back up at the top all right but I am trying to answer
168. 23:50everyone's okay can I explain the add-ons I just did that okay tree of thoughts and other uh I don't know what that word is
169. 24:02discretized I don't know if that means prompting methods have been shown to increase accuracy yeah there's there's a
170. 24:08lot of different methods where where the answers that a large language model gives you are going to be run through
171. 24:16other checks and balances basically like anthropic is one they've got a thing called the Constitutional learning model
172. 24:22where they write a set of rules and then when it produces an answer it checks that answer against the rules and if it
173. 24:29violates any of those rules it rewrites it until it gets it so so it follows them I think that whole line of sort of multi-step
174. 24:38um answer review before it gives you the answer I think that's going to be an incredibly common and and really
175. 24:45important thing for things like the Air Force where where the plane went back and attacked the operator
176. 24:51in the simulation okay cool that helps I'm getting better with these prompts yeah keep playing with the prompts keep
177. 24:57again with the prompts don't I I think prompt engineering is the wrong term it it is kind of like programming with natural language
178. 25:06um but I prefer the term prompt crafting think of it like a conversation that you're having with a really smart
179. 25:13intern or a really smart creative collaborator that's that's kind of what it's like and if if you start to mentally shift
180. 25:20from I've got to write the perfect prompt to let me just start talking to this thing and see if if together we can solve a problem
181. 25:30it's like it's like collaboration it's like a Saturday afternoon movie from when we were kids when I was a kid
182. 25:38we had Saturday morning TV I'm new to this love the topic awesome welcome um if if you're new to this please go to that website that's the
183. 25:48chat.openai.com is the official chat gbt website I don't get anything from that I just have it because that's the official
184. 25:53one and I think it's important for people to play with it prompts.chat is a pretty cool site that'll teach you how
185. 25:58to do prompting it gives you kind of this never-ending list of prompts that you can just copy and paste and play
186. 26:04with and start to explore how [ __ ] powerful this thing is and then futurepedia.io is just a list of lots
187. 26:11and lots of AI tools there I mean there are so many tools being released in the wild right now it's impossible to keep
188. 26:18up with it do I think open AI will go public soon I do not I think they're going to keep it private for as long as they can um
189. 26:27I I think I think being public puts them in a whole different place and Sam Walton's got it as a capped prophet
190. 26:34for-profit meaning that the investors I don't think can make more than 100 x on their investment or it might be 10x but
191. 26:42I think it's 100x might be a thousand X I don't know whatever there's some cap on the on the limit so it's not quite a
192. 26:49non-profit which pissed Elon Musk off enough that he went and started a competitor uh because because he's Elon
193. 26:58um and he did that a week after saying we should take a six-month pause on developing uh AI technology started a
194. 27:05competitor open I AI a week later I thought that was pretty pretty funny um all right it was just scrolling and
195. 27:13found this great to see you at dinner tonight oh hey nice nice Danny we were at a CEO Soiree we were just a bunch of CEOs hanging out
196. 27:24eating sort of Mexican food Talking Shop you know what you do all right uh do you see the FDA approved elon's neural link for human trials I
197. 27:36know that's crazy if you don't know what neurolink is basically you Elon musk's sort of contention is that
198. 27:44these these fanciful phones we have in our pockets um are are you know we're cyborgs already but we have a really shitty
199. 27:53interface and the shitty interfaces are eyeballs and how long it takes the signal to get to our brain and how long
200. 27:59it takes our brain to process all that crap uh and he said if you could just connect your you know basically the the
201. 28:07data in your smartphone directly into your brain uh then you can do it so it's it's a remarkable piece of technology if
202. 28:14you haven't watched videos on it it's well worth it he basically had to invent a robot company to do the surgery to
203. 28:21install these little you know thinner than hair like electrodes into your brain and it literally sort of punches
204. 28:28it in and figures out where the blood vessels are and punches the electrode in between the blood vessels it's insane
205. 28:35um so anyway yeah so they're doing human trials on that and there was also there was a guy in was it France or Greece somewhere in Europe
206. 28:42um a guy been paralyzed for 12 years uh that they they used large language models to read his brain scans and had
207. 28:49him think about moving his legs then they did a Bluetooth Bridge down to an um and a set of electrodes on his spine and
208. 28:59now he's walking so they basically use AI to bridge a motorcycle injury after 12 years it's amazing this is so hard but I'm trying tell me
209. 29:10what's hard about it I'll do what I can to help um just just keep trying just keep playing here's here's one of the
210. 29:15tricks of if if you're if you're a little flummoxed about what to do with chat GPT it's very overwhelming when you
211. 29:21have a tool that can essentially do anything from like let's talk philosophy of Plato to you know tell me about
212. 29:27quantum physics and anything in between um it can be very overwhelming so one thing that I found that helps is pick a really
213. 29:35small project in your life like a specific project I want to eat better okay have it create a new nutrition plan
214. 29:42for you and then have it generate um the shopping list and then have it organize the shopping list by section of the grocery store
215. 29:51um to your specific grocery store and it will probably do it uh but but pick small little discrete projects that you
216. 29:57have to solve a problem and then use the tools to solve that and you'll get better at using it because otherwise it
217. 30:04just gets completely overwhelming is the model's capacity to remember and build conversations limited by the max
218. 30:10amount of tokens in a single session yes if you use something like um Lang chain which is a series of tools strung
219. 30:19together so Lang chain you've got the large language model right so think chat GPT GPT three or four you've got an embedding engine where
220. 30:29you're taking some amount of data and you're bet embedding it you're making it sort of large language model friendly
221. 30:36you've got a uh what's it called a vector database that stores that the vector databases where you store the
222. 30:43memory um and then you've got like tools like connecting to Google and things like that so Lang chain strings these
223. 30:49things together and it can have longer term memory but if you're just talking about chat GPT it's limited to the
224. 30:56context window which is is three thousand words four thousand tokens so yes that's if you're talking within a
225. 31:02single session within chat gbt however long your context window is that's kind of how long your memory is which can be
226. 31:09a pain in the ass if you're if you're working on on something long generating lots of content you can forget the
227. 31:15earlier [ __ ] and you have to kind of go back and grab stuff and keep moving it down into the conversation um let's see llamas freak
228. 31:26oh llamas leak from meta is showing you can get 99 similar performance with local use yeah exactly this is so so if
229. 31:34you don't know So Meta Facebook their large language models like chat GPT the core models actually leaked and
230. 31:41then the developer Community was like cool we'll take those and what they actually did this was brilliant and this
231. 31:49happened within like two weeks um they took this leaked large language model and they said hey why don't we
232. 31:55take gpt4 and have it train the large language model to do what it does and they figured out a way to basically have
233. 32:03chat CPT generate all the training that trained the smaller large language model and so they're sort of knocking these
234. 32:10things out like once a week they'll come out with ones that like that are now approaching gpt4 level quality passing similar kinds of tests
235. 32:21um where they train it on like a hundred dollars of compute time I mean it's staggering and and they're they're
236. 32:26they're optimizing them one of the things they figured out they call it quantit quantization I think is that
237. 32:32what they call is that the right number of syllables quantization I think yeah um basically where they're they're
238. 32:39stripping out like 90 of the data out of these large language models and they still work like they still work
239. 32:45remarkably well so what what they're doing is like pulling out 90 of the volume of data compressing them and and
240. 32:52making like five ten you know 50 megabyte installs that you can install in a phone and run on a phone
241. 33:00um and and they're incredibly high level of of performance it's amazing um okay check out text generation web UI
242. 33:10web UI similar things exist for image gen but it makes it easy okay I will do that what types of jobs will be displaced I think almost
243. 33:25every knowledge working job creative job consulting job anything where you're using knowledge is going to be disrupted massively like
244. 33:37massively and then I think any job that's impacted by complex systems so think you know I.T security um those are all knowledge workers
245. 33:52Logistics um all sorts of things like that where you might have people who've got jobs where they're doing data entry or data
246. 34:00analysis or things like that where kind of the low level execution of of all of the really complicated stuff are
247. 34:07probably going to be automated out and and and just disrupted away but you're gonna you're gonna see disruptions I I
248. 34:13mean I'm already starting to see it in the advertising world where um the just the rules are changing kind
249. 34:21of in real time for what you can charge for creative and how long how long you can take to turn it around like clients
250. 34:28are already expecting higher quality results faster and cheaper there used to be this old adage in the ad game called
251. 34:35you you know good fast and cheap you can have any two of the three right it can be fast and good but not cheap or it can
252. 34:43be fast and cheap but not good well with generative AI you can have all three um and I think clients are starting to
253. 34:50know that so I think I don't think that any jobs are safe safe from disruption um largely speaking um I almost want to let AI save Earth at
254. 35:02this point I you know what I'm with you like I feel like who's that Alexander yeah I I feel like we've we've certainly
255. 35:09here in America but but it seems like we're we're kind of tanking this thing globally as well it just feels like
256. 35:16we've gotten so divisive and weird in in this country that um as painful as a reset is going to be I think we need a reset and I think kovid
257. 35:25was the beginning of that reset and I think generative AI is the second half of that reset the good news about
258. 35:30generative AI is that while it's going to displace a lot of people and disrupt a lot of jobs it's also putting these
259. 35:39superpowers in the hands of the people that are being displaced so the implication of that the way I think about it is okay
260. 35:48um someone decided not to learn Ai and they got laid off from their job or they just you know were doing their job and
261. 35:53all of a sudden AI you know the boss said hey we're gonna have ai do that now good luck um a lot of those people are gonna be like
262. 36:01wait a minute and and uh you know I didn't ask for this and they're gonna find that job doesn't exist for them anymore
263. 36:09and some people are going to be you know upset and pissed off and sad and kind of stay in that place right they'll go into sort of
264. 36:17um just kind of Despair I think some people will be um you know displaced and they're like okay well I'm Gonna Learn these tools and I'm
265. 36:28gonna go back and I'm going to compete against my ex you know my ex-employer so I think we'll see some Innovation like
266. 36:34that and then I think you're going to see a whole other wave of people that are like they get laid off from the job
267. 36:38it's going to be initially really painful and then what they're going to realize is I hated that job anyway I'm
268. 36:46glad I don't have that job you know what I've always wanted to do blank and then they'll use generative AI to to
269. 36:52completely reinvent themselves I think we're going to just see an absolute explosion of people Reinventing
270. 36:59themselves inventing things creative expression I I think I think we are entering what I'm calling the great
271. 37:06Renaissance that's going to make the traditional Renaissance look look quaint by comparison um I Heard it can access hyperlinks
272. 37:14posted yeah that's I just showed that before so now I've got two plugins installed one's called ask your PDF
273. 37:21and another one I have turned on is called link reader and you can basically just go get any website so let me pop over here to the Twitter
274. 37:33let's go find a an article let's see oh that's just my profile
275. 37:48um let's see let's type in white paper by the way I'm in Denver and the Denver Nuggets won their game against the Miami Heat tonight we're
276. 38:02going to see a a uh a Denver championship well I mean we are seeing the championship are we going to see them as
277. 38:09Champions I can't get it to focus anymore come on God dang it there we go um let me go grab a white paper advancing the next generation of
278. 38:23download it now okay this is a white paper is there anything here oh download white paper it's gonna make me fill out a form no okay
279. 38:34so that's PDF so I can just grab that URL and I can pop over here to chat GPT and go based on this and there's the URL please
280. 38:50tell me what this is about as if I were a five-year-old Bang and then assuming it works it should
281. 39:06fire up one of the plugins okay using ask your PDF it's doing it and let's see what it thinks a 50 year old should know about that article
282. 39:17I don't I didn't even look at what that article was called so yeah I got nothing so let's see what it comes up with
283. 39:26I'm sorry but there was an issue processing the PDF all right so that one didn't work uh [Music] upload your document oh that's
284. 39:36interesting I have a feeling click here to upload your document I've never seen this before I don't feel like doing it anyway so so
285. 39:45yeah so what the plugins do is is they kind of extend what what chat GPT can do uh um it says chat GPT you can't refer
286. 40:01it says chat EBT can't refer online website yeah so if you're not paying for chat GPT plus you don't have access to
287. 40:10gpt4 if you do have access to to plus then you can turn on web browsing the other thing you can do is you can use
288. 40:17bing.com so if you go to bing.com it says ask me anything there's TPT and there's sort of a more creative version
289. 40:30a more balanced version and a more precise version this is gpt4 and it's connected to the internet so
290. 40:38um so yeah you can use all that and that and Bing is going to support plugins so if you're new here that's the official
291. 40:45chat GPT URL so go there and start playing and then ask me any questions you have about AI it repeats itself
292. 40:53sometimes too yeah it does it also lies a lot I like to think of chat GPT as mansplaining as a service because it will confidently tell you
293. 41:02what it thinks even if it's wrong jump scare you mentioned zapier do I like it how do you use it so I've been
294. 41:11using zapier to create I'll show you a couple of things I've done with it one organization that I'm actually the
295. 41:18chief generative officer for I don't know that there are any other Chief generative officers in the world I think I might be the first
296. 41:26but we created this thing called The Challenge engine which you can give it a business challenge so this is a little
297. 41:32Google form here that's embedded on the website you put in your email your business category you give it a
298. 41:39challenge that you're facing answer a couple of questions about it and then within a few minutes let me pop over here to Gmail
299. 41:51within a few minutes of asking the challenge engine about your challenge
300. 42:05so I gotta find one of the answers here challenge engine response it generates an email that talks about you know it restates your hang on
301. 42:23there we go it restates uh what your challenge was and then it provides you with three questions in six different
302. 42:31categories and so zapier takes the input of the Google form it sends the the data that it collected in a prompt that that I created with the
303. 42:43team at content Evolution it sends that to open AI um it gets back a bunch of raw content and then I basically send that raw
304. 42:52content back to open AI to do all the the HTML formatting so so all of the HTML formatting in this email was done automatically by um
305. 43:03buy some just some decently clever coding and buy a a nice prompt a well-constructed prompt to open Ai and
306. 43:12then it takes the output of that HTML pops it in an email and sends that via Gmail so that's one of the things I've
307. 43:20built so I've been building these these content generation engines um with zapier do I like it it's a janky piece of [ __ ] but they've
308. 43:29got like 5 000 Services connected to it um it's the the challenge with all of the no code um tools is that they are they
309. 43:41to make them no code they have to put a lot of interface in front of the code which makes it unnecessarily complicated so you kind of
310. 43:49have to learn all of the intricacies of how their no code stuff isn't coding which is kind of like coding like it's
311. 43:56it's probably about as easy to learn coding as to learn zapier but not really once you kind of learn the quirks of
312. 44:03zapier you can knock stuff out pretty quick the problem with it is just you're using what zapier effectively
313. 44:10does is it is it links together multiple API services so you're taking multiple cloud services and just kind of stacking
314. 44:17them on top of one another to make these automations if any one of those cloud services goes down your whole application kind of
315. 44:25fails I mean but that's that's cloud computing in general but when you're using something like zapier if if they
316. 44:30change some code or the third party changes code it messes stuff up but I like it it's pretty good there's one out
317. 44:36there called make m-a-k-e that looks good looks really pretty I think it's newer so but it's probably got some some
318. 44:44uh you know some some quirks to it as well but it looks prettier than zapier um others need to download to use I don't
319. 44:53know what you're talking about there um maybe you need a very maybe you need a new provider you unfortunately lagging very often yeah my
320. 45:02um Wi-Fi here is horseshit it's Xfinity and it's bad if anyone knows anyone at Xfinity tell them to get their [ __ ] together in North
321. 45:12Denver it's really bad it's really bad can you upload files uh you can't right now on chat GPT but there's a there's a
322. 45:21a mode coming in chat GPT called code interpreter where you can upload files and it is remarkable um zapier last I used IRT I don't know
323. 45:34what that means IRT in in real time [Music] ascots analysis for the Ukrainian Russian conflict nah I've I've been I've
324. 45:47been booted off of Live Twice now for seven days each for Community guidelines I have no [ __ ] idea what I said but
325. 45:54the second time I was booted off I was talking about um someone asked me about bias in the models and I was explaining how the bias
326. 46:01might get in there and and just in talking about it they kicked me off so I don't know I'm not going to get into
327. 46:06anything politics you can explore that all you want uh uh uh uh uh map or esri integration ideas I don't know what
328. 46:18those are but you know what I can do what oh wait I'm just gonna go to normal I'll go to I'll go to uh chatgpt with
329. 46:33Bing what our map what is it ersi e s r i n Integrations I have a degree in acting so I don't know a lot of the super technical stuff here
330. 46:47environmental systems Research Unit GIS software okay oh map map okay so basically mapping software um
331. 46:59boy like a lot of ideas I think I think I think you're gonna see Google come out with a [ __ ] ton of AI tools sitting on top of Google Maps
332. 47:10um but you but you know so so let me let me see um what are some scripting or coding languages uh for mapping
333. 47:24which are the most commonly used so let's see what it's going to tell us here so here's some so python okay
334. 47:36Python's a scripting language sure of course python does everything JavaScript um sequel yep
335. 47:50um can you make me code in uh let's see yeah I'll do python um mapping census
336. 48:07data to major U.S cities that's a really crappy prompt but let's see what it gives me to map Census Data you'll need a few
337. 48:20different components the Census Data itself the list of cities you can do geo you're going to need libraries geopandas
338. 48:32matplotlib and pandas libraries and then it wrote the code and then it explains what the code does this is a very basic example
339. 48:41now it's talking about creating a census data CSV and how you would use that so so yeah I think it gives you
340. 48:49explanations of what to do so so yeah I think you can do all sorts of crazy stuff with it and and with actual programming
341. 48:57um where you're doing some coding and connecting data sets together I think you can do remarkable things but I
342. 49:01Google's going to come out with some crazy stuff um also Apple has their worldwide developer conference announcement on Monday Monday at
343. 49:1310 a.m PST ific time I think it's 11 A.M Denver and um they're going to be announcing their fancy AR goggles
344. 49:23um that are probably going to kick the [ __ ] out of meta's uh whatever the hell those are called their VR goggles that
345. 49:31um are kind of big and bulky but I think they're also going to be announcing some AI stuff uh maybe things like agents like one of
346. 49:41the things they've already announced is that in the next IOS operating system you're going to be able to clone your
347. 49:47voice and I would be very surprised if they didn't connect that with their memoji technology where you'll be able
348. 49:53to take you know some version of your face some version of your voice and create a generative clone like like an
349. 50:00avatar that's that sounds like you looks like you um and they're going to do some interesting stuff with generative agents I do have a
350. 50:09feeling all right a resume plug-in let me upload a file to chat GPT that's cool yeah there's so many plugins right now I'm very behind
351. 50:18on I think there's 160 or some some crazy amount of plugins let me go see how many are there um plugins plugins plug plugins go to the
352. 50:30plugin store um all so there are 56 times four pages so 224 224 plugins it's a lot of plugins
353. 50:47it's a lot of plugins it's a lot of plugins I don't know if you knew this there's a lot of plugins why is he talking like that I don't know
354. 50:55it's a lot of plugins a lot of plugins all right Quest bought us West okay bias is inherent I agree with that although I think anthropics
355. 51:05constitutional learning model um it's not going to get rid of bias but I think what it will allow you to do is
356. 51:13they're going to be different models where you can opt in to the bias of your choice if you want one that leans more
357. 51:19this direction or that direction you can do that or if you want something that's particularly neutral you can do that
358. 51:24don't ask it for code why not it's it's brilliant at code all right what else do we got here other questions sorry I lost uh Tick Tock if
359. 51:36I'm scrolling when new questions come in it jumps me all the way to the bottom and then I lost where I was in my little
360. 51:41line it's it's annoying it's annoying this Tick Tock thing Universal basic income I think that was probably in response to me talking about
361. 51:55Job disruption um I think Universal basic income is going to be a tool that's that's important I think it's going to be a short-term
362. 52:04important thing um I finally understand it Andrew Yang was talking about it in the last presidential election and I I was like
363. 52:12conceptually I'm like yeah okay that that sounds right but I didn't quite get it and uh now that I've been digging
364. 52:18deep into this AI stuff I'm like oh I get it is chat GPT plagiarism so it's well okay that's an it's a really interesting question um
365. 52:30the model itself is trained on kind of all of the internet all of the public internet there's a bunch of stuff out
366. 52:37there that's copy written so it's trained on some copywritten data and then it's trained on lots and lots of
367. 52:42non-copy written data So within the model within so think of it kind of like the large language model as a brain the
368. 52:50brain went to lots of universities and learned lots of [ __ ] right and a lot of the [ __ ] that it learned is copywritten
369. 52:56[ __ ] and so if you ask it to give you some specific you know copywritten thing it will give you things that look like
370. 53:03that it might quote it accurately it might quote it inaccurately um but you can get it to write in the style of a certain person and and
371. 53:11there's a very much a gray area there but the one thing that these large language models are not doing
372. 53:17they're not copying and pasting chunks of copy written text and and merging them together that's not how it works
373. 53:25they're literally typing or sort of retrieving word by word the next most probable word in a whole string of words
374. 53:34so you type in a prompt and and when you submit that prompt it figures out mathematically based on all of the words
375. 53:41that you said there what is the next most likely word in all of the words that I know which is like all of the
376. 53:48words across the internet is the next most likely word and then what's the next most likely one after that so it's
377. 53:54literally generating original content out of this massive set of knowledge um the other thing about the the way
378. 54:05you actually use these tools versus how they're written about in The Press a lot a lot of how how these tools are
379. 54:14described as like the robots are stealing you know are stealing from the humans and the robots are doing this
380. 54:21that's not quite right the humans are using this tool called a large language model which is a bot like thing but it's
381. 54:31a it's a large language model it's a prediction model actually humans are saying hey all of this vast knowledge can you help me write a
382. 54:40paragraph that sounds like Emily Dickinson and out it will come is that plagiarism no because it's writing original content is it is it a
383. 54:50very much a gray area that this the the base model is is trained on all this stuff I think it's a gray area because
384. 54:57it's kind of like this if I went to we'll take images for example if I went to art school for for four years and I
385. 55:04studied pop art and I love Andy Warhol and I learn everything I can about Andy Warhol and I get out of school and and you know
386. 55:13I'm totally inspired by Andy Warhol and I create pieces that look like Andy Warhol um is that stealing um well you know
387. 55:22depending on I I think is it the Keith Herring it's it's one of the one of the the foundations that for one of the
388. 55:28artists that have passed sues the [ __ ] out of people if if pieces look too much like that artist
389. 55:36um others don't really give a [ __ ] um so you know I think it's very much a gray area I think the IP attorneys are gonna
390. 55:42make a fortune in the next 20 years um I'm I'm not worrying too much about any of that stuff I'm just trying to
391. 55:49keep my head down and do the work um can I AI help bury a body I mean it's actually an interesting question there are all sorts of safety
392. 56:00um guard rails built into chat GPT that if you ask it what it thinks is dangerous stuff it will avoid doing that
393. 56:06but then there's a whole hacker community that works around what they're called jailbreaking chat GPT to do all
394. 56:13the stuff that it's not supposed to do uh if you want to um if you want to explore that go to Reddit and and search for uh chatgpt
395. 56:21damn d-a-n do anything now so there's sort of this whole battle going back and forth between the open AI team and this
396. 56:29group of hackers on Reddit writing these Dan Bots which are basically just super long sophisticated prompts that trick
397. 56:35chat GPT into ignoring its safety guidance because why not because this is how we play in the internet worlds all
398. 56:45right I've been working on AI work I was wondering what was up with the tokens yeah tokens are a pain in the ass right now
399. 56:52um I saw a thing today if so Nvidia just announced a new chip that some crazy AI you know happy thing um we're probably in not too long an
400. 57:08amount of time we'll have context lengths of like a million tokens like it will be it's going to be kind of like
401. 57:14you know RAM on computers like when I was growing up in the 80s like you could you could you know you could dial up to
402. 57:20the Internet and literally watch the characters type across screen like everything was so slow and you know
403. 57:25every time it got better and better and better and better and now it's just good enough and it'll be the same thing with
404. 57:31these context Windows right now they're they're a pain in the ass limitation but but still like even with the limitations
405. 57:37what what this stuff can do is remarkable loving this convo thank you very much I appreciate that will AI be
406. 57:44the end of the world I don't think it will like here's the deal you know are there a lot of dangers there and could it absolutely
407. 57:54be you know put in the hands of Bad actors to do bad stuff with it yeah but these tools are also in the hands of the good actors
408. 58:03and I also think that we've got 50 years of Hollywood that's told us pretty convincingly that the robots will kill us
409. 58:11and so I think I think at least in America I don't know how it is globally I think there's a default fear of AI
410. 58:19just because of Hollywood that they're gonna [ __ ] kill us what's a what I think some a more interesting um
411. 58:32um sort of way of looking at the future is well what if they dramatically improve things what if they amplify our humanity and and
412. 58:41we channel the power of them in a good way so there there's a non-zero chance that it goes incredibly good as well
413. 58:50right I think we all know the incredibly bad it can go I think it can also go incredibly good and that's things like
414. 58:56solving disease solving energy um you know cancer genetic diseases neurological diseases um just scientific breakthroughs you know
415. 59:11happening on a weekly basis that historically would have taken decades I you know I if you look at the advancement in what's happening in AI
416. 59:21right now that's what's happening like every week I do a newsletter called everyday Ai and every week is dozens of
417. 59:30remarkable things that are happening um so you know it could it could kill us all but I'm I'm trying to live on the other side of
418. 59:40that because otherwise it'd be pretty [ __ ] depressing what's the cost for me to walk again I don't know if you're
419. 59:45not walking I'm sorry about that but you know I they just I mean they just successfully did the first Bluetooth bridge between
420. 59:54the brain they they jumped um with Bluetooth they jumped a spinal cord injury and and that guy's walking
421. 1:00:01after 12 years of not walking I I assume it was done at a university um there's limitations to it obviously
422. 1:00:08but but it's the first time that they've done that so um so I think breakthroughs are coming in that which is really really amazing
423. 1:00:15what's the base AI tool that can predict outcomes of football matches um you know start using um when code interpreter comes out dump
424. 1:00:26all the all the football match data into code interpreter and go crazy um tell me again how llm is used to create another llm I don't quite
425. 1:00:37understand it exactly but so how they train these models is they they use what's called human reinforcement learning or something like
426. 1:00:48that human something reinforcement learning so they they build this base model and then they start asking it questions and
427. 1:00:57and and there's a bunch of tag data in in the set that they train and so there's part of it is getting good data into the base model
428. 1:01:07that's when Sam Altman talks about open ai's Advantage he said they spend a tremendous amount of time and money
429. 1:01:13getting the right data into the model so the the first thing is getting the the base data right and then they use humans to kind of
430. 1:01:22write questions get answers back and then say yes that was good or no that wasn't good and then I assume they also put in like
431. 1:01:31here's what might be better and so after tens of thousands of this human reinforcement learning that was good
432. 1:01:39that was bad that was good that was bad that was good that was bad the the large language the fine tuning of what's
433. 1:01:45called you got the base model down here you got the fine tuning up here at a certain point there's this Tipping
434. 1:01:51Point where it just gets really good like like remarkably good so so what the what the open source community did was they basically said
435. 1:02:02okay we've got the base model that meta built and it leaked so why don't we use gpt4 rather than having humans do all
436. 1:02:10that typing and we have to pay them 5 or 10 or 15 bucks an hour why don't we just have chat GPT simulate humans and have
437. 1:02:17chat GPT write the questions and answer if they're good or not and so they're basically using a fine-tuned trained
438. 1:02:25model gpt4 to train an unfine-tuned you know open source model with no restrictions on it so that's how they're
439. 1:02:31doing it it's remarkable Geeks are amazing when when Geeks get their mind put their mind on something you know
440. 1:02:38they just you know the hacker mentality is amazing what do I think about companies using their own version of
441. 1:02:44chat EPT to help their their software um engine yeah or engagement or whatever oh Engineers um yeah I I think that I think that
442. 1:02:53personal data sets or company data sets internal data sets are going to be massively powerful I think uh Microsoft
443. 1:03:01has announced that that large language models are going to operate at the operating system level so Windows 11 is
444. 1:03:08going to have you're going to be able to just ask it you know fix my you know Windows like configure this printer
445. 1:03:15driver whatever it might be and it'll just go do it so these language models are going to be throughout the Microsoft
446. 1:03:21suite and I think they're going to make it very very easy for companies to train on their own data I think Google will do that as well
447. 1:03:28um so I think that's I think that's going to be huge um how do you break into AI as a business analyst start learning AI start using so
448. 1:03:37what I would do if if you're a business analyst is I would start using chat GPT to support you in your work um have it help you do research
449. 1:03:49um triple check the research there was a lawyer that was just busted he wrote a legal brief that cited
450. 1:03:58citations of of you know cases and the judge busted him because he just did it on chat GPT and it made up some cases
451. 1:04:07that didn't exist and the judge was like uh you referred me to these cases and these are all [ __ ] what's going on
452. 1:04:15um but but I would start using chat gbt to support you in your business um and and just start learning about it
453. 1:04:22and start talking about it because your customers are going to start asking you about it medical coding I don't quite know what
454. 1:04:29the question is there but yeah um any job that has a routine task I agree with that um it's already massively affected my
455. 1:04:37job as a software engineer yeah I'm hearing what I'm hearing uh I heard this tonight I was at a CEO Soiree and uh the
456. 1:04:46there were some investors talking on a round table and they said that the their portfolio companies are seeing 30 to 40
457. 1:04:53percent increases in efficiency uh because of chat GPT um set grad student here how can I have chat GPT do statistical analysis how
458. 1:05:04would I share a spreadsheet okay so that's Chad B so Chad B go look for I'll I'll just go hands on it but go look for
459. 1:05:16let's see we'll hop over here to the YouTube and you're going to want to go search for code interpreter
460. 1:05:32chat GPT and then there's going to be a whole bunch of videos talking about code interpreter and what code interpreter is is mind melting so
461. 1:05:47with code interpreter what you'll do is you'll go to your chat GPT window like you normally do and in your little box
462. 1:05:53where you type in your query there's there's an icon where you can upload a spreadsheet or a document and when you upload that document
463. 1:06:02um chatgpt learns it and I think you can put in like 100 megabyte which isn't huge from a data perspective but pretty beefy piece of data
464. 1:06:13and it just learns it and then you can start asking it data analysis questions like what's interesting about this data what are the trends
465. 1:06:22um okay how do you compare this and that then you can ask it things like could you make me a map of the United States
466. 1:06:28where you change the height of the state based on the Census Data or whatever data you want and they will then write the python code
467. 1:06:37to generate a map of the United States with the states at different heights generate the image and hand that back to
468. 1:06:42you like it's mind-blowing so go watch videos on code interpreter uh and and just be patient grasshopper it will come
469. 1:06:51when it comes pounce on that [ __ ] there are probably some third-party companies out there that are using things like
470. 1:06:57Lang chain to allow you to upload spreadsheets and and play with them also if you haven't played with there there
471. 1:07:05are now plugins for like Google Sheets there's one called GPT for Sheets there's probably a ton of them now where basically every cell in
472. 1:07:14your Excel spreadsheet can be a call to open AI so so you can have like put one word in a spreadsheet and then the next
473. 1:07:21cell over says based on this word generate 10 other words and then you know the cells to the right of that can
474. 1:07:27can generate their own things based on that so you can have a spreadsheet that self-populates original content so if
475. 1:07:33you want to make synthetic data you could set up a spreadsheet in probably 15 or 20 minutes that generates you know
476. 1:07:40massive amounts of synthetic data that's exactly what you need it's amazing everything that's coming is
477. 1:07:47so amazing if you know what I mean it's amazing it's amazing the what's coming on the Internet is the Mad thing
478. 1:07:56all right can you tell it's late here and I'm a little Punchy and I had a drink or two at the CEO Soiree all right
479. 1:08:03why is AI not working in my language I would love to help I would love to help to improve it how can I do it okay I don't know what
480. 1:08:13language where where you are but that thing chat.openai.com understands all languages it should work in your language
481. 1:08:24um it's remarkable like it you can just tell it to translate this thing into that language and it does it I had a guy
482. 1:08:31I told I told the guy I said it's probably good enough at languages that it can invent a language any any those
483. 1:08:38were my tick tock on on my recorded things any and he sent me back a comment he said he said yeah I I uh had an
484. 1:08:45interaction with it were we talking about Star Wars and I had to invent a language for me so it should be able to
485. 1:08:51understand your language my guess is that you're using a non-official um app that someone put together that's just a piece of [ __ ] so
486. 1:09:01go there chat.openai.com and you should be unless it's blocked in your country you should be able to use it how do you
487. 1:09:09break into okay that was the business analyst one uh uh um why is it called generative AI okay so GPT those three words chat GPT where
488. 1:09:21does that come from generative pre-trained Transformer for Transformers I'll get back to generate in a second
489. 1:09:29pre-trained means so historically with with large language models or with with machine learning you had to bring your
490. 1:09:36own data so there were these tools out there and you could bring a massive data set and then you could put it into these
491. 1:09:42large language models and then you could search on it pre-trained means it's pre-trained meaning it's trained on what in the case
492. 1:09:50of gpt4 kind of all of the internet all of the public internet um Transformer what's that about that's a piece of technology that Google
493. 1:10:00invented in 2017 that allowed these large language models to be trained on massive data sets so so basically before
494. 1:10:09you had to have small data sets the Transformer allow you to basically stack Transformers and and it was kind of like
495. 1:10:16the more data you threw at it the better it got and that's what openai got really good at once they once openai sort of
496. 1:10:22cracked the nut of the more we throw at this thing the better it gets they said let's throw as much as we can at it and
497. 1:10:29it just kept getting better and that's kind of where why we are where we are so so generative means it's literally generating word by word
498. 1:10:38the response to your query using a pre-trained data set that's sitting on top of these this Transformer
499. 1:10:45architecture so that's what those three things mean generative means it's generating based on your input original
500. 1:10:51content or original images essentially from nothing um one word at a time or one pixel at a time in the case of the images it's it's
501. 1:11:00pretty remarkable stuff the more you learn about how it works it's just like like your face just goes you're like
502. 1:11:06holy crap who invented this [ __ ] back to school training entrepreneurial Innovation Universal basic income
503. 1:11:16um kind of I'm weight loss you know what I'm going to do I'm going all the way to the bottom here and I'm just going to
504. 1:11:21grab a couple of the questions most recent questions thank you for that look at that cool little heart hat thank you
505. 1:11:27whoever sent me that all right how do I use AI to expedite social sciences research and article finding um I would do something like krita
506. 1:11:38zapier automation to you know anytime you know some something is tweeted or posted on LinkedIn that's got certain
507. 1:11:49keywords in it I would sort of use zapier to flag articles and then um pull them in and um pull them into openai to summarize them
508. 1:12:03and then like I would use zapier to create automations because I think you could do lots of stuff with that you
509. 1:12:10could dump all the all the stuff it finds into a spreadsheet and then you could do analysis on the spreadsheet I I
510. 1:12:16think there's lots you could do there GitHub copilot has statistical and program assistance too oh oh yeah that's
511. 1:12:22that's that's true so there was a question someone was asking about statistical stuff uh depending how geeky
512. 1:12:28you are go over to GitHub because co-pilots got some pretty powerful um pretty powerful tools on that front
513. 1:12:36um what happens when the AI gets the wrong info um so this is an important things thing to know it's called hallucination in the
514. 1:12:45Craft um they call it uh these models hallucinate hang on a sec um I'm apparently not allowed to drink liquid
515. 1:12:57on camera because that violates Community guideline violations it's the dumbest [ __ ] thing I've ever heard
516. 1:13:03but apparently I can call the guidelines the dumbest [ __ ] thing I've ever heard but not drink water on camera whatever
517. 1:13:14okay so what happens when AI gets it wrong it gets it wrong a lot the way the models work they're literally prediction engines of what
518. 1:13:24word is the most likely to come next in a series so it's not actually conscious of what it's writing it's literally
519. 1:13:31saying the next word should be this and the next word should be that and the next word should be that so it doesn't
520. 1:13:36have awareness of what it actually wrote I think there will be systems in the future that analyze what it wrote and
521. 1:13:43somehow keep track of that [ __ ] right now it doesn't know so if you say give me your opinion on this court case
522. 1:13:52and cite case law that supports your argument it will do that sometimes it cites case law and it'll put hyperlinks in
523. 1:14:04there too and sometimes it cites case law and those hyperlinks just go to nothing there's nothing there it just
524. 1:14:10makes [ __ ] up so what I would say is this if you're doing creative writing it doesn't matter if it hallucinates if
525. 1:14:16you're doing research um it matters a lot if it hallucinates so if you're doing anything where data Integrity or
526. 1:14:23accuracy data accuracy is important then you need to double and triple check your work because you don't want to be like
527. 1:14:31the lawyer that just got busted by the judge that said did you write this on chat GPT This legal opinion This legal
528. 1:14:37brief and he did and he got busted yes Ben can use chat GPT plugins there's an Innovative browser extension called chat GPT for Bing chat
529. 1:14:52you don't need chat GPT for Bing I don't think because Bing is chat gbt so but whatever I I am I am not a fan of extensions
530. 1:15:03right now because I think most of the extensions are just kind of these thin layers sitting on top of
531. 1:15:11um simple open AI calls most of them are just going to become features within the browsers or within operating systems
532. 1:15:19anyway so I'm just I'm just waiting I'm not doing anything with extensions what would you say is currently the best
533. 1:15:26usage case for AI there are so many like it's it's almost like I mean that's why it's so mind-blowing I mean
534. 1:15:33here's this will give you a sense one of the things I have and show you this chat so so this is awesome chat GPT prompt so
535. 1:15:48the beginning of this document is basically just teaching you how to do prompting but the bottom of this document is
536. 1:15:57a a seemingly never-ending list of prompts of how you can have chat GPT Act so act as a travel guide act as a character from any movie book
537. 1:16:12or anything act as an Advertiser act as a Storyteller act as a football commentator act as a stand-up comic a
538. 1:16:19movie critic a relationship coach a poet a rapper a motivational speaker right this just goes on and on and on every
539. 1:16:27one of these is a rabbit hole you can run down with chat GPT so the answer to the question of what's currently the best usage like I don't I
540. 1:16:37don't have enough time in the day to explore enough to know what what it's best at it's really [ __ ] good at diagnosing
541. 1:16:46diseases according to doctors but it's really bad at citing the cases that sit underneath its decisions it's really
542. 1:16:54good at law it's really good at code it's really good at creative writing um it's really good at at um at things like reading complicated
543. 1:17:05documents and and you know giving you interpretations of it act as a florist act as a self-help book
544. 1:17:15um act as a statistician act as a prompt engineer act as a password generator a Morse code translator an instructor in a
545. 1:17:24school I think education if you go look at what um what Khan Academy has done with with Khan Migo so conmigo is a GPT for chat bot that so
546. 1:17:38um Khan Academy was given access to GPT for last August August of 22 and they took until I think it was early may this
547. 1:17:47came out they launched this thing called conmigo which is this um it's it's like an instruction assistant
548. 1:17:55like a t it's like a teacher and it will teach you [ __ ] and it and if you just say give me the answer to this it won't
549. 1:18:02just give you answers it gives you like this Socratic back and forth thing to understand if you actually understand
550. 1:18:08the concepts it's brilliant so I think that these large language models are capable of being good at almost anything in the
551. 1:18:18right hands so I don't have a better answer for you there but it's good at a lot it's good at a lot a lot of things it's
552. 1:18:26so many things I don't know if you know that's because a lot of things it's a lot of things all right companies rely
553. 1:18:33on AI and it gets it wrong could end the company absolutely companies that use powerful tools stupidly will be punished we will see
554. 1:18:43many headlines of idiotic things like the the group in Samsung that put a bunch of sensitive company data on what's effectively an open
555. 1:18:55and open at the time chat GPT and you know and and leaked their secrets ah [Music] all right what other questions do I see
556. 1:19:10here oops hold please I got lost in my questions damn
557. 1:19:25cash flow Queen oh there's a couple of people sharing this live thank you for sharing the live I appreciate that
558. 1:19:30there's a bunch of you in here so if you're new here the the reason I have those things there is that's the official chat GPT website
559. 1:19:41chat.openai.com they do have an official app but there's a lot of Faker apps out there so you don't need to pay for apps
560. 1:19:47you don't need to pay for chat GPT you can you can pay for What's called chat GPT plus that gives you access to
561. 1:19:55plugins web browsing and uh gpt4 uh natively but you can also use GPT 3.5 which is still incredibly powerful and very fast
562. 1:20:06all right if this Tech is so powerful and potentially dangerous why would they open source it well they didn't so
563. 1:20:13okay so a couple of things so here's here's some of the scoop of what's going on um so Google invents this thing called the
564. 1:20:25transformer in 2017. I think as they start exploring what this stuff can do they realize that a tool that just gives you answers is a
565. 1:20:40threat to Google's business model because their business model does not want to give you answers they want to
566. 1:20:46give you a list of 40 possible links and have you go click on as many of them as possible because every time you click
567. 1:20:52they make money so they kind of mothballed this technology they kind of you know just said Ah that's good we'll
568. 1:21:00just keep this back here but they released this paper to the world this thing on Transformers and so the rest of the development
569. 1:21:07Community kind of adopted this and over five or so years people were developing things open AI was created as by Elon
570. 1:21:16Musk as a counter to Google all the stuff Google was doing with AI Elon Musk saw that as a threat so he founded um openai as a non-profit
571. 1:21:32to be an open source company to counter the proprietary [ __ ] that Google was doing so that that was kind of how those two became rivals
572. 1:21:45and then as Sam Altman realized the more compute power we toss at these large language models the better they get
573. 1:21:55we need way more compute power than we have than money we could plausibly raise as a non-profit so we're going to take
574. 1:22:03it for profit Elon Musk huffed out of there um didn't like that at all Sam Altman said we're going to try to control this
575. 1:22:11so so open AI is not open source it's not non-profit anymore it's a capped for-profit so that happened when chat GPT was released
576. 1:22:22it went from zero to 100 million users in six weeks it was the quickest adoption of software in history I mean
577. 1:22:30the World Wide Web took six years to get to 100 million users chat CPT took six weeks because it took six weeks everyone kind
578. 1:22:40of panicked Google said oh [ __ ] we got to get our AI stuff out there meta said oh [ __ ] we've got AI stuff too Amazon
579. 1:22:47said we've got a whole AI Division coming like everyone's coming out of the Woodworks Apple's announcing on Monday
580. 1:22:53I'm sure there's going to be some AI stuff in there when meta announced that they had these large language models
581. 1:22:59like chat GPT they they inadvertently kind of released um the core large language model out to the developer Community kind of kind of
582. 1:23:09in a controlled but not controlled kind of way and the developer Community went thank you very much they took their core
583. 1:23:16models the Llama models and they just sort of blew them out into the open source world and so the developer
584. 1:23:22Community has been pouncing on these open source uh or these these leaked models fine-tuning them creating their
585. 1:23:31own and all sorts of stuff like that so um so it wasn't a mistake that they're open sourced but it I think it was just in
586. 1:23:43the Panic of people of companies trying to stay relevant in the game as openai went from zero to like a million overnight
587. 1:23:53um I think that's what caused some of the open source stuff so I think that's that's what's going on there I think
588. 1:23:59what stocks would be good to invest in for the AI boom I don't know like listen I'm you know I've got a degree in acting
589. 1:24:05I'm really shitty with money um you know I think that what micro Microsoft is acting like a startup like like nothing I've seen since they
590. 1:24:14decided to adopt the internet um they're kind they're kind of gleefully um kicking the teeth of of Google um
591. 1:24:24I think open AI is really interesting they're not public I think um stability AI is very interesting um I don't know I like Nvidia they just
592. 1:24:35did a trillion dollar market cap um and show no signs of slowing so you know but it's a trillion dollar market cap how do
593. 1:24:44you spell zapier uh z-a-p-i-e-r z-a-p-i-e-r um [Music] they're trying rewarding the process now which should reduce hallucinations quite
594. 1:25:00a bit oh yeah they're they're basically putting weightings on if if it gets it right it yeah I there's gonna be whole there's
595. 1:25:11gonna be whole multi-layer systems that that deal with accuracy like anthropic a big part of why they were started was to
596. 1:25:18to deal with some accuracy stuff so mm-hmm why say hallucinate versus lying it's called marketing if you say that AI lies
597. 1:25:29no one would use it if you say it hallucinates it's like that's kind of trippy oh it's like it's on acid in the sticks these [Music]
598. 1:25:42all right it hallucinates when writing code too yeah it does listen I I mean here's the deal with this thing I my you know my experience
599. 1:25:52with using these large language models in a work environment is that it gets you to 80 percent almost
600. 1:25:59instantly and then the last 20 is up to you and you can use chat GPD to help you get from you know 80 to 100 but it's
601. 1:26:07really up to you to take the steaming pile of [ __ ] that chat apt put in front of you and turn it into something you
602. 1:26:14can sculpt it into something um some use cases it's you get to like 99 really quickly some use cases you get to 50 and then
603. 1:26:25but I would say on average it kind of gets you to 80 very very quickly guys it's past your bedtime good night
604. 1:26:32it is past my bedtime all right it's 12 20. yeah I got to go to bed um I started this late I need an AI wrap
605. 1:26:39male voice using text input which app is free for this purpose there's no free ones well I'm sure there's some open source ones if you go
606. 1:26:49to GitHub there's probably [ __ ] that you could you could install on Google collab or whatever um I use um 11 labs and did
607. 1:27:01um I think did's got some free like you can do some amount of free stuff with did so it's d-id.com or eleven labs e-l-e-v-e-n-l-a-b-s
608. 1:27:15um for some of the voice synthesis stuff let me see the title I don't know what title you're talking about if you tell
609. 1:27:22me what title I will why not include an act as oh dang it dang it tick tock yeah let's see if I can find that
610. 1:27:31question again I was in the middle of a question and I was scrolling please let me see the title that's where it was okay
611. 1:27:39why include the act as prompt it gives similar results without so there's something about the way that um
612. 1:27:51hang on there's something about the way that um these large language models work that the act as thing makes a difference
613. 1:28:07creative mornings here we go I've got I'm downloading damn it my my hard drive on my laptop is full so
614. 1:28:16everything's in the cloud so anytime I want to open something it's not there do I have that do I have it in there
615. 1:28:30yeah well it's not there anyway so so the reason they act as kind of works is there's this thing called the latent
616. 1:28:40space or vector space the latent space is when when these large language models go out and they learn all of the words
617. 1:28:51they're not just learning words as discrete entities they're learning words in semantic forests right so if you're learning about
618. 1:29:01you know I don't know dogs and and you're talking about pack animals you know that's one thing if you learn about
619. 1:29:09dogs as you know men who philander that's another kind of semantic ocean So when you say act as um you know a Freudian therapist it it kind
620. 1:29:22of points the large language model into a semantic direction that all of your prompts from that point forward sort of
621. 1:29:30live in that semantic area region so I think that's why it works better um but you you don't have to to tell it
622. 1:29:37to act as a certain thing it can improve things um I don't always do it I think one of the things that that works pretty
623. 1:29:45universally well is when it gives you an answer you have it you say review review your answer and tell me how you would improve it
624. 1:29:54um and then have it improve it that apparently works incredibly well is there automation data an information
625. 1:30:01database that AI language models base their answers on or do they actively look for info so so the core model
626. 1:30:14they stopped they basically stopped trading the core model in September of 2021 for open AI um Bard might be different but I think
627. 1:30:24it's it's around the same time so they basically were ingesting all of this data and tagging all of this data and
628. 1:30:31whatever they do to build the base large language model this is very expensive time-consuming thing so that stopped in September of 2021 but
629. 1:30:40they've been fine-tuning it and now that it's connected to the internet um so so when I go this this option that
630. 1:30:48says browse with Bing so that's now connected to the internet so so I'm now querying this massive you know data Lake
631. 1:30:57that's that includes up to September 2021 but I'm now augmenting that with with um with current data so um who won the nuggets
632. 1:31:11game tonight question mark so that says browsing the web and now it's searching Bing Denver Nuggets game result June 2nd 2023
633. 1:31:25and the Denver Nuggets won their game 104.93 and it's got a citation and that citation takes me too probably some Denver paper odds whatever
634. 1:31:36so that's how that works hope that helps hope that helps all right you're a fascinating person I'm staying tuned and
635. 1:31:45look forward to learning more from you thank you very much I appreciate that [Music] thank you for explaining absolutely open
636. 1:31:53AI was created by a bunch of people to be honest not just Elon Musk I know and he takes he takes all the credit for it
637. 1:32:00yeah it was a group of whatever seven or eight people but the point was was that if you hear him tell the story he was
638. 1:32:07like you know Larry Page is really pissed off at me because I started open AI to you know because he was doing it wrong
639. 1:32:15um but yeah there were a lot of people that that got that thing off the ground so yes I'm sorry if I just gave all the
640. 1:32:22credit to Elon he's got plenty of plenty of credit going on right now so I'm sorry I missed where you were sharing
641. 1:32:28about how to train chat GPT um yeah I'll talk about that again I I I I go over stuff a fair amount on this channel so just
642. 1:32:37hang out on my lives I've been trying to go live daily except for when I get um banned and put in detention I fear AI
643. 1:32:46I believe it will be used for pure evil and was will destroy Humanity one day you've seen too many movies yeah I think
644. 1:32:52Hollywood has absolutely trained us that uh AI is going to destroy us um I personally think that's kind of a boring future
645. 1:33:01um if that happens that happens I'm of the belief that we're we're early enough right now and because we have
646. 1:33:08access to these tools that anything positive individuals do right now with AI is going to have a disproportionately
647. 1:33:15positive effect on the future um so yes Bad actors can play with this not play with Bad actors can leverage
648. 1:33:23these tools to do really bad things but good actors can as well to counter balance that um so go do some good [ __ ] with it so
649. 1:33:32rather than just sitting on the sideline start learning it and start doing good things with it would be my suggestion I'm sure uh
650. 1:33:40I'm sorry sir you shared something I really wanted to take note okay sorry ask me again I'll see if I can say it
651. 1:33:48again I don't know what I said maybe an AI fund a blended AI fund zapier yes I use zapier to automate
652. 1:33:55all wait onboarding all my new customers and automating estimates yeah zapier is really powerful I mean you have to really
653. 1:34:03the thing about zapier is it's kind of like chat CPT it can do so much that it's hard to get your head around like where do I start
654. 1:34:09so I've just been using it to create these content generation engines which it's a cool they're powerful
655. 1:34:17thoughts on Air Force Air Force drone incident I'm glad it was a simulation um if I asked chat GPT who I am it makes
656. 1:34:24things up a lot yes that it does it it often gives me really powerful credentials I kind of like what it makes
657. 1:34:32up about me I think I might keep some of them sounds pretty accurate yeah it gets you to 80 pretty quickly I've been using
658. 1:34:38chat CPT to write copy for classic cars and they've been selling like crazy yeah because because that's the thing like
659. 1:34:45you know the large language models have been trained on all the internet all the internet all the public internet and if you think about
660. 1:34:52all of the car Geeks that have been reading writing about every obscure you know detail of every classic car for
661. 1:35:02the past 30 or 40 years all of that's in there so if you say you know talk about you know a 69 GTO Judge in you know in
662. 1:35:12the classic orange color that's my favorite muscle car from the 70 or from the 60s um it'll know a lot of details about it
663. 1:35:21you can have it really geek out I think that's brilliant it's a brilliant use of it and it's fun like that's a that's a
664. 1:35:25fun one and it's it's you know I can imagine it's incredibly good at that do you uh let's see do you have the most wait do you have
665. 1:35:36the most amazing camera lens I've ever seen no really it's just an iPhone 11. yeah it's an 11 it's an 11. it's an old
666. 1:35:43iPhone with a cracked with a cracked uh front front um when will quantum computers combine with chat gbt we're all dead most likely
667. 1:35:54I don't think we're all dead we're all dead or maybe it's curious cancer when quantum computers when when they
668. 1:36:01can start training these things on quantum computers I think you'll get real-time training so the large language
669. 1:36:05models won't be cut off at a certain date you'll you'll just get things that are trained in real time and I think
670. 1:36:11you'll start we'll start solving the world's problems Sam Altman talked about he thinks In Our Lifetime there's going to be
671. 1:36:18a single year where we have as much scientific advancement as we've had in the last 500 years so go to chat GPT and
672. 1:36:26have it give you a list of the top 20 scientific advancements in the past 500 years just look at that list or past top
673. 1:36:3450 whatever some long list look at that list and imagine a single year where we have that level of scientific advancement compressed into a
674. 1:36:43single year like that's the kind of thing where when you when you start saying quantum computers get overlaid with with
675. 1:36:50um with AI that's possible now could it just go ahead and [ __ ] exterminate us all sure it could I don't like that
676. 1:36:58future so I'm not giving that any power I give you no power negative future I give you no power uh concession stop gerrymandering and
677. 1:37:09draw Fair boundaries well it's not chat well AI is probably generating them but who's asking the AI to generate them
678. 1:37:20it ain't at the a the AI ain't prompting itself that's human beings being [ __ ] all right AI is going to make us build
679. 1:37:29stuff faster faster better I yes I think we're entering a great Renaissance we are at the beginning of a great
680. 1:37:36Renaissance we're going to look back at these times and go holy crap remember when chat GPT just did text oh that was
681. 1:37:42so stupid you could only put 3 000 words in it wow it didn't make movies back then no we had to watch movies on this
682. 1:37:50thing called streaming services what's that oh it doesn't matter wait you only had one movie for all the people
683. 1:37:58yeah Gone With the Wind was just like Gone With the Wind and it just had that one actor in it yeah it was just Clark Gable
684. 1:38:07you couldn't put Superman in it yeah that's what it's gonna be like hey I made up legal cases and citations yes it
685. 1:38:14did and the lawyer didn't check it and he's an idiot I like the Will Smith world the first half yeah I know the first half of all
686. 1:38:24the AI movies is great it's the second half that we want to avoid the best camera image ever well thank you I don't
687. 1:38:30know what's going on with my camera image but yeah maybe it's just a when you got this going on maybe it's not the
688. 1:38:37camera you know what I mean you know what I mean I just felt like I sounded like Ted lasso there for a minute did
689. 1:38:43you see the last episode of Ted lasso is bawling my eyes out God damn that was good all right I used AI to help with my
690. 1:38:49non-profit kitten rescue cool save the kittens you say so what is AI good for saving kittens see you heard it here first that's awesome
691. 1:38:59I like that all right I lost my place again I'm going back I'm going back I'm doing it uh-oh all right did you see the yes I saw the AI
692. 1:39:12drone that went rogue but it was a simulation so we're all good I went back too far I've lost my place okay I just got to the kitten rescue
693. 1:39:24thing okay I think it's about I think is about how to train AI prompts you know folks put complete sentences in and that won't
694. 1:39:35sound like such an Egypt um all right can I use AI to write my resume and make me look like a rock star yes you can
695. 1:39:45um just so go get a resume format that you like or have it recommend a resume format that you like pop that into chat EBT then go copy and
696. 1:39:55paste all your LinkedIn stuff paste that into chat EBT and say write me a resume that follows the format of the resume up
697. 1:40:03top and populates it with the details of what's from my LinkedIn profile for a job in whatever it might be
698. 1:40:12what's the Air Force incident so so the Air Force had a simulation going of an armed drone that was set to take out enemy positions
699. 1:40:23and the the the person sending instructions to the Drone basically called it back and said you can't go
700. 1:40:30kill that thing and it didn't like that so it turned around and attacked the operator now it was a simulation so so
701. 1:40:39no shots were fired but uh yeah assuming they're a missile attached to that their drone um that wouldn't have been good so that
702. 1:40:46was the Air Force incident so um good that they're testing good that they're doing it in a simulated way and
703. 1:40:53they got some Kinks to work out you know Elon Musk said we'd have full driving cars by what four years ago and it's
704. 1:41:01still not quite there but it's close well imagine flying things with bombs on them they might take a while to get it right
705. 1:41:10uh why did you say 69 GTO Judge that's creepy I just sold one oh dude I I don't know there's something about a 69 GTO
706. 1:41:18Judge that's just the most beautiful car car in the world [Music] uh do you have a website for more info wait prompts for code Generation
707. 1:41:33Um I don't but that's the the SE prompts right here so prompts.chat that'll take you to the prompting Thing
708. 1:41:41Future PD is a bunch of tools and then chat.openai.com is uh the official chat GPT website um if you're asking about information
709. 1:41:49about me I've got a link Tree in my bio that that's got a bunch of my [ __ ] there like my company and I think I put a link
710. 1:41:56to my email and things like that so um where can you get your own model to run locally on a server cluster just go
711. 1:42:06search for uh just go to Twitter and search for open source llm um and and go to go look up trending projects on GitHub and hugging face
712. 1:42:18hugging face is a place to get all those things there are so many of them out there that you know probably the place to go is
713. 1:42:25Reddit and and and just find the large line I'm sure there's an open source large language model subreddit
714. 1:42:33um and just go start hanging out with those Geeks and you'll find some some amazing stuff I'm a fantastic teacher
715. 1:42:40thank you very much I appreciate the kind words and I gotta go to bed I gotta go to bed people what what are we doing here what
716. 1:42:48are we doing why are you watching me stop stop it's not that interesting thing at all who are your people
717. 1:42:59um let's give AI the goal of making the Star Trek Society yeah I think it could I think we're gonna I think we're gonna see some really
718. 1:43:08really amazing things I think there's going to be some headline making catastrophes um but I think we're also going to see
719. 1:43:15some remarkable remarkable stuff coming out of this don't forget we're so early with this and and you know
720. 1:43:24ai's been around for a long long time so the foundation of all this is has been around for a while but what's new is
721. 1:43:32what what started on November 30th last year 2022 is when Chachi BT came out it gave regular people access to these
722. 1:43:43tools that were largely the the you know the territory of data scientists and researchers and and basically Sam Altman said here's a gift
723. 1:43:54do you know that they wrote chat GPT in two weeks do you know that in the documentation for gpt3 which came out like a year and
724. 1:44:04a quarter before chat GPT the instructions for how to build chat GPT were in the documentation for gpt3 for like a year
725. 1:44:15and Sam Altman said we assumed someone would build it and no one built it and then last November uh I think it was
726. 1:44:24because Microsoft was going to announce their billion dollar investment Sam Altman said hey uh engineering team can
727. 1:44:30you whip up that chat GPT thing they built it in two weeks and six weeks later it hit 100 million users I mean that's remarkable
728. 1:44:41fastest adoption of software in history all right I want to be a part of this Renaissance what's a good field or thing to study with AI
729. 1:44:54ubiquitous so so I would sort of push that back on you and say what is the thing you're most interested in and then I would dig deep on
730. 1:45:02How can I do what I'm interested in better faster more powerfully with AI so if it involves images start learning
731. 1:45:12mid-journey or stable diffusion or um the new Adobe generative fill if if you look at what's happening with
732. 1:45:20General Phil within Photoshop it's remarkable um if it's around words or business or analysis or programming or things like
733. 1:45:30that then dig into chat EBT but I like I I just think it can't hurt anyone right now to go to that address like why I
734. 1:45:37have it on screen that's the official chat EBT website just go there start using chat GPT to solve problems in your life um
735. 1:45:47I would say lean into what you're passionate about what you're good at um what you have a point of view on and
736. 1:45:55just figure out which tools can support you in doing that because again it's so [ __ ] capable that it can do anything if I ask chat
737. 1:46:03CPT to write a book about vampires and you ask the same question the same way will we get the same story we will not
738. 1:46:11um in fact you can do that yourself um create a new chat you know put in a decently sophisticated um prompt about your vampire book have
739. 1:46:22it writes a chapter or whatever or an outline and then start a new chat and use the exact same prompt and generate
740. 1:46:29it and you'll you'll see it'll give you it'll give give you different things um and that's largely because they've got
741. 1:46:37there's a thing called temperature if you if you use the the back end of these GPT tools I think chat GPT is set for
742. 1:46:450.4 it goes from zero to one zero is the most factual answer so if it's zero you you'll likely get things
743. 1:46:53that are all the same but if you dial that up to one you'll get things that are much more creative so I think
744. 1:46:58default chat GPT is somewhere near the middle everything using chat GPT apis are going to be paid Services right because
745. 1:47:06developers using apis need to pay for its subscriptions yeah but they're relatively cheap I mean unless you're
746. 1:47:12doing unless you're doing training and embedding which can be expensive um just the calls to open AI like I set a
747. 1:47:21Max monthly spend of 20 bucks I don't think I've ever exceeded 10 and I like I do a decent amount with this I don't do
748. 1:47:28a ton 65 Riviera is more beautiful I don't know I don't know 69 judge the the Riviera's it's a bit more of a boat it's I don't
749. 1:47:41know it's a little rounder I see where you're going it's it's maybe more unique I like the I like the muscly aspect of the GTO
750. 1:47:49which job sector will it be affected negatively from a AI from Full stack cloud and devops I don't know like all of them
751. 1:48:00I don't know if they'll be negatively impacted they're just going to be dramatically impacted the negative part is
752. 1:48:09some jobs are going to go away some jobs are going to get very very different and people might not like them there's a
753. 1:48:15really sad Reddit post about a guy that did 3D modeling and he said I hate I'm quitting my job he was lamenting his job because he's no
754. 1:48:24longer a 3D modeler he now just writes prompts and then curates images that this thing generates and he and he hates
755. 1:48:30his life and he's he's leaving the profession so I think you'll see some of that but I think I think every single
756. 1:48:36sector is going to be impacted by this especially for things like devops and a lot of that stuff the coding and the all the tech stack
757. 1:48:45um just every all that complexity is just gonna get subsumed by the Simplicity of the large language model sitting on top of it
758. 1:48:53so so this is really interesting so right now you've got this incredibly complicated technical world right with
759. 1:49:01cloud services and you know devops and you've got all these machines and platforms and services and then you have entire companies that
760. 1:49:10build software to manage and glue all this [ __ ] together right and you have entire Tech divisions to do that so some
761. 1:49:18combination of services and Technologies and human beings make sense of this madness down here well what large language models are
762. 1:49:28going to do is they're going to put a layer up here that just does all this [ __ ] automatically and so companies that serve that that
763. 1:49:38thrive on the complexity of the ship below it because they figured out how to simplify this stuff they're going to get
764. 1:49:45displaced and people who spend their days doing menial I don't know data analysis they're going to get displaced and so
765. 1:49:55that that whole middle layer sitting on top of all that complex stuff down beneath I think that's where you're gonna see the biggest
766. 1:50:03the biggest thing the algorithm has me I don't know what that means invest in stocks and cryptos and invest in metaverse as well I would
767. 1:50:15not recommend investing in crypto or metaverse right now let AI take take off so I spent about a year deep diving in
768. 1:50:23blockchain and nfts and what was interesting was the deeper I dug into that the less I believed in it and with
769. 1:50:29AI it's the complete opposite where the deeper I dig dig the more my mind is blown I think AI is going to drive
770. 1:50:36adoption of blockchain but we're going to come up with a use case first and then the technology as opposed to right
771. 1:50:42now where crypto and and blockchain and nfts are kind of leading with technology and and so many of those companies are
772. 1:50:50just money grips they're just it's just bad um is that GPT how to build guides still available GPT how to build the prompts.chat yeah
773. 1:51:05prompts.chat is I think that's the thing you're talking about you should be if be an affiliate for them I know I talk about them all the
774. 1:51:15time they should give me they should give me access to code interpreter the [ __ ] 0.7 over here it was preset to 0.7 oh
775. 1:51:25chat GPD was preset to 0.7 GT all day let's go GTO people truck drivers are the worst well the Riviera is not a
776. 1:51:33truck it's a boat boat drivers are the worst Maybe um platform is preset to 0.7 yeah so it's decently creative I thought it was
777. 1:51:41lower than that for some reason how do professionals in fields using AI set themselves apart from the folks
778. 1:51:50who can now fake the skills that's an interesting question um the part that's still important is the critical thinking and the strategy
779. 1:52:06so how you decide what what you want these tools to do and then how you're going to use it and how you deploy it that's still humans
780. 1:52:17and so if you have someone that's just using chatgpt to generate documents that look like that that's just a different thing than
781. 1:52:27someone who can look at a bunch of output from chat GPT and go that I like that I don't like based on their 30
782. 1:52:33years of experience so I think how you combat charlatans is how you always because here's what's going to happen
783. 1:52:41everyone has access to these tools so the perceived level of idiocy in the workplace is kind of is gonna kind of rise like this where
784. 1:52:54oh you know maybe Janie isn't such an idiot because her her writing quality went up because she's using a tool but
785. 1:53:03then what you're going to see is people with actual experience and brains and strategy are going to learn how to use
786. 1:53:09these tools just at a diff a slightly different level so you're going to start to see some work do this so all work's
787. 1:53:15gonna go like this and then the good people are going to rise to the top as they always do because these are just tools in the end
788. 1:53:22and the people that are smartest at using and deploying the tools are going to be the ones that rise above the noise and so
789. 1:53:30be smart one and start learning AI start getting good with this stuff so that you're one of those ones that rises
790. 1:53:37above the charlatans because there will be lots of them good thing about chat CPT is it will force a human to see that their mind
791. 1:53:45is not what are what they are is that their mind is not what they are um yeah I I think that one of the Silver
792. 1:53:56Linings of this the disruption that's coming is is that I think it's going to allow people to be a bit more free to say what do I really
793. 1:54:09want because like if what I really wanted was to make beautiful images I've got photography and I'm good at that but like when I see a good
794. 1:54:21illustrator it like makes my heart ache because I want to be able to do that but I suck at drawing I suck at it um and then
795. 1:54:30mid-journey and stable diffusion come along and I can take the idea that I have in my head of something I want to
796. 1:54:36visualize and now I can just visualize it and so I think you're going to see a lot of people discovering creative sides
797. 1:54:43of themselves or creative expression of who they are um that they might not have otherwise ever gotten to I think it's going to be
798. 1:54:51beautiful I I think people I think the amount of reinvention human reinvention that's coming is just going to be amazing
799. 1:55:03preach go chat GPT I'll tell you I tell you it's gonna save you the AI is Gonna Save You It's Gonna Save You Brethren It's Gonna Save You
800. 1:55:14all right I can do a few more questions and then I gotta go it's one in the morning here what are you people doing
801. 1:55:20to me what are you doing with your lives I hope you're drinking all right I use chat tvt to mimic a human AI Persona with the experience
802. 1:55:32and skills I use it as a CFO yeah there's a cool project there's a guy on uh Tick Tock called McKenzie Bose who's
803. 1:55:39got a project called GPT boss and he's it's basically a wrapper app that sits on top of uh gpt4 but it's all these different characters and
804. 1:55:50generators and so there's like a CFO and there's a coder an elite engineer and a copywriter and a marketing expert and an
805. 1:55:57SEO expert and so he's built all these sort of pre-prompted tools that you can use to run your business it's pretty it's pretty slick
806. 1:56:05all right what happens when the tools become so easy anyone can use them that's what we're in the middle of finding out I mean yeah like
807. 1:56:19writing uh a uh like like you can do anything right now you can do anything so so the tools are easy enough that anyone
808. 1:56:29can use them right now but you have to have a little bit of skill to sort of piece part the different things together but
809. 1:56:37within two or three years there's going to be slick interfaces on top of all this stuff it's going to be [ __ ]
810. 1:56:42miraculous so what what that means is the people that start learning it now are going to be way better at this [ __ ]
811. 1:56:49than when it gets easy to use two years from now so all the opportunity is happening right now let's call this the
812. 1:56:55summer of opportunity get curious about ai go to that website chat.openai.com that's the official chat GPT website
813. 1:57:04go to prompts.chat and just start playing with the prompts there and just learn about prompting start learning it's just a tool I'm a
814. 1:57:13senior Oracle DBA and I use it to help you write documentation yeah it's really good at documentation and and parsing
815. 1:57:19things I got chat cheap knowledge up to year 2021 guess who sings the song anti-hero accurately well you can I mean
816. 1:57:33you might be able to get it to I I mean it can't guess anything unless it's been fine-tuned to things after 2021 it can't know that so maybe
817. 1:57:43you had a trained model but gpt4 is connected to to the internet now via Bing so that is all water under the bridge the
818. 1:57:52it only goes to September 2021 thing that's over that's over because it's all connected to the internet now with the Bing the
819. 1:58:02Bing from the Microsoft Microsoft company out of Redmond Washington they made software they made shitty software
820. 1:58:09for years and then they discovered this AI stuff and all of a sudden they're like [ __ ] Apple what the hell what
821. 1:58:16the hell is going on in the world I used to hate Microsoft for years I'm an apple fanboy can I just say this Apple's got their
822. 1:58:26worldwide developer conference on Monday if they do not blow it out of the [ __ ] water and make themselves
823. 1:58:32relevant again what is going on in the world uh that was an interesting comment I think it will make real artwork even more valuable
824. 1:58:42absolutely agree with that I think that authenticity things like in-person music real art um you know video of actual human being
825. 1:58:53sharing their stories authentically I think the value of authenticity is going to go up dramatically so so we're going
826. 1:59:01to see the sea of synthesized content that just sort of boils up and I think there's going to be a hunger
827. 1:59:08and and I the the 2024 election in the U.S there's going to be so much disinformation and misinformation that's
828. 1:59:15going to be AI Amplified it's going to be insane I think the backlash of that so think you know 2025 will be an incredible
829. 1:59:27demand for authenticity and I think the value of authenticity is going to go up I think you're going to be able to build
830. 1:59:34businesses around it I think authentic Arts authentic experiences I think value goes up dramatically so I totally agree with that Holly
831. 1:59:45what's the best way to use this little wait what's the best way to use this with little technology I like again I would just go to
832. 1:59:55um just start learning chat gbt like you use it as as an Arts major as a Humanities major um have it right song lyrics have it
833. 2:00:04write poems have it help you um create outlines for a book write chapters of a book have it help you rewrite you know if if you wrote wrote
834. 2:00:15like I wrote the shitty beginning of a screenplay and I had to help me rewrite the first couple of scenes of it just to see how it did
835. 2:00:22it did great like um use it for things that you're passionate about don't worry about the tech piece of it if you want to do
836. 2:00:31technical stuff have chat GPT start to teach you that like you can start to geek out and run down rabbit holes if
837. 2:00:37you want to but leave it leave it up at the tech leave it up at the at the humanities level um it's it's remarkable at that [ __ ]
838. 2:00:47like gpt4 in particular like it's writing is really good you have to you have to get good at prompting it if if
839. 2:00:53you give it shitty prompts it will comply and give you shitty answers all right what's the best way to use
840. 2:00:59okay that was that that's fascinating all right have you tried prompt AI PRM I have not it's really good okay I'll check it out
841. 2:01:10um I've been doing some Consulting for ad agencies interested in texted video I think my summer will be short-lived well I I
842. 2:01:21think listen I think companies are going to be hungry to figure out what the hell these tools are so I'd say Consulting
843. 2:01:27with this stuff even though I mean here would be my only advice if you're Consulting with AI don't tell them
844. 2:01:32you're an expert no one's a [ __ ] expert right now everyone's trying to figure this out and everything's changing week to week
845. 2:01:39um so I would just say be humble about it just say hey I'm learning as much as I can about this AI stuff you know can I
846. 2:01:44come in and tell you what I know um and just you know try not to put yourself on too much of a pedestal because
847. 2:01:51this [ __ ] is changing so fast it's it's very humbling like I I kind of pride myself in being able to quickly grock
848. 2:01:57[ __ ] and keep ahead of stuff I'm not even close to keeping ahead of this stuff it's moving so fast [Music]
849. 2:02:08National digital banking starts in America on July 17 2023 are you ready no I'm not that's going to be interesting
850. 2:02:20to see what they do with it we'll see I'm not all that worried about it right now thanks for sharing your time and
851. 2:02:26knowledge you're welcome you're funny thank you agreed but looks aren't everything [Laughter] all right do you think there will be
852. 2:02:41a regression toward the mean AI content of AI content of AI content that's a really interesting question Krista I don't know I mean
853. 2:02:51I suppose if you well yeah it's really interesting I suppose that the point at which you pull humans out of the contribution
854. 2:03:06Matrix right if you just if you say right now humans are still very actively prompting and and refining
855. 2:03:17and crafting content in conjunction with AI if the AI gets so good that that humans increasingly kind of back out of
856. 2:03:26the process and just let the machines do the bulk of the work I think at that point you might start to see kind of diminishing quality but
857. 2:03:37but at the same time as as the tools get more sophisticated I think we're going to see the tools get trained on
858. 2:03:43increasingly large data sets like like at some point the large language models are essentially going to be the entire output
859. 2:03:55of the history of humanity and and it will be updated in real time so at that point I don't know I think I think we'll just
860. 2:04:04be able to do magical magical [ __ ] with it but I don't know it's possible I don't know I don't know enough about large data degradation to
861. 2:04:16to talk to that at all so sorry about that I'll just I just think I don't have the book smarts all right I gotta go to [ __ ] bad
862. 2:04:30people I gotta go to bed eight to see what the military is going to do with it they just shot their own well in a
863. 2:04:36simulation a drone using AI went back and attacked its Commander because it didn't like the the uh the command it gave the Drone so went
864. 2:04:48back and and shot the operator so they're gonna have to they're gonna have to tweak that little algorithm all right
865. 2:04:54all right I got a few more here for sure I mean if I welcome for anyone who's new how do you know this stuff I just
866. 2:05:01started so so here's how I know this stuff I I went through I I discovered the World Wide Web personally in
867. 2:05:11um 1994 when when I when I discovered you know what what had been out for a few years um I had a massive Epiphany that holy
868. 2:05:20[ __ ] this is going to change the world but like the world has changed it's just no one knows it yet and I started one of
869. 2:05:24the first digital agencies so I've been through this before and and this AI thing had the parallels to what
870. 2:05:33I was going through with the world wide web in the mid 90s and what's going on with this stuff now are insane
871. 2:05:40um and so the minute I had this Epiphany with chat GPT like I've been playing with language models before that and
872. 2:05:46then I had just a breakthrough with chat GPT like holy [ __ ] everything's different um I just dove in I started this Tick
873. 2:05:53Tock Channel I started an AI Salon I started in a newsletter um I I basically started saying yes to every speaking gig I could get every
874. 2:06:03Workshop someone said hey can you do a workshop I'd go do that so I've just been insufferably diving deeper and deeper
875. 2:06:13into learning everything I can about it and it's impossible to learn because there's so much of it um but but just got curious about it
876. 2:06:20because I think it's incredibly important just enjoying the ride thank you stranger in a stranger land we are in
877. 2:06:27surreal times I feel like we're in post we're entering a post sci-fi world I feel like there's just all these Sci-Fi
878. 2:06:34movies it's like check yeah we can do all that [ __ ] now check we can do most of that [ __ ] now check oh we don't have
879. 2:06:40time travel yet but you know we'll probably figure that out in the next decade um excellent perspective thank you very
880. 2:06:48much very cool thanks so much for sharing that okay listen I gotta go it is one in the morning here peace out
881. 2:06:54everybody I appreciate it thumbs up to all of you I will check back in tomorrow probably and uh I hope you're well go
882. 2:07:03there if you haven't go to chat.openai.com start using chat GPT do it good night