aiLL 7/12/23 - How Generative AI is Already Impacting Business and Society

Transcript

1. 0:08[Laughter] um [Music] [Laughter] [Music]
2. 0:27welcome welcome welcome to the AI learning lab we're gonna do some learning about AI in the lab or at least a home office
3. 0:40welcome welcome welcome welcome welcome welcome welcome welcome welcome welcome welcome welcome to the Future good day J word what's up Phil
4. 0:55how you guys doing what night is tonight Thursday Wednesday Wednesday happy Wednesday what's up earthlings good day Kevin um
5. 1:10a lot of people have showed up but Tick Tock has not yet told me how many people are here so I don't know if you all can see it
6. 1:19looks like zeros looks like none right now nothing I got nothing except I see people I see people coming in welcome welcome 25. cool is it
7. 1:29possible for AI to act like people via a bot yeah look at me I'm sorry I need to catch up to speed oh you're
8. 1:39you're being serious okay is it possible for AI to act like people via a bot uh yeah you know I think I think um
9. 1:51it you know if if you're if you're sort of chat interacting with this generative AI stuff it can they did a study about three weeks ago some
10. 2:01results came out that a bunch of patients interacted with chat GPT doctors and then they interacted with real doctors and the patients
11. 2:12largely preferred the uh GPT doctor over the real doctor they said they were more competent and more empathetic so
12. 2:20so yes you can uh welcome everybody this is the AI learning lab I'll show you what these URLs here and are in a minute
13. 2:28I actually added one it's been an interesting two or three days um chat GPT all of a sudden has some real plausible um competition
14. 2:39and uh that's that's interesting I mean I figured that would be coming and I said I wouldn't change this list until
15. 2:45some new [ __ ] happens you know till someone emerges that that knocks someone off the list and it's funny I haven't knocked anyone off the
16. 2:54list but I added one so so we'll see we'll see how that goes 108 can AI get rid of trump well he's still he's out there I don't know he's
17. 3:05he's got he's got some issues I don't I don't know that we need AI for that he seems to be managing that himself
18. 3:14and also I don't do politics here I do not care to speak about it I'm tired of the divisiveness and I'm hoping AI can
19. 3:23make us all human again and I think it's actually a possibility um these Bots actually um speak to you with nuance and logic
20. 3:40and um empathy ironically enough wouldn't it be amazing if the Bots rather than killing us were the things that freed us up to be decent
21. 3:51human beings again that'd be fun anybody done with that um anyway welcome everybody this is AI learning lab my name is Kyle Shannon a
22. 4:01little bit about my background I'm an entrepreneur I've started 12 or 15 companies in my day I was an innovator in the early days of
23. 4:08the World Wide Web I started one of the first digital agencies a company called agency.com and I was Chief creative
24. 4:14officer there and the parallels between the early days of the web and the early days of this generative AI stuff are
25. 4:21insane and it's got my mind all a flutter and that's what this channel is about just talk about what I'm seeing
26. 4:29answer anything I can if you have questions about AI pop them down below if you like this kind of content that's
27. 4:34what the channel is about so give me a follow and I see you're tapping the screen there that's great that that'll let people know we're here
28. 4:42and uh yeah let's get let's get rolling on this fantastic Thursday night Wednesday night [Laughter] all right oh I've also been giving out
29. 4:55thank you very much I've been giving out free business ideas like the last three nights it's been at least two business
30. 5:01ideas a night that that are gonna be billion dollar ideas I just give them away for free you don't even have to sign it it's not
31. 5:09even like a raffle or anything I just give them I just you just take them you take these business ideas I come up with
32. 5:15I'm like answering someone's question I'm like hey here's a business I give it away give it away for free why because
33. 5:20that's the kind of guy I am that's really more just because I have ADD and I have 4 000 ideas a minute and
34. 5:27so there's more where that came from so and and and quite frankly the idea is never the thing it's always the execution so um
35. 5:37I've learned that many day many times in my life what I've also learned is if you say an idea out loud it is happening it
36. 5:45is being started while you say it out loud it's being started at the same time and if you don't do yours three months
37. 5:51later that business will show up in the world I've seen it happen like so many it's freaky it's freaky how many times that
38. 5:59happens so if you have an idea and you you don't want it stolen just don't say it out loud but the minute you say it out loud you
39. 6:07better do it um okay so let's talk about the if you're new to generative AI if you have not played with chatgpt that's what the top
40. 6:16URL is that's the official chat GPT website and then Bing Claude and poe.com cloud.ai um are are all different flavors of it so
41. 6:30so Bing is Microsoft's implementation of chat gpt4 so you can play with you know the official gpt4 there that's
42. 6:38Microsoft's version of gpt4 Claude is a competitor to gpt4 and they just launched clod two where you can upload files and it'll
43. 6:50read them and you can interact with them so you can upload a big PDF it also has a hundred thousand token context window
44. 6:56so that means if you go to claude.ai you can input a prompt as big as 75 000 words so like your whole novel The Great
45. 7:06American novel that you wrote you can just pop it into Claude and start you know having it edit it for you give your
46. 7:12other ideas and then poe.com actually lets you play with the Claude models the open AI models and their own model and
47. 7:20plus it's got a or waitpo.com and plus it's got a bot Builder you can make your own bot and the bot now can uh you can
48. 7:30you can use gpt4 and Claude 100 000 token in your bots so the Bots that people are going to make with po.com are
49. 7:39going to be insane so these are all just different flavors of chat GPT if you've not played with one of them go do it now
50. 7:45prompts.chat will teach you how to interact with them it'll give you a primer on prompting and then give you
51. 7:50some ideas and then futurepedia.io that's just a giant pile of AI tools that I can't keep up with them anymore
52. 7:57so Matt wolf I think is his name that created that he's keeping up with that [ __ ] I can't do it anymore so I am I am
53. 8:05I am delegating that uh to some uh some YouTuber dude dude am I Gen X yes I am Gen X Gen X in the house
54. 8:20I ain't no bloomer you know what else is I'll tell you what you got to be careful with Gen X because we know how to use computers and we're
55. 8:29going to be really good at this [ __ ] so uh you yield little young whippersnappers we're gonna figure out how to use this
56. 8:36AI stuff you're gonna be like what do you mean I lost my job to an old man it's gonna happen all right
57. 8:51will AI reveal the Illuminati and the aliens if the data is there um one of the things that's really good
58. 8:59at is looking at disparate pieces of data and analyzing it and pulling it all together if you haven't played with code interpreter
59. 9:06go do that because uh if there is day to be hat data to be had that can prove that and you can get to that data why
60. 9:14indeed you know you might be able to reveal some [ __ ] it's uh it's pretty amazing stuff what it's capable of doing
61. 9:22uh yep let's Unite Here yes I am Gen X can AI make sure we get a monthly stipend since they're going to take over
62. 9:28all our jobs well AI can't make sure that we do that but we as a society better [ __ ] figure something out
63. 9:35um I don't think we're gonna need um Universal income for long I think that um my my suspicion or my the way the way I'm seeing things play
64. 9:51out I think there's going to be a two to three period two to three year period of painful disruption where people lose
65. 9:59jobs and then we're going to start seeing entirely new categories of jobs being invented because of AI so this happens at every
66. 10:09major technological innovation but this one's gonna this one's gonna be big profound and fast but I think also the the kind of the
67. 10:18prosperity kick up because of this stuff and the um new jobs that are created because of this stuff I think they're gonna they're
68. 10:26gonna kick up as fast as well so I don't think this is I don't think the pain is going to be like a decade I think it'll
69. 10:32be more like two three five years maybe and then it'll start to get interesting AI seems nicer than humans a it's Danny that's it it yes
70. 10:45especially in the states and it seems like this is not just the states but there's there's just a lot of
71. 10:51polarization going on right now so we're not taking each other in very much these tools are actually really good at
72. 10:56listening and responding reasonably to a to an inquiry um and we as humans have gotten really shitty at that in the past decade
73. 11:07um I mean I think we've always been generally bad at it but so these things are good I have a feeling that these
74. 11:11tools are going to help us be better people all right what's the xai about Elon Musk is behind it so Kevin so yes Elon Musk
75. 11:25put 50 million dollars into open AI as a non-profit and he said he did it in reaction to having some dinners with
76. 11:33Sergey Brin from Google who was a little bit too flippant about uh AI safety and so um Elon Musk said I'm going to start a
77. 11:43non-profit to make sure that someone is doing something with AI that isn't just Google because he didn't like
78. 11:49his attitude and so it started as a non-profit and then depending on who you believe I'm gonna go with Sam Altman because Sam
79. 11:58Altman actually seems reasonable and thoughtful and Elon Musk seems a little like emotional and rammy but what ended up happening was
80. 12:10as open AI learned more and more about how to use the Transformer model that Google invented ironically enough what
81. 12:16they realized is that the more data they threw at Transformers and the more processing power they threw at them the better they got
82. 12:25and what they realized is they needed to throw so much compute power at it that it thank you for the Roses appreciate
83. 12:32that they needed to throw so much compute power at it that it um they couldn't raise that much money as a
84. 12:39non-profit so they went to what's called a capped for-profit which I had never heard of before I don't know if they
85. 12:45made that up I assume they didn't but a capped for-profit which just basically says no investor can make more I think
86. 12:51it's 100 x their investment which is still a lot um and it's got some limited rights Sam Altman doesn't own any of the stock
87. 13:01which is interesting he's the founder and CEO he doesn't own any of the stock um so so that's all the background
88. 13:11and then about what is it two months ago three months ago Elon Musk signed a letter that said anyone who's doing AI research like open
89. 13:21AI why don't you all take a six month pause because this stuff's dangerous and we don't want it advancing so why don't
90. 13:28you all take a six month pause and a bunch of people signed it and but Elon Musk was kind of at the front of the line there and then
91. 13:36a week later Elon Musk musk accepted a shipment of 10 000 gpus and started x dot AI which is a competitor to open Ai
92. 13:46and to anthropic and to inflection um which which also probably deserves a spot on that list Pi dot a i p i dot a i
93. 13:57um so yeah so that's what that's what x a i is about so at some point Elon Musk is going to come out with now what he's
94. 14:05been saying is these things should be open source they should be free to the world um we'll see what he does
95. 14:12see what he does Biz idea for someone AI adjacent adjacent a geeky chicken wings place that shows star Star Wars no
96. 14:22sports ball that's actually not a bad idea you could you could just do it would just be all sort of geeky kind of cult
97. 14:30films you could you could narrow it to sci-fi but you could also you could also just take you know cult
98. 14:37films that's kind of that's not a bad idea what could you do with AI with that I don't know you could remix your movies in real time
99. 14:45I want to see Star Wars with uh the cast of Indiana Jones there you go have you been on Shark Tank no but I was
100. 14:55on there was a show called American inventor um that was before um before Shark Tank it only lasted one
101. 15:04season but I I was on one season of it um I had a my inventions were a long sled so it's basically like a winter
102. 15:13sled but it had wheels on it so you could go down Hills in summertime and in the South and then um we had this we had this thing called
103. 15:22zip lights which was was like a magnetic base and and like you know those little zip things that the string would pull
104. 15:28out and zip back in with a little light on the end of it a little LED light and like one of the inventor dudes there
105. 15:35said um said so you mean tell me people are gonna buy a light and stick it in their closet and just leave it there and
106. 15:43I'm like yeah and it was like that's ridiculous and like two years later the those Puck lights came out that they
107. 15:51sold billions of so I don't think my idea was stolen I just think that that guy was an idiot pension company called invention Asylum
108. 16:04and so uh my buddy Steve session who um is the CEO of a company called zero shoes x-e-r-o-s zero zero zero shoes um
109. 16:17he uh he was on Shark Tank he and his wife were on Shark Tank he has no qualifications none do not believe anything I say I have a degree
110. 16:27in acting started using chat gbt to review my novel interesting uh what's your name imminent I think it's a good name it's hard to
111. 16:39say it's it's like too many other words uh but I'm curious like you know what's what's your what's been uh the result
112. 16:46like has it been good has it been helping how's it helping you edit I'd love to hear pop it down there Gen X rocks yes all right
113. 16:57what are your qualifications sir I have no qualifications I resemble that remark you people with your desire for knowledge and credentials
114. 17:14[Music] thank you for that what's the best AI related stock in my opinion uh I think it's been established I have no qualifications
115. 17:25um I also have a degree in acting that's been established by Penn State and uh so I wouldn't take any advice from me but I
116. 17:32can tell you the companies that I think are interesting right now um let's see that's one of them that's another one
117. 17:42Microsoft that's another one anthropic po no I don't really care about that's quora quora has been around for a while I I
118. 17:52don't I don't know about Poe it's not I don't think they're doing I don't think they're doing interesting research I
119. 17:58could be wrong I don't know I mean pose at least showing up as a competitor um I think inflection is interesting
120. 18:06that's the the one that was uh the newly relatively newly founded Company by Reed Hoffman of LinkedIn and the dude from
121. 18:16deepmind and they just raised 1.3 billion dollars so that's inflection so of the startups those are the ones I'm paying attention to
122. 18:26um I think Microsoft's interesting I think Google is struggling I think meta is struggling but meta meta seems like
123. 18:35they've got a more interesting approach to this AI stuff at least short term than Google does but Google invented a lot of this [ __ ]
124. 18:45so I just I can't count Google out Nvidia seems like they have a quasi-monopoly um and that that um the the pi dot AI that inflection
125. 19:01company from the deepmind dude they're gonna spend a good chunk of that 1.3 billion dollars on um building a data center with twenty
126. 19:11thousand um gpus like the big honking ten thousand or twenty thousand dollar Nvidia cards so that's going to make nvidia's stock
127. 19:20do something interesting so I don't know I assume all that shit's baked into the price already but because you know and
128. 19:26what Nvidia hit a trillion dollar market cap a month ago but amd's trying to compete with them but it doesn't look good
129. 19:37I don't know Nvidia seems like they have an interesting kind of lock on the uh on the processor business the GPU business
130. 19:46I don't quite know how that world works but holy [ __ ] it's it must be Mighty expensive pretty impressive I don't know what was
131. 19:55impressive the hair oh the all the hair the hair is doing a special thing tonight it's it's parted it's parted in like six
132. 20:05different directions this is like a Vortex there's like a wormhole in my head where aliens are like entering and leaving the universe
133. 20:13all right well what are you gonna do [Laughter] why don't you have a podcast I had a podcast if you go look up uh you
134. 20:25should have been there I did a podcast for a year about the early days of the internet where we interviewed my buddy
135. 20:31ritesh Patel and I interviewed a bunch of the early Founders from the early days of the web Craig canarick and Jeff
136. 20:39Duchess from razorfish and just a bunch of people a bunch of people from agency.com it was cool um I might do a podcast I'm listen I'm
137. 20:47just doing the the tick tock thing because it started as as I was learning about this generative AI stuff I was like my
138. 20:56mind was so overloaded with ideas and crap and learning and I didn't have anyone to talk to about it because like
139. 21:04all of my friends and family were like we don't give a [ __ ] shut up go away and so I started the tick tock just as a
140. 21:12way to kind of process my thoughts and get it out there and then I started doing the lives just as a way to
141. 21:20um not blow a couple hours a night doing Sim racing which I was doing for like two years in a row which is fine but I
142. 21:27just I feel like there's there's an important I think talking about this shit's important I think getting people to try
143. 21:34it and be curious about it is really important so guys it was a joke from a steam stream last night he has no qualifications exactly
144. 21:45what's the best AI platform to build an AI sales agent for texting leads and moving them through a CRM
145. 21:52um probably the easiest thing to do if you want to do that is just do a zapier automation or a series of them you can start really simply
146. 22:02um and you know you could you could probably do something you know when a lead is created in Salesforce that's a
147. 22:08trigger that then generates an email and that email could be based on you know some text from that Salesforce record um
148. 22:20so I would say zapier is probably probably it's it's it's it's got It's an ecosystem that's got lots and lots of
149. 22:28services connected to it I think it's like five or six thousand services that you can connect connect to through
150. 22:34zapier so I would say just start with you know what's your triggering event like like what's going to trigger you know some
151. 22:43communication and then how do you need to process that through open AI or or anthropic or whichever one you use probably open AI
152. 22:53um and then how are you gonna send it are you sending an email is it going into a blast email or is it just updating
153. 23:00um a record in a database lots of ways to do it so if you have more specific questions that's probably where I'd
154. 23:06start and then it's not an AI platform but it's just a it's just a a platform that you can incorporate AI into and tie
155. 23:14it into lots of other systems you're probably already using and then go to futurepedia.io that bottom URL there and
156. 23:23I'm sure there are sales specific AI tools that people have created because obviously this is something that's going to be big
157. 23:31my caveat with going to futurepedia is that there's probably 4 000 Tools in there of which maybe 400 or not abject pieces of [ __ ]
158. 23:42so the signal to noise ratio there's going to be low there's gonna be a lot of crap and the problem with this new AI
159. 23:48world and because everyone's got templates is that all of these apps are going to look really slick and it's not
160. 23:54until you've like committed and start using it that you realize oh this is just a Hollow piece of poop that I
161. 24:03shouldn't have paid for so I'd be careful there there's probably something there to to look at interesting why wouldn't you just
162. 24:13execute and run with the idea because I have too many of them and because see so so I got a couple of things going
163. 24:23on I've got my current company so I'm the the founder and CEO of a company called storyvine it's an 11 year old
164. 24:29automated video production platform um I'm now incorporating AI into the the product vision for for that company and
165. 24:40re-architecting the value proposition and just everything I'm sort of rethinking everything about it so that's a lot
166. 24:48and then I've got this I've got uh the AI Salon that I run which is a bi-weekly Salon like that doesn't take a ton of
167. 24:56time but that's some time um I'm also the chief generative officer from for for a like a Federation of
168. 25:03marketing companies and chief generative officer doesn't mean anything I was just given that title so the guy the guy that
169. 25:11runs it goes hey do you want a title Chief generative officer I'm like sure he goes I think it's the first one in
170. 25:17the world I'm like all right um but but that takes up some time and so so I just have a lot of a lot of stuff going on anyway and here's the
171. 25:27thing I know about any idea is that you shouldn't start it if you you're not willing to see it through the dark times because as exciting as
172. 25:39anything is once you put it in the world you're gonna hit brick wall after brick wall after brick wall and so if it was
173. 25:47just a brain fart that was like oh here's a good idea and it really could have been a good idea if I'm not like
174. 25:52you know tied to it I I it won't survive my ADD for one thing um but it also won't survive just the gauntlet of starting a business because
175. 26:03it's just running starting and running businesses is hard it's hard it's like I I I wouldn't and couldn't do anything
176. 26:11else like at this point I think I am 98.9 unhirable and plus I don't really want to be hired you know you know someone can make me a
177. 26:24you know a ridiculous offer and then I'll probably reconsider that but whatever um yeah so so my heart's got to be in it or
178. 26:32I can't do it I just can't do it can't do it can't do it some comments in the live were filtered to protect the
179. 26:39community's experience don't be [ __ ] to each other thank you all right Sean I built one I can show you oh okay cool
180. 26:46Christopher connect with Sean um I Love Actually this has been happening a lot in the lives and it makes me very happy of people just
181. 26:54um kind of connecting with one another one of the things that is a a Hallmark of my lives is that I try to answer all
182. 27:01the questions so I get really behind and so people don't think I'm actually live but whatever um and you've all been helping one another
183. 27:10as you have questions that's been really cool um do me a favor double tap the screen get some likes going to see if we
184. 27:16can get some more people here share the live if you know how to do that uh and you're willing to like if you think like
185. 27:21who's this damn old man I don't I don't want anyone else to see this stupid [ __ ] my friends see this
186. 27:28I'm not endorsing this you don't need to endorse me that's fine coffee by the way what is the best one what is which best
187. 27:41one Danny what were you asking why don't you have a podcast what's the best podcast oh for I I don't know for for this AI stuff
188. 27:55I don't know the only one I listen to really consistently is the uh Paul rateser's marketing AI Institute I listened to that Weekly
189. 28:05he's got his [ __ ] together he's been doing this for like five or six years on this generative AI stuff around
190. 28:11marketing so I think it's uh that one's that one's a good one mine's like 120k I don't know what that means told my
191. 28:22idea to someone 20 years ago and heard it in the speech recently with all the same examples yeah that happens I'm
192. 28:28telling you man you put something in the world it's just in the world I don't know what it is there's something about
193. 28:33it it just happens especially when there's like a lot of energy there's so much energy around this stuff the other
194. 28:39thing that's going to happen with these tools is it's going to be incredibly easy for people to steal other people's
195. 28:50businesses like incredibly easy not just the look and feel not just the marketing like all the typical [ __ ] you see today
196. 28:58the barrier to entry historically was doing the execution part like doing the coding of a particular kind of platform
197. 29:07or particular kind of service that was always the barrier to entry but with these tools you're going to literally be
198. 29:12able to say grab me all the marketing positioning rewrite it so it sounds different grab me the design you know
199. 29:18redesign it so it looks different but it's still functionally the same and then let's go ahead and just copy that functionality
200. 29:26um I don't know what that does to the world of investing and startups and entrepreneurs I don't it seems like it's going to be a really
201. 29:39stressful time like like just like just like it's going to be stressful for people who get displaced out of current you know nine
202. 29:48to five jobs I think there's going to be similar displacement in the startup world and in the certainly in the investing world
203. 29:56um so it's gonna be I that's that's an interesting time how do I access code interpreter is it a plug-in or separate
204. 30:02service here I'll show you um okay so here we are at chat gbt you can make money
205. 30:16um so in order to get access to code interpreter you need to have a chat GPT Plus account why is this not there we go
206. 30:29um you have to have a chat GPT Plus account and and then you'll get access to it now if you have a GPT Plus account and don't
207. 30:38see it that means you haven't enabled it so you have to go to your settings and then in your settings there's this
208. 30:46thing called beta features and turn them on so you got plugins and you got code interpreter and at some point they'll
209. 30:53bring bring back browse with Bing I don't know what's taken them so long but whatever um just turn them on and then you will
210. 31:03see code interpreter and if you haven't seen code interpreter let me show you what it do it's amazing so normally in
211. 31:10chat gbt you just type your thing here like write me a poem and it'll do that what you can do with code interpreters
212. 31:16you hit plus upload a file and then I forget what this was a data set I grabbed last night I don't even remember what it was
213. 31:28but it's called archive five zip and it's uploading I know slowly [Music] um
214. 31:42and I can now go what is in this file question mark so it says the uploaded file appears to be a zip archive to see what's inside
215. 31:56I'll have to unpack it let's do that and then it fires off this plug-in looking thing and if you look at what it did
216. 32:04it actually wrote python code it it loaded some libraries wrote python code to unzip that file unzipped it and then
217. 32:12it says oh this is crime data so it says it it contains the following crime offense and crime XML would you like to
218. 32:21preview the content of these files to or or perform a specific let's say preview um preview what's here
219. 32:31and so again it's writing python code to read those files and give us a little example of what's there you don't have to look at this
220. 32:40code this is all just happening in the background so there's there's what the files returned so it seems like there was an issue
221. 32:48reading one of the files because it was encoded weirdly so now it's going to try another way of encoding it or reading the encoding uh
222. 33:08looks like it hung so it's probably going to apologize and try something else and that's another thing that's wild about this is
223. 33:16if it if it fails it it um it'll it'll try again and it apologizes
224. 33:31um I'm gonna say stop generating let me say um please visualize VIs you are lies the interesting
225. 33:46um Trends in this data something went wrong regenerate response something went wrong okay I gotta reload it you failed I'm sorry you failed
226. 34:04um if you're sitting there thinking well I thought code interpreter was supposed to be so cool why did it just poop out on
227. 34:11you because all of this technology right now is early it's just a sign that we're early if it were perfect and easy to use then we
228. 34:21wouldn't be at this phase of the uh the the beginning of of a remarkable transformation in all of humanity um so let's see um please
229. 34:35visualize the interesting Trends in the data here okay
230. 34:49let's see what it does first let's unzip it foreign [Music]
231. 35:13seems like there was an encoding problem so now it's going to try a different encoding thank you for the hearts
232. 35:25all right so now there's the data incidents Offense Code offense type offense category first to current state last to current state
233. 35:33incident address Geographic coordinates of the incident neighborhood so there's a bunch of interesting stuff there so
234. 35:43it's unzip the file figured out the encoding of the CSV files read the data now it's reading the data of another CSV file
235. 35:58so it says before analyzing the data let's take a quick look at the XML just so we understand the hierarchical structure
236. 36:06so it's got there's there's your metadata from the XML so now let's move forward with the analysis of the CSV file based on the
237. 36:16contents there are several several interesting Trends we might explore number of incidents over time
238. 36:21distribution event uh offense by type or category distribution of incidents by geographic area number of victims per offense so let's do um
239. 36:36a beautiful um visualization around three so it knows what I meant by three it
240. 36:50gave me four options and I just picked the third one so it says sure let's do it by geographical area we can start
241. 36:56with distribution of incidents by District let's create a bar chart incidents for each district
242. 37:11so that's boring can you make it a heat map um on a geographic map
243. 37:28let's see if we can figure that out unfortunately I'm not able to generate a geographic map that's not always true oh
244. 37:37because lack of the internet access but there it tells me the code I could do if I wanted to do it myself um
245. 37:49let's see can you do some other more visually interesting um look I don't know
246. 38:05sure I understand you're looking for something more visually interesting how about we explore temporal Trends in
247. 38:11the data we can visualize the number of crimes I'm just going to say yes see what it comes up with
248. 38:30to do this we'll need to convert the first occurrence date to a date time format then extract the hour for each time stamp
249. 38:39so now it's doing some full-on you know data work some data cleansing and and reformatting
250. 38:55we're waiting we're waiting if it times out it'll it'll often give itself a new task too [Music]
251. 39:10I have no idea how much data's in here I think it's a fair amount of data now what this thing's actually doing is
252. 39:17when you start a chat instance with code interpreter it's actually creating a virtual machine so this is like our own
253. 39:24little stand-up computer or like you know virtual computer that is you know setting up all of its RAM and its places
254. 39:31to write data and it's installing Python and installing the libraries and then running all that stuff within that virtual machine
255. 39:41so that's number of incidents by hour of the day I mean that's that's fine it's it's not all that interesting but but
256. 39:48anyway we could go on but that's that's what code interpreter does that's how you find it so hopefully that gives you what you need [Music]
257. 39:59d d um I'm parsing it out in chat GPT chapter by chapter and it's it's keeping up so far um the only thing to be careful with if
258. 40:10you're doing it chapter by chapter is just be cognizant of the context window it's only 3 000 words so
259. 40:19um it may forget your earlier chapters so what you might want to do if you haven't already done this is for each
260. 40:25chapter have it summarize the chapter so that um you've got a shorter representation of what's happening in the chapters leading up to
261. 40:35the current one you're on that'll keep as much as possible in the in the context window the other thing you could
262. 40:41try is claude.ai has a hundred thousand token context window so you could pop all of your writing in that and then
263. 40:50just start interacting with it that's another option um I don't think it's as good as gpt4 but it's pretty flipping good
264. 41:02I feel the same way I can't keep up I know same here can't keep up at all it's just there's too much there's too much Buck you know
265. 41:10you know it's like it's like I told you there's just a lot there's a lot there's a lot to deal with it keeps coming at
266. 41:16you it keeps coming at you you see it's a lot yeah that's champ quick question can turn it in detect AI
267. 41:34is turn it in homework like plagiarism software probably um if you're University or your high school is using AI detecting software
268. 41:48I would I would transfer um we should be past the point at this point where universities are saying you know
269. 42:01we're just going to use AI detector software they should be engaging with a Ai and teaching their students how to deal with what's coming
270. 42:09um they should be teaching them critical thinking skills and understanding how this [ __ ] works and what the
271. 42:14implications are but that's a whole separate conversation love your haircut thanks I don't know that this is a haircut so much as a
272. 42:23like three toupees got in a fight and this one lost that's just really bad but thank you I'm just here for the shirts oh yeah uh-huh
273. 42:40this shirt is you can almost read the text in front of it it's a little busy there's a lot going on here it's it's a little hard to
274. 42:47read it hurts the eyes but thank you any AI crypto making tokenism type apps you can I mean chatgpt does Smart
275. 42:57contracts you could just copy your favorite smart contract paste it in there and say write me a new smart
276. 43:02contract and give it whatever you want it to be and it'll do it um probably go to futurepedia.io there's
277. 43:13probably some crypto Bros have figured out a way to put a front end on front of it in front of it but what I would do I mean
278. 43:19seriously what I would do is it's good enough with with coding and with with a lot of the the smart contract stuff that
279. 43:27I would probably just figure out how to do it in chat GPT just go learn and maybe you start a business right maybe
280. 43:33that becomes a business you know click here to build an nft collection here's your images here's your this
281. 43:40here's your smart contract here's you this do you want to deploy it start a Discord server and get ready to rock your followers
282. 43:49it's a web 3 tradition you can now rug people without even knowing code [Laughter] oh web3 time was huh
283. 44:04showed such a promise web3's got three years of wound looking well-deserved wound looking but I think AI is going to drive adoption of web3
284. 44:15in the end because I think it's actually really important technology it just it just had the wrong people at the helm for a while there
285. 44:26all right I'm trying to figure out where I was Champion stop it hey what'd you do tonight I don't know I
286. 44:37watched some old man yell at his dog for an hour why would you do that I had no idea oh that's way old questions hold please
287. 44:50I lost my place I lost my place it's hard you don't know how hard it is to yesterday laughing my ass off because why
288. 45:07you don't think they have a monopoly on this AI [ __ ] tell me someone who can compete with them right now all of
289. 45:14the software infrastructures based on their um based on making calls to their chips no one else seems to be doing it maybe Intel gpus but
290. 45:28maybe AMD go listen to that if you want to hear if you want to hear a really interesting take on that stuff go listen to The Lex
291. 45:38Friedman George hotz interview from three days ago and Listen to George Hots talk about you know he almost bailed on AMD and then they're
292. 45:48trying to get their [ __ ] together and then he almost went to Intel and didn't want to do that but he doesn't want
293. 45:53Nvidia at the top of the heat but he's like who's gonna compete check EBT novel analysis was pretty impressive it picked
294. 46:01up on what I intended to do thematically that's cool from a fiction narrative perspective but it's hard to trust
295. 46:09at first it responded like I'm a great writer when I that's interesting when I asked if there were problems it found
296. 46:16problems and contradictions to previous analysis and after several chapters it crashed several times but the basic analysis was at least on
297. 46:26point with what I intended that's pretty cool hey Troy Decker it is Kyle that's me I'm Kyle um yeah the The Narrative analysis thing
298. 46:40having having chat GPT be like your your writing coach and your writing partner is a really interesting idea that would be that would
299. 46:49business to build is you could you could almost build a tool that works in stages right there's the ideation stage there's the blocking and
300. 47:04tackling rough draft stage and then there's different levels of refinement and if you could build tools that make
301. 47:12it really easy for writers to navigate those stages of of writing a novel where they're still in charge but you're kind
302. 47:19of supporting them and doing it in the sort of pre-structured way with a really slick interface that's a business
303. 47:27now what's the downside of that business writers are broke who's gonna [ __ ] pay for it so it's not a business it's a good idea
304. 47:39one of the things you should always be asking yourself because there might be a brilliant idea and people might tell you
305. 47:45holy crap yeah that's awesome I'd use that don't mistake that for here's a check that's what you need to figure out is
306. 47:56build a something that's a good enough solution someone's willing to write a check for it I'm sorry that's very Gen X that
307. 48:04someone's willing to venmo you money for it or give you six diamonds [Laughter] hey hey Kyle you're like a fine wine
308. 48:11better with age Gen X yeah thanks Troy in the either have you heard of road map to riches I have not I will check it out though my
309. 48:24novel is about 120 000 words oh interesting so yeah that's a that's a big fatty of a novel does it need to be 120 000 words if it
310. 48:35does it does but that could be an interesting exercise is to see if see see how chat GPT might might um
311. 48:44how it would suggest tightening that if you want to tighten it but if it's like a big epic story then maybe it needs to be that big
312. 48:52I don't [ __ ] know I don't read so 80d 80d in reading for me they do not go well together I want chat GPT to rewrite
313. 49:01and design my website how would I go about it um did you I don't know there's a lot of ways to do it uh
314. 49:13I don't know enough about that world I mean I would go to futurepedia.io and look at some of the tools that people
315. 49:19have already created for AI assisted web you know design and redesign and design analysis I'm sure there's some tools there that do it um
316. 49:31you can also just try taking chunks of your code pasting it into chatgpt and say you know tell me about this website
317. 49:40and then describe how you want it changed and just sort of go back and forth copying and pasting code like
318. 49:46describe how you want things changed go back and see how it works on the site you might do that if you're doing that
319. 49:52or just use one of the no code tools you know Squarespace or or or webflow or whatever he's so polite to the robot I am I do
320. 50:01say please and thank you because quite frankly the robot's polite to me and it and it kind of just triggers
321. 50:07it's less like I'm prompting a robot and more like I'm having a conversation with a remote co-worker
322. 50:14and so I do I find myself I I think I've gotten over this part of it but this has taken me five or six months is
323. 50:23like if it would be getting things wrong or I just have a new idea and I'm like oh I want you to rewrite this oh rewrite
324. 50:29it again rewrite it again about the third time I'd say rewrite it again I would start to feel guilty like maybe
325. 50:36I'm asking it to do do too much and they're like oh no it's a robot you [ __ ] idiot um but it's but it's like a there's
326. 50:45something about because it feels like you're interacting with a person you're having like this back and forth collaborative thing
327. 50:52that experience I have with a real human like about the third time I send something back for a rewrite
328. 50:59I just go oh [ __ ] it I'll just do it myself it'll be faster for me to just do it myself and and that's you know just
329. 51:06me being a spineless ninny and not wanting to offend someone but whatever um he's Apocrypha I don't know what that
330. 51:16Rachel good word good Lord this is this is the AI learning lab apparently this is the AI this is the vocabulary learning lab as well
331. 51:27apocrypha that seems like what I should be able to figure out from the Greek a collection of writings that are not
332. 51:39included in the standard Canon of religious tradition wow fantastic that is I I it is worth noting the term profit can
333. 51:52also be used more broadly to refer to any non-canonical or disputed writings okay I I I don't I I like it I like it I like it okay
334. 52:03is your GPT on a local machine no I haven't done that oh he is Apocrypha Apocrypha is a person in here
335. 52:12so Rachel's referring to Apocrypha not saying I am Apocrypha but you're saying I am whatever she was asking her he was
336. 52:19asking before but now we know what Apocrypha is that's cool am I saying that right it's got to be apocrypha okay agree all
337. 52:29right oh crap Tick Tock just jumped me to the bottom dang it tick tock love your lives bro thank you meta Victory I appreciate that um
338. 52:42I feel like like I'm saying nothing today I feel like I'm just rambling tonight so if I am just tell me I'm wasting your time
339. 52:56[Music] novel analysis was pretty impressive Oh wait we're already here oh Tick Tock man it's just jumping me all over the
340. 53:07place what's happening look at tiny bucks do you think Universal income is inevitable okay is your GPT on a local
341. 53:15machine no my GPT is not a local machine um I will I'll probably do a local llm install when they get better and smaller and uh
342. 53:30more stable they're just they're just evolving too fast right now um although I met with our Tech team today and we've
343. 53:39got an instance of Lang chain up and running so that we can start running um Lang chain prompts on all of the transcriptions in our videos
344. 53:48um so I sort of have a local machine running but not on like my local personal machine so I do for work do you think Universal
345. 53:56income is inevitable um I think something is going to be necessary I do I think I think maybe a year from now
346. 54:09yeah probably about a year from now the disruption is is the disruption from AI on the workforce is going to be significant enough that um
347. 54:22I don't think I don't know my sense is it won't be as dramatic as kovid you know covid we just sort of like we were just like okay
348. 54:31tomorrow don't leave the house right it was just like bang and it was Global it was [ __ ] surreal I don't think it's gonna be like that
349. 54:40but I do think that there's going to be a level of disruption that's bad enough that it's going to have to be dealt with
350. 54:48and I don't think companies are going to deal with it because they're just money machines and they'll be like well we're
351. 54:55generating the same amount of product with a third of the people awesome profit Bob you going golfing this weekend no I'll
352. 55:04be on the yacht sorry um so I think they'll be doing that and then I think the workers are gonna the X workers are gonna be
353. 55:12um you know increasingly pissed off and freaked out and so I think something's gonna have to happen hopefully
354. 55:20it doesn't get to the point where people are freaking out and they have to do something hopefully they do something
355. 55:25proactively but we'll see like like it could be you know it could be Universal income or it could be incentives to
356. 55:34companies to not lay off people like if they're seeing you know increased productivity with AI to maintain their
357. 55:43current employer levels there were a lot of those incentives with kovid you know don't lay off your people so I could see
358. 55:49something like that but I could also see there being some sort of universal income I don't think it's going to be
359. 55:55necessary to maintain it for maybe I don't know more than two or three years but I don't know I don't know look at
360. 56:04tiny bucks Tiny Box is really interesting that's Tiny Box is George hotza's new company he's creating or
361. 56:10it's called tiny Corp and then the the computer that he's building is called Tiny Box and it's a fifteen thousand
362. 56:17dollar six GPU AI monster machine that basically what he wants to do is he wants to make the most powerful AI
363. 56:26super computer that you can just plug into a regular wall outlet and not have to run special power to it and not have
364. 56:32to run special cooling to it and not have it be so noisy that you can't work next to it but it's going to be like you
365. 56:38know some ridiculous amount of flops that you can do you can run a 65 billion parameter large language model on it
366. 56:49he's pretty brilliant brilliant dude that Duty God these guys looking Marvel's going on in the old uh Gumball Factory if you know
367. 56:57what I mean all right what I liked was when it got critical about Pros it wrote a disclaimer stating that it was a
368. 57:07subjective analysis that's interesting silver fox do you know how long code interpreter keeps your project results to refer back
369. 57:14to not long like this is one of the okay so here's a real frustrating thing about code interpreter right now I don't have
370. 57:21my little URLs up do I um code interpreter like if you go away from it for like 30 minutes and sometimes it's faster like I can't tell
371. 57:29when it when it shuts it down so so you upload some some stuff you run some data analysis on it it makes you
372. 57:37all the charts and graphs if you go back to it like half an hour later it says you know this this is gone
373. 57:43and then it it just everything's gone and you can redo it but it's it's gonna write it's gonna you're gonna have to start from scratch
374. 57:50so a thing you're gonna have to do with code interpreter short term because they'll fix this is um once you get results that you like
375. 57:58migrate that stuff into other documents um that's it's a really good question it's a problem so so you just need to
376. 58:05just need to keep on top of it just know that whatever's in your active session anything you want to keep get it out of there
377. 58:13um I'm in the road map to Rich's Marketing Group you would be an asset oh cool um let me go so do me a favor just Rachel just send me a DM
378. 58:22and I will uh I'll check it out for sure he's Apocrypha that's really funny that
379. 58:36I thought you were calling me Apocrypha I'm like I'm a lost book of the Bible can Chachi bt4 interact with Google Maps
380. 58:51or other map apps let's go look at plugins there might be I think there's some map plugins so let's go I'll show you some [Music] um
381. 59:05uh rut row I wonder if plugins are turned off oh your session is expired no [Music]
382. 59:21name that comedian come on you gen xers no plugins enabled um plug-ins down I think plugins is down
383. 59:37reload plugins okay they're back okay never mind false alarm don't go posting that on Twitter they're back
384. 59:51okay never mind false alarm don't go posting that on Twitter AI learning lab said that openai canceled plug-ins all right Maps
385. 1:00:08Okay maps map app takes latitude longitude and latitude coordinates and generates a map Tom Tom explores Maps gets traffic updates
386. 1:00:19how many of these oh there's only eight of them Whimsical diagrams mind maps mixerbox powered by Google Maps API
387. 1:00:30mixer box chat map is the world's first AI chat box for Matt chat bot for maps powered by Google Maps API mixer box
388. 1:00:42chat map is the world's first AI chat bot for maps okay for the marketers in the bit in in the audience
389. 1:00:53if you ever write a sentence like that and and actually put that out to the public you should be ashamed of yourself
390. 1:01:01you you you cannot physically say this sentence much less understand it powered by Google Maps API mixer box chat map is
391. 1:01:10the world's first AI chat bot for maps if you write something like that have chat GPT rewrite it but let's try this
392. 1:01:20and see if it works okay so I'm going to install mixerbox mixer box chat map if you can't say it too if you can't say
393. 1:01:29it say it one time fast it's too complicated I'm turning off instacart so we only have mixer bot chat Maps I can't say
394. 1:01:42that's [ __ ] hilarious okay um let's see um like um I'm trying to think I don't I'm not connected to the internet so let's just
395. 1:01:57say um show me um the water around Coney Island
396. 1:02:14so this should fire off the plug-in it did here's some water related places around Coney Island New York Aquarium no no no right
397. 1:02:28uh no no I want to see um the Hudson River
398. 1:02:44on the map near Coney Island see if we can get it to actually generate some map things but what what this is telling me is um
399. 1:02:59so there's Coney Island Coney Island Creek in Brooklyn oh that's that's an interesting one let's go look at that
400. 1:03:14all right so it's taking us here's Coney Island Beach and there's the creek I've fished out here this is good
401. 1:03:22I've walked out here and fished that's fun um I don't know what this does Beyond that you couldn't do on Google Maps
402. 1:03:37right so um like what is what it says to me is that the the fact that there's only five plugins that have maps and one of them isn't
403. 1:03:48even map related is that this is probably the this is probably something that's coming well here's let's go let's go try another thing um
404. 1:03:59let's go try Bing because Bing's connected to the internet and Bing is also a search engine with a map
405. 1:04:08so let's go to chat and then let's say um um show me the let's see um overlay today's high and low
406. 1:04:26temps um on a map of Coney Island
407. 1:04:41so so I went to bing.com went to chat and then just typed that in I'm sorry I'm not able to Overlay this over a map house can I help you
408. 1:04:53yeah so it looks like they're not really dealing with maps right now so there's a business there's something to play with
409. 1:05:03I'm not sure what you want to do with it you know it might be that you know there's a there's a map API for Google so
410. 1:05:09I would think it would be relatively trivial to build some automations or some applications that do that
411. 1:05:20um you are invaluable thank you by the way your idea too and then it cuts off thank you Sean Kyle hi quick thoughts on making a
412. 1:05:29choice on AI tools and how many to concentrate on for beginners awesome question and hello AI for beginners um why I put these urls
413. 1:05:43this is my answer um and the only thing that I would add to this are the [Music] um are the image generation tools but I
414. 1:05:55think I think as a foundation where where to start is with chat gbt just the official website if you can't afford
415. 1:06:03chat GPT plus then use Bing or Claude now so it used to be Bing was the free version of gpt4 but now Claude
416. 1:06:14which is a bunch of ex-openai people now that they've launched this this is Claude 2 with a hundred thousand token
417. 1:06:21context window which for beginners it just basically means you can put a shitload of words in it and it doesn't
418. 1:06:27break gpt4 still breaks you know after 3000 words or so it might be up to six thousand words now but it's still not 75
419. 1:06:37000 words like Claude is um and then poe.com lets you play with a bunch of different models you know three
420. 1:06:47from anthropic two from open Ai and one from themselves plus they've got bots so as a beginner I would focus on learning
421. 1:06:58prompting and so prompts.chat that document right there it's just a it's just a a document called awesome chat
422. 1:07:05GPT prompts and it it's really just a a single flavor of font of prompt which is act as a you know Persona here
423. 1:07:16um so act as an accountant act as a screenwriter act as a therapist act as a coach act as a nutritionist but you can
424. 1:07:25really quickly get your head around what prompting is how to do it and also how incredibly [ __ ] profoundly deep these tools are
425. 1:07:37this is not these are not fancy Google searches so anyone beginning I would say you know basically this is what I've
426. 1:07:46been saying in here you owe it to yourself to play with these tools long enough for you to have your Kevin
427. 1:07:51McAllister moment which is this moment oh Malone moment where at some point you'll do something with one of these and it'll just blow
428. 1:08:02your [ __ ] mind and and that's the point at which you should start getting that then you should start digging in
429. 1:08:08and then you can start to focus about what tools you want to use and what where you want to explore on the image side of the house
430. 1:08:16you can do free images at bing.com you just tell it to make an image and it will that's using Dali I personally pay
431. 1:08:23for mid-journey because I think it's the best stable diffusion if you're geeky is has the the by far the greatest flexibility
432. 1:08:31of the tools out there Adobe Firefly and generative fill within Photoshop are commercially safe because they train
433. 1:08:39their models on open sort or uh public domain and Adobe stock libraries that they have the rights to so if you don't
434. 1:08:49want to risk being in any sort of legal Waters I would steer away from mid-journey and stable diffusion
435. 1:08:58and probably even Dali and I would go toward the Adobe products so there's that's where I'd start and then if you
436. 1:09:05really just want to geek out futurepedia.io that's just a big old pile of tools so you can feel free to go
437. 1:09:12if you well what about I want to compose music go there and find all the music tools um I accidentally hit send skip my comment okay
438. 1:09:24Kyle you big fat oaf get a real haircut and what's with the glasses yeah that's okay that's fine that's fine whatever
439. 1:09:33you said I'm not I won't take it personally I spent too much time on AI lab my wife is getting jealous you think
440. 1:09:39your your wife is getting jealous my wife's like hey what are you who are you talking to why do you are you gonna go
441. 1:09:44do that again so yeah yeah I get it I understand uh I felt bad for a split second when I kept having to add the letter
442. 1:09:54to add the letter I was writing I felt bad for a split sec I don't know what that's in reference sorry all right we're building an AI
443. 1:10:04call center agents I'm just a dumb entrepreneur no you're not in my opinion this will displace 80 of call center reps I saw I saw a really
444. 1:10:14unfortunate headline today um it said Indian call center replaces 90 like I don't know whatever 94 96 like fires 94 96 of its staff
445. 1:10:31and I was sort of imagining like a hundred thousand person call center but it was it was a 22 person company
446. 1:10:37um but yeah I agree I think that um I think call centers are going to be replaced and I think from a an end user
447. 1:10:45experience standpoint they should go because here's the deal I want I want to be able to call a company
448. 1:10:55and rather than sitting on hold for 45 minutes talk to a synthesized bot that knows everything about the company
449. 1:11:04can answer all of my questions has the authority to do some level of [ __ ] and is smart enough to know that if my
450. 1:11:12problem is non-standard and needs to be dealt with by a human it immediately connects me to a human that means you could get rid of 80 to 90
451. 1:11:24percent of call setters take the 10 or 20 percent of the people that are left upskill them into high level customer
452. 1:11:32service people that have the authority to [ __ ] fix problems that that are escalated let the llm-powered chat bot talk to everyone
453. 1:11:45answer the phone immediately detect from their area code what part of the world they're in and have the chat
454. 1:11:53bot you know respond to them in an appropriate dialect so it sounds like them have it know all the [ __ ]
455. 1:12:00have the conversation have it record the conversation so that if it does escalate it to a human a nice little report is
456. 1:12:07given to that human here's what this person's dealing with here's why I escalated it to you you don't have to be that's that's the
457. 1:12:16only transfer you do it's a two transfer call center and there's no phone trees that's what I want go build that they're
458. 1:12:23they're second business of the night second business business of the night that needs to be built we're going to transform customer
459. 1:12:33service to actually not suck again thank you I got I got one hard on that I don't know I seemed really upset about call centers I don't know
460. 1:12:47I don't think he likes India say it ain't India that's not the problem the problem is Scripts and phone trees
461. 1:12:59and bad shitty over complicated logic and people that are paid below minimum wage that are not allowed to actually have a
462. 1:13:10conversation they can only read the script that's the problem you can flush that entire [ __ ] system down the toilet and it and it will
463. 1:13:22improve dramatically Andy Kaufman hey Ellie come to save that day there's a jet some gen x's in the house oh yeah Wolfman
464. 1:13:37all right yeah Sean McCracken I think that's really interesting I I would be curious to you know keep me posted on how that
465. 1:13:51evolves because I think that's uh that's a big deal um do me a favor if you would like uh tap the tap the screen make the hearts
466. 1:13:58go and if you would share the live so we can see if we can get some more folks in here Silver Fox for code interpreter I've
467. 1:14:05been saving the code when I have my results oh that's interesting Underdog it's Mighty Mouse is copy and paste the only way to save
468. 1:14:16migrate a chat GPT history no um if you go to go to your settings and then go to what's it called Data
469. 1:14:33data controls and there's an export data button pop that button now to Silver Fox it was a silver fox yeah the silver Fox's Point
470. 1:14:48um I don't know that the code the code interpreter writes will get exported with that if your virtual machine has been shut down so
471. 1:14:59I would say for code interpreter if you want the code that it wrote to do to generate whatever charts and graphs you like um
472. 1:15:09that you might want to copy and paste but that's worth an experiment I had ice cream tonight because because because I'm a bad bad keto
473. 1:15:24person I'm doing okay on the intermittent fasting but I'm doing bad on the Keta on the non-uh carb part of it
474. 1:15:36I've lost a little bit of weight this shirt didn't fit at all um I don't know three months ago so I'm attending a meeting about llms at my
475. 1:15:48Corp today wow that's interesting I'm in their chat group and learning group lab that's cool man is it is it a big company that you work
476. 1:15:58at cabrino and by the way do you say your name cavruno I never know how to say your name it's it's like Bruno KV
477. 1:16:06Runo k k Bruno Bruno I don't know anyway you could change it so that it's like sayable oh you'd be like llm Mama or
478. 1:16:21[Laughter] what did I do last night I misspelled AI learning lab when I made a bot so I made a bot on Poe called AI leaning dude
479. 1:16:34which is which is not bad is gmaps plug-in wait is gmaps plug-in a publisher AKA marketer in my opinion for now
480. 1:16:46a lot of the whoa what's that Sean that was cool I've never seen that one I got the cowboy hat and a little heart every
481. 1:16:54now and again but the uh oh that's cool uh is gmaps plug it wait gmaps plugin is a publisher AK marketer in my opinion for
482. 1:17:06now a lot of the plugins right now a lot of the plugins are um just shitty ass ads for websites come on focus focus you can do it you
483. 1:17:21can do it come on there we go all right here that generative AI could add 2.6 to 4.4 trillion to the economy
484. 1:17:31yeah I think the I think the um the the long-term prosperity and and that's low like I that that was a reporter it was an article that that
485. 1:17:42came out like what 10 days ago or something like that Bruno damn dang your name um and and what it was talking about it was
486. 1:17:51kind of comparing the early predictions of what the World Wide Web would add to the economy and what it actually added
487. 1:17:57which was you know significantly higher and and then you know they're saying yeah 2.6 to 4.4 trillion um I think it's got the chance of of
488. 1:18:07contributing you know see even more than that like a significant amount like we could exceed that um again I you know I'm not an economist
489. 1:18:16but I I just in in my simpleton mind when I when I hear anecdotal evidence of when uh when a tech firm goes all in on
490. 1:18:28code on uh GitHub co-pilot when they go all in on AI pair programming they're seeing a 30 to 40 percent increase in productivity
491. 1:18:37so so if you take that as an average take the low end of that let's say there's a 25 increase in productivity
492. 1:18:44thanks to these tools and let's say these tools you know get spread across all the stuff that we do so you've got this amount of people
493. 1:18:51working producing this amount of economy and with a 25 increase like that's a lot that's a lot of increase when you know
494. 1:19:01when we're measuring things in you know half a percent one percent two percent if you see a 25 increase that's a big
495. 1:19:08gap that gets that gets created there and then you know what do you do with that Gap is interesting do you maintain
496. 1:19:13the economy and lower the workforce right that's where you need Universal basic income or do you maintain the
497. 1:19:20workforce and dramatically improve the prosperity and if you can do that in a way that the billionaires don't just buy a
498. 1:19:28second and third yacht but do that in a way where that money can go to deal with the displaced and upskill the displaced
499. 1:19:35then then we potentially start to to move into an economy that that is is quite prosperous and that's that's a very real possibility
500. 1:19:45um I want route optimization input several addresses and it gives me gives you the most efficient driving I think
501. 1:19:52there's some that do that I think um is it Bing maps that does that I know Waze does that you can do it within Google Maps but
502. 1:20:04it's really uh convoluted it's the interface for it's really shitty um what would be worth trying is is go to
503. 1:20:14chatgpt and have it tell you which mapping tools allow you to do that because I I know there's I I oh what it was uh
504. 1:20:25MapQuest MapQuest had a pretty good version of that I don't know if MapQuest is still around they were big for a while
505. 1:20:32trading in llm is terrible CO2 emissions equivalent to 60 or 70 cars yeah I I I agree with that although the if you're
506. 1:20:41looking at what's going on with how they're training the smaller large language models in the open source
507. 1:20:46Community where it's costing them like 100 bucks in training to get it nearly to GPT 3.5 performance um I think I think the optimization of
508. 1:20:56these large language models for the for the non-big ones right I think I think open AI inflection um Google meta
509. 1:21:08you're gonna have big players that are going to spend Millions on you know GPU centers to train the massive data sets
510. 1:21:17but I think the efficiencies and the the open source things are going to get increasingly powerful in increasingly
511. 1:21:23smaller packages so I think there's going to be a balance and equilibrium that's hit there where I I don't think
512. 1:21:29it's going to be runaway energy costs I hope but I don't know did you see the uh the uh the Google Sycamore um
513. 1:21:40um quantum computer now now I don't think this is related to AI because it's a different kind of compute but maybe it
514. 1:21:46is I don't I don't [ __ ] know um but the the Sycamore so so Google has a 70 Cubit quantum computer and they
515. 1:21:56just did a benchmark that it it it calculated in six seconds what it would take today's state-of-the-art super computer 47 years to compute
516. 1:22:08six seconds versus 47 years it's like [ __ ] so um I have no idea what's going to happen with that Quantum stuff I mean other
517. 1:22:19than you know all of your high security passwords are now compromised that shit's gone anyone can get into any system at any
518. 1:22:30time uh when you have that level of compute going on luckily those things are still expensive um oh it just jumped me around again
519. 1:22:42dang it dang dang it uh Portland is homeless due to call center decline in automation that's interesting
520. 1:22:51listen convert to text query response text to speech play response audio already running oh that's interesting Sean yeah I'd be interested
521. 1:23:01to learn more about that because I I mean I really do think like that's that's where it's got to go like like
522. 1:23:07just just if you think in the end the inefficiency of the like the the incredible amount of time that must go into programming phone
523. 1:23:18trees like I've got a small company and programming our little voicemail thing was an abject nightmare I can't even
524. 1:23:23imagine doing it for a bigger company so so just just the labor alone to program call center tree logic is
525. 1:23:34like that's a whole industry that that you know needs to go away it's it's a horrible experience for the end User it's really expensive it's
526. 1:23:47really time consuming um it's not necessarily better for the company now is that bad for all of the companies that have spent decades
527. 1:23:56building phone tree logic programming companies yeah it's [ __ ] devastating but which of those can pivot into this
528. 1:24:05new world because maybe they become experts at creating the kind of system that you're talking about Sean oh Chris Bruno
529. 1:24:17largest private company in the world interesting pretty big 155 000 employees wow that's really uh I'm curious what sector
530. 1:24:25um okay I'll call you Chris from now on because crevuno is making my tongue hurt but nice to meet you Chris
531. 1:24:36um it's really interesting that that your company is leaning into this I'm curious what sector it's in hey Kyle hello Joy
532. 1:24:45have you ever asked chat gbt to write some good jokes I have and it sucks let's go try some now we're gonna try it
533. 1:24:52on a couple of different tools yeah let's do this let's do this okay so we will do this well this actually this could be fun um
534. 1:25:05so we're gonna go to so where I am is I went to Poe so it po we we've got Sage we've got Claude 2 100K Claude instant 100K chat gbt4
535. 1:25:17Claude instant and Chachi BT so chat EBT is 3.5 Claude instant is anthropic's equivalent of that so those
536. 1:25:26are roughly equivalent Sage I think is like those two Claude 2 is like gpt4 yeah all right so here's what we're gonna do we're gonna go
537. 1:25:40we gotta pick a comedian oh we'll do Jim Gaffigan okay I want you uh like you are a stand-up comedian uh wait stand up comedy writer
538. 1:25:56who writes um for let's see um big name comedians like Jim Gaff again and
539. 1:26:12um John Mulaney they write their own stuff I know but that's all right um your expert at writing short
540. 1:26:29jokes about the funny things in life most people Overlook okay so we've set up we've set up who this
541. 1:26:44person is now we say write me a joke okay here we go about how fast this is a good one how fast avocados
542. 1:27:00go from unripe to rotten which is that's a universal frustration in all of the world okay um
543. 1:27:15sure here's your joke for you avocados are like gold the Goldilocks of fruit you buy them unripe and they're too hard then you wait a day
544. 1:27:24and they're just right but if you wait one more day they turn into mushy green sadness it's like you're auditioning for a spot
545. 1:27:31on The Bachelor they go from not ready to over emotional in just 24 hours there there's the essence of a joke there but that is
546. 1:27:41hugely not funny okay so let's go let's go take our little uh let's take our little prompt here and let's pop down to we'll do Claude
547. 1:27:55instance so this is anthropics version of uh like GPT 3.5 about two days to use it before it turns
548. 1:28:11into a rotten stack of Goo Ah I don't get it that's it okay that's a good setup of a joke but the punch line
549. 1:28:21is disturbing is it like did you not feed the puppy and now it's dead and it's a rotten sack of Goo did you adopt
550. 1:28:27a dead puppy yeah that one's not good okay um Jesus that was bad all right we're doing no edits here this is first time joke
551. 1:28:37first time models okay so now we're on chat GPT this is chat CPT 3.5 avocados are like Goldilocks of produce
552. 1:28:49interesting that it used that same thing they're either too hard mushy or too rotten it's like they're trying to tell
553. 1:28:56us to enjoy them at the exact moment they hit Perfection or else they'll turn into a mushy Brown mess faster than you
554. 1:29:02could say guacamole so that is structurally much better than the sage one but it's it's still a bad joke like
555. 1:29:09it's just bad it's not funny okay all right let's go we'll go Club two so Cloud 2 is supposedly closer to gpt4 here's a short joke
556. 1:29:26my avocados are like an impatient toddler one minute their heart is a rock and then next they're a slimy mess
557. 1:29:34I swear I just bought them and blinked now they're brown and mushy avocados age faster than a child star in Hollywood
558. 1:29:42so again what's interesting here is it's got an interesting setup it's age faster than a child star in
559. 1:29:51Hollywood is that could be a punch line but my avocados are like an impatient toddler one minute their heart is a rock
560. 1:30:00okay that's just [ __ ] disturbing it's like no no no that's that's not how you do that okay now we'll try gpt4
561. 1:30:11GPT before better not suck at this well it can't get much worse like like I don't even know which one Claude instant
562. 1:30:18was the the one that was was better so far wasn't it or was it Claude two no it was Claude instant all right
563. 1:30:27this is gpt4 why do why do avocados have the worst sense of timing I mean you buy one and it's rock solid like a
564. 1:30:35green-handed grenade okay we're getting Maybe you check on it every day but it's still not right then one day you go to the
565. 1:30:43bathroom and boom you come back and it's turned into a mushy Brown guilt trip it's like the avocado waits for you to
566. 1:30:49leave the room and goes quick let's get rotten before they make us into guacamole that's a joke gpt4 wrote a joke
567. 1:30:58now is that like does that need to be edited and tightened yeah but that's a joke right it's like a hand grenade just
568. 1:31:06waiting to go off and then you go to the bathroom one day you know you keep checking on it day by day that's totally
569. 1:31:11a joke and that's totally a Jim Gaffigan joke all right let's go try um pie there's pie and then there was another one
570. 1:31:22so Pi this is the inflection company um start over
571. 1:31:37you're a tough crowd no problem let's start over shall we start there and avocado's lifespan is like a blink
572. 1:31:51of an eye you look at it one day and it's rock hard the next day you look at it again and it's a it's mushy and brown it's like
573. 1:31:59they've gotten the lifespan of a fruit fly how did I do horrible okay so Pi is not a joke writer um and then there was one more that what
574. 1:32:11was the other one there's pie oh Claude we did cloudpo.com okay so there's your verdict
575. 1:32:25for Jim Gaffigan style jokes gpt4 is the only game in town oh you know what'd be interesting to try let's go try Bing
576. 1:32:32because Bing is gpt4 but Bing has also been neutered so I won't flip the camera but I'll just I'll just pop it in here and let's see
577. 1:32:40what it does why did the avocado hire a personal trainer because it wanted to go from unripe to overwrite in 0.2 seconds
578. 1:32:59that's a very Microsoft joke structure and joke writing that's awful all right gpt4 at openai.com is the only place you're
579. 1:33:12allowed if you're going to use AI to write jokes you can only use gpt4 from open AI That's the new law
580. 1:33:21if I hear anyone in here say oh I used Claude to write a joke for my stand-up act I'll kick you out you'll be banned
581. 1:33:29forever that's it get out get out gpt4 is the only one that wrote a joke it's just like children they start out rock hard what no
582. 1:33:43no you can't say that you go to jail for that where are the safety guard rails yes to Sycamore six seconds down to 45
583. 1:33:59years yeah 47 years six seconds I know it's Insanity Insanity Insanity by the way if you're new here my name is Kyle Shannon this is
584. 1:34:08the AI learning lab we like to have fun here we write jokes We compare different large language models and see which
585. 1:34:13one's the funniest guess what none of them are funny but gpt4 at least understands what a joke is um what else do we do if you got
586. 1:34:22questions about AF i a i pop up in the questions below I'll do my best to answer them uh if you're wondering who
587. 1:34:28the hell I am and what are my qualifications I have none uh yeah I'm an entrepreneur I'm the CEO of a company called storyvine which is
588. 1:34:37an 11 year old automated video production platform a video storytelling platform uh I drink coffee at night
589. 1:34:46I live in Denver Colorado I'm a Taurus and I got some unfortunate things going on on the top of my head all right oh
590. 1:34:53yeah and the shirts are always a topic of conversation whether or not they're appropriate for tick-tocking because of
591. 1:34:59reading text on top of them you answer after I'd already hit send oh that's that's not a problem you can you
592. 1:35:07can put things in here tree logic is a heavy task now very heavy Portland lost most of its big centers years ago that's
593. 1:35:14interesting they lost their jobs due to automation years ago that's the thing about this AI stuff is the the AI stuff
594. 1:35:20has been around for what two or three decades and it's it's really been being incorporated into almost everything we
595. 1:35:27interact with for the past decade the new the new stuff about these generative tools is just this idea of
596. 1:35:33you know the the speed with which you can process everything and the the scale of what these things are trained on um
597. 1:35:42so yeah so I think it I you know I think the acceleration continues but yeah that that logic for phone trees is it's just it's got to
598. 1:35:54be excruciating to work on on the build side Bill Burr yeah that would be interesting to let's see I'm gonna go do a something here
599. 1:36:07was it Poe yeah I'm gonna say Bill Bill Burr and Tom Segura are my favorite um two of my favorites um so I'm gonna say you're a stand-up
600. 1:36:22comedy writer who writes for big name comedians like let's change this Bill Burr and to Tom Segura [Music]
601. 1:36:33you're an expert at writing short jokes about the funny things in life that most people overlook that can really piss you off
602. 1:36:47write me a joke about how how okay let's see if the joke gets any better so this is gpt4 but we switched out the comedian from Jim Gaffigan
603. 1:36:56to Bill Burr and Tom Segura you ever feel like avocados are just messing with us I mean you buy an avocado and it's rock hard I'm waiting
604. 1:37:04for days hoping it'll finally be ripe enough to use but the second I turn my back do a little Bill Burr second I turn my back
605. 1:37:11the avocado goes from unripe to rotten faster than politicians promise it's like they're sitting there smirking
606. 1:37:17thinking hey you want to eat healthy right well here's a challenge for you get me at the right time or enjoy the taste of disappointment
607. 1:37:24that's not bad gpt4 can write jokes GPT for walking right jokes jokes jokes are they really funny no but I'll tell you what if you were
608. 1:37:38you know you know you could do a gpt4 you could do this in a spreadsheet with the chat with the GPT for Sheets plug-in
609. 1:37:47you could you could automatically generate a whole bunch of premises like why do avocados go you know you know go rotten so fast
610. 1:37:56whatever just have have it right like 50 joke premises and then just have it just keep iterate just keep writing jokes
611. 1:38:04just keep writing jokes I you know you spit out 300 jokes in an hour there's got to be five in there that are worth worth saying
612. 1:38:15Bots aren't overly new the logic displaced a lot of Entry levels CC work on the high end yeah well I I get the
613. 1:38:22Bots aren't new but like what what I haven't seen I mean maybe it exists it's just not any company I've ever [ __ ]
614. 1:38:28called is that there should be no reason that I'm on hold like I shouldn't need to go into a phone tree waiting to talk
615. 1:38:37to a person I should just immediately be able to start talking to a bot and then that bot should be dynamically
616. 1:38:44like it should just have enough access to all of the information that it can answer all the questions and if it can't
617. 1:38:51answer a question then it escalates hang on Champs gotta go pee you want to see Mama yes
618. 1:39:06your boy needs to feed me Jeffy needs to be [Music]
619. 1:39:21[Music] um all right um
620. 1:39:37Sean I hit enter before I was done again I'll just I'll just listen hands are too sweaty exercise bike I'm glad I'm glad
621. 1:39:47you clarified why your hands are sweaty oh food egg by bio industrial trading shipping so not funny but you can feel a tinge of
622. 1:40:00it yeah exactly like those at least GP T4 understood joke structure it was a little wordy and it was a little not funny but
623. 1:40:15like I feel like the GPT 4 joke was it was at least close enough that if that if I just were if I were just stuck with
624. 1:40:22writer's block like I can't figure out how to make this premise work I can't figure out how to make this premise work
625. 1:40:28and I told chat gbt4 to give me you know 10 possible jokes for that and and I and I seated it with some of my work or if it knew my work um
626. 1:40:39I it it would it would at least get me unstuck I've been live for non 90 minutes that's dark comedy that's disturbing exactly
627. 1:40:49that was the kid one this is a great experiment that's funny best one interesting snooze yeah I know exactly
628. 1:40:56is bring is being free with Microsoft 365 it's just free um time for bed on the East Coast peace Jen thanks for being uh entertaining me
629. 1:41:07tonight you're welcome um Kyle I'm too humble thank you Becky I appreciate that uh it was great fun tonight thank you
630. 1:41:23disturbing their conceptualization yeah the conceptualization of puppies and babies yeah it's just like a puppy you bring it
631. 1:41:31home and you have two days before it's a gooey pile of you know mess hmm avocados aren't funny anyway try a rabbi
632. 1:41:45and a priest yeah exactly all right you have become my total nightly entertainment education I certainly appreciate you teaching thank you
633. 1:41:57appreciate that bummed out yesterday when chat GPT 4 servers went down yeah I know that was weird because of code interpreter Maybe
634. 1:42:07yeah it's possible because because code interpreter is spinning up a bunch of virtual machines and that could be the way that
635. 1:42:18um chat gbt works but I I have a sneaking suspicion that's different so it it might be code interpreter might have might have
636. 1:42:27hosed might have hosed the servers but I was it was acting really weird last night so I ran a I ran a melt Mount malware scan after my
637. 1:42:41my live last night I was having weird malware like captcha things show up and I didn't I didn't have any I'm on a Mac
638. 1:42:48so so generally we don't have a ton of ton of malware stuff on Macs but I don't know I don't know what's going
639. 1:42:56on T just gave me this joke why does why don't skeletons fight each other because they don't have guts
640. 1:43:03it's if you're writing a joke for sixth graders and not even these days doesn't feel like a joke for fourth graders these days third graders
641. 1:43:16um someone in the chat just made a great Point avocados they're not funny it can be funny but you just need to make fun of
642. 1:43:25hipsters with them on toast right isn't that the way to do it someone uh oh Sonia so so bing.com requires you to be an edge
643. 1:43:41for it to be free is that true wait let me go look I'm gonna go run Chrome if that's the case I'd like to
644. 1:43:49know that because I don't want to steer people wrong because I've been telling everyone it's free just go there you can use it you can do it
645. 1:44:02in calm chat
646. 1:44:21oh yeah look at that Bing chat is only available in Microsoft Edge look at Microsoft getting all monopolistic and all
647. 1:44:36they're like oh yeah that Chrome thing from that Google company you can't use that no no you're gonna have to go on
648. 1:44:46ahead and come on over to the dark side we got your free gpt4 all you can eat it's here for the taking
649. 1:44:55just download Edge mcping your default it'll be fine that's pretty [ __ ] evil all right I'm glad you said that thank you for saying that
650. 1:45:08so Bing is free salon because you sign your soul to the devil oh Microsoft your ways are dark and gnarly no
651. 1:45:25it gives me a perverse thrill to see how you run your business leverage people extract money from them it's delicious
652. 1:45:39avocados are like children they start out rock hard what what no that's not the joke what was the big one the Bing joke was
653. 1:45:53even worse why did the avocado hire a personal
654. 1:46:07trainer because it wanted to go from unripe to over ripe in 0.2 seconds yeah Microsoft I know it's listen they're doing they're
655. 1:46:22they're doing the good work they're they're getting this AI stuff out in the world we're gonna we're gonna give them we're gonna give them
656. 1:46:29some benefit of the doubt but you know I do have 35 or 40 Years of Microsoft bashing to get over code interpreter wasn't working for me
657. 1:46:40today mid-afternoon EST but maybe because I was waist deep in H2O are you in Vermont hopefully not good Lord that looks like a mess up there
658. 1:46:55my MacBook Air made me download Edge to use Bing chat feature yep looks like that's what they're doing it
659. 1:47:02won't write dirty jokes yeah they won't you're gonna have to go to the open source models for that and then I bet I
660. 1:47:07will write some jokes that are so dirty that you're like oh good God geez is this connected to the internet is the FBI watching yeah
661. 1:47:20don't go to those unsupervised models if you you know if you don't want to get smacked upside the head with some some freedom
662. 1:47:32wait a few weeks wait wait a few weeks ago you said they were ahead in the AI space what's changed I said who was ahead in the AI space
663. 1:47:42um open AI what's changed is in the past two days anthropic came out with Claude um inflections uh
664. 1:47:58Pi is out there um Poe now has gpt4 in it they got rid of browse with Bing
665. 1:48:13um and it was just wonky yesterday so but I don't know if that's what you're asking a few weeks ago said they were ahead in
666. 1:48:19the AI space I don't know who you're talking about Becky if you if you let me know I'll say it yeah Bing no yes oh yes
667. 1:48:26Vermont oh I'm so sorry sending all of my best wishes and thoughts to Vermont Vermont I used to I did a summer theater in Warren Vermont
668. 1:48:39for like four or five years and uh damn it's beautiful up there but holy crap that flooding is insane like really sad really sad
669. 1:48:53can I just hear AI do you get it do you get it is that something I say you get it you get it I don't know I don't know if
670. 1:49:04that's something I say it's one of the joys of add I've got all these you know verbal tics that I have no [ __ ] clue
671. 1:49:10what I'm saying half the time so um is it true that chat gbt cannot spell lollipop backwards I don't think it can
672. 1:49:18do things like spell it backwards or give me exactly this number of things or tell me how many things that you've
673. 1:49:25created it all of that stuff it does poorly um we can try it um spell Lolly pop back words
674. 1:49:52pillow Paul no it did not come close nope it can't but it makes sense it if if you understand how it does what it
675. 1:50:05does it has no awareness of of what it's actually created it only has the awareness of what's the probability what's the highest
676. 1:50:15probability of the the next word in the sequence based on your prompt and so until it gets better at self-reflection essentially it's going
677. 1:50:27to be shitty at stuff like that because that's bad you know it would be interesting um am I in code interpreter
678. 1:50:36no let me try it in code interpreter um right a python right let's see right
679. 1:50:52and run a python script that um spells
680. 1:51:09lollipop backwards let me run it for you yeah it spelled it wrong again
681. 1:51:25oh no it didn't pop it law so so it if if so I asked it to write a python script so because it's writing a python script
682. 1:51:41it can actually write the logic and and it it knows the logic of the script and it can run the script and it ran it and
683. 1:51:47it got it right so if you have it if you have it right code to do that it can do it right but just the way the large
684. 1:51:52language model Works things like counting predicting like predicting length of things um getting an exact number like give me
685. 1:52:0113 bullet points and it'll give you 12 or 14 or 16 and you're like well you give me 14. it's like oh my apologies
686. 1:52:08here's 13 bullet points and then it gives you 16. that's just the nature of how the thing works but if you use code interpreter
687. 1:52:16and have it write scripty scripts for you then then you can start to do things like actual math and actual prediction and actual
688. 1:52:23word jumbles or or you could just go down to the grocery store and get the word jumble in checkout in the checkout line do they
689. 1:52:32still sell those is that a Gen X thing do those still exist the word jumbles no that's just because I'm old right
690. 1:52:46that sent that one through the nose is it true yeah cannot spell lollipop backwards that is correct oh Microsoft so so I think Microsoft
691. 1:52:57Becky is um I think for a company the size and nature of Microsoft they are acting more like a startup than they have in a long long time
692. 1:53:09I think the fact that they are so aggressively um incorporating open AI chat GPT into all of what they do they're placing a big Bet On It
693. 1:53:23um I found out today they're one of the major investors in um inflection which is the which is a competitor to open AI
694. 1:53:34um so Microsoft's a big company they're going to do what they're going to do um you know they will always tend toward
695. 1:53:44the more conservative to protect the revenue of what they've got um but yeah I think I think listen for a company as big as they are
696. 1:53:53what they've done in the past six months is nothing short of remarkable and you know whether or not they can sustain that
697. 1:54:01um and really take it to Google long term is going to be really interesting to see my my gut tells me yes they're
698. 1:54:09going to be able to do that because I don't I just don't I don't see goo see Google has the engineering skill to do the science of this
699. 1:54:20really well I don't know that they have the usability and human factors expertise like an apple does you know Microsoft
700. 1:54:29isn't great there but they've gotten a lot better in the past six months um and maybe before that like frankly I
701. 1:54:36just I haven't interacted with Microsoft products in so long that you know it's it's possible they've been improving
702. 1:54:42over the years but the stuff they've been doing with the AI integration is is really solid act like a spelling teacher and spell
703. 1:54:53lollipop in reverse order did it right oh that's interesting yeah I mean that's the thing the the the
704. 1:54:59code that Microsoft used to program Bing chat which is sitting on top of a gpt4 large language model and what openai did
705. 1:55:09with theirs like it's two very different code bases sitting on top of the same model so it's fascinating to compare those two
706. 1:55:17and and I get you know my my sense is that long-term Microsoft's Bing chat is just gonna get better and better and better like it's
707. 1:55:29kind of like this like GPT three 3.5 and 4 for for open AI did this and now they've kind of plateaued and it's even starting to come down a
708. 1:55:41bit I feel I feel like Microsoft went up to maybe like three quarters of where they were like you know open AOS here
709. 1:55:48Microsoft came to here and then Microsoft had to bring it down to like 50 because of the New York Times thing
710. 1:55:55and I just feel like Microsoft is going to be on this slow improvement over the next two years or three years it's just
711. 1:56:01going to get better and better and better slowly steadily um and like when they roll out co-pilot or chat GPT integrated into
712. 1:56:12office 365. it's going to be very well thought out it's not going to be some janky ass add-on so I think I think Microsoft's a really
713. 1:56:22interesting one to to pay attention to do when little kids tell ridiculously toward jokes they always say if you get
714. 1:56:37it oh do you get it do you get it I get it I get it that's the that was the Bing joke that's so bad why did the avocado have a personal
715. 1:56:49trainer because I wanted to go from unripe to ripe from 0.2 seconds you get it get it oh that's bad it's so bad
716. 1:57:00Celebrity Jeopardy was a rerun last night so why are you here tonight about last night about last night and
717. 1:57:09failed miserably wait what I don't know if that's a reference to wait Claude allows PDFs as input just like that yes
718. 1:57:17I just had to summarize my employee handbook yes so look this is so a couple of things to show overnight from from when we talked last night
719. 1:57:28there's new things so right now I'm in po po.com um you used to have access to Claude instant 100k now you have access to Claude 2 100K so
720. 1:57:45Claude 2 is like the equivalent of gpt4 but they now have a version with a hundred thousand token um context window so both of these clods
721. 1:57:56in Poe have 75 000 word context windows and gpt4 right now has four maybe eight I don't know what it is here but I think it's four
722. 1:58:07um the other thing that's going on at Poe I got this email today this morning see that little paper clip you can now
723. 1:58:15upload files into Poe as well so that's pretty remarkable
724. 1:58:31um so that's new um what else is I going to show you if you go to claude.ai
725. 1:58:46um hair growth geek if if you want to hop up on stage your thing just popped up here if that was you doing that just
726. 1:58:50do it again I'll bring you up on stage that might have just been Tick Tock going hey do you want to do this but if
727. 1:58:56you do if you have a question or want to hop up here let me know um okay this is pi so if you go to claude.ai foreign
728. 1:59:11this now also has um a little paper clip here you can upload up to five files so five PDFs or I I don't think this will do zip files
729. 1:59:22let me let me see if this does zip files I don't think it does but you can do five PDFs um yeah you can't do zip files but you
730. 1:59:33can do csvs and PDFs well dogs singing Goodwill Hunting screenplay so let's open that so now it's got the Goodwill Hunting
731. 1:59:47screenplay in there so I'll say um who are the main characters and who likes apples
732. 2:00:04I don't have enough content to know so it didn't do that well all right failed there anyway Claude and Claude
733. 2:00:13dot AI also has a hundred thousand token context window so um so a lot of things changed overnight like I I feel like there are legitimate
734. 2:00:25chat gbt competitors in the market now um and there haven't been there really haven't been I mean being in being an
735. 2:00:33open AI we're kind of in cahoots Bard just wasn't anywhere close meta didn't all of their stuff was crap um
736. 2:00:43the Claude stuff was okay but not great but then Claude 100K now Claude 2 with 100K plus you can upload files
737. 2:00:52thank you for that Sean I appreciate that um all right why do things that will take longer to input
738. 2:01:07the prom oh yeah but why do things that will take longer to input The Prompt than do the thing manually that's that's
739. 2:01:14just one of those things with computers you know how you spend two hours trying to get you know some piece of software
740. 2:01:20working you download it try to run the demo then you find out the demo doesn't actually two hours later
741. 2:01:27you realize oh I could have just written that thing in that amount of time because it's new technology and we're
742. 2:01:34geeks and Geeks Geeks are gonna geek so you know and you know quite frankly the the reason to invest the time into overwriting a prompt
743. 2:01:49is that when you get it right it's scalable like if I if I really [ __ ] around with that joke writing prompt like there's I
744. 2:01:58guarantee you there's a way to write a joke joke writing prompt with a few example structures in there um where you could get it to
745. 2:02:08consistently write the kind of jokes that you tell like it might take you two hours a session for ten sessions five sessions
746. 2:02:22where you could have been writing jokes but if you get that [ __ ] dialed in you could you could really start
747. 2:02:28cranking out some interesting stuff and and change up your act and you know so so so I I think there's I think there's
748. 2:02:35there's an investment in learning how these things do what they do that's worth worth that waste of time but it's a it's a great point
749. 2:02:46funny funny jumbles I know it uh have you have you ever asked it to list U.S presidents who died in office I have not
750. 2:02:57I'm I'm wait I'm going to get a raise with this so much of what I do is hyperspecific so the PDF helps definitely definitely go get a raise
751. 2:03:08with this I'm telling you we have a short window where if you've got Senior Management that's slightly curious about
752. 2:03:16this but doesn't really feel like learning it if you can pick on the expert and make them look like Superstars
753. 2:03:25there's there's going to be a whole lot of new VP of AI generation in companies that were like you know had some sort of like you know
754. 2:03:34entry-level associate level project management all of a sudden wait why are they the VP of AI prompting integration
755. 2:03:42I don't know she went into Bob's office and came out with a raise and a new office go make Bob look like a superstar
756. 2:03:56I mean Bob's a dick but he can give you a raise foreign thank you all enjoyed it bye Joy thank you iced coffee man oh yeah Charles
757. 2:04:10oh ah it's a lovely bouquet do you think Chad GPT is bad no I think chat GPT is going to fundamentally change the world
758. 2:04:24I think chat GPT is the inflection point after which nothing will be the same I think Chachi PT is remarkable I think
759. 2:04:34it's the most powerful tool put in the hands of regular everyday people probably in history I think it's remarkable um
760. 2:04:44and it's not perfect it hallucinates meaning it lies it's overly wordy sometimes underly wordy sometimes um
761. 2:04:55will confidently produce shitty work and think it's good but doesn't think anything it'll confidently produce shitty work
762. 2:05:02and then if you really push it hard then it starts to get better and then if you really push it hard after that you can
763. 2:05:10start you realize how remarkable it is it's it is truly remarkable technology the thing that makes it weird right now
764. 2:05:19is that it's so raw and early and it was tossed together chat GPT was written in two weeks two weeks now the underlying model underneath it
765. 2:05:34that was in development for seven years one gpt2 gpt3 GPT 3.5 chat gbt just a little shitty thin application layer sitting on
766. 2:05:49top of this powerful model and then that little thin application layer 100 million people started using it in
767. 2:05:58the first six weeks fastest adoption of technology in history so I think chat EPT is [ __ ] remarkable and it's imperfect and new and duct
768. 2:06:11tapey and spit and string are holding it together it's just the nature of being so it's the price you pay for being
769. 2:06:21early if you want all this stuff to be slick and perfect wait for a year and a half and it'll be you won't even the whole idea of having
770. 2:06:29to prompt stuff I think that drifts very quickly into the background you'll just say hey I want this and
771. 2:06:36it'll go figure out how to do it right let's see I'll do a few more questions here and then I gotta go
772. 2:06:44isn't even on building a competitor as well in AI never bet against Elon yeah he just hasn't come out with
773. 2:06:52anything yet he's a little distracted with he he's got a cage match to prepare for and and Zuck you know Zuck decided
774. 2:07:00to to clone Twitter just to piss him off so I think I think musk is is going to be a little distracted I mean don't get
775. 2:07:09me wrong the man could he shouldn't be able to run two companies and he's running what seven right now and you know many of them are
776. 2:07:18changing the world so I'm not going to put anything past the guy but he's a little busy right now so I assume that
777. 2:07:24we're not going to see anything from X a i for probably till early next year and then they'll probably come to Tesla so all he has to do is
778. 2:07:36pluck three of them out of that and say hey go take all that [ __ ] you did here and go spin it up on these 10 000 gpus over here and
779. 2:07:47make it you know 100x what open ai's doing and tell me when you're done that I mean that conversation's probably already happened
780. 2:07:55how long before AI will replace news anchors and other types of Information Services I think I think that AI May reintroduce news
781. 2:08:10anchors for local news that no longer exist so I I don't see it as an as a net loss I don't I I think big networks and
782. 2:08:23things like that you're going to have you're going to have um you're gonna have news anchors you're gonna have regular news anchors I think
783. 2:08:32though what happens in the smaller markets where maybe the the local newspaper the local TV station has been nationalized
784. 2:08:41because it got bought up by you know Gannett or whatever the [ __ ] it's called now and uh tegna I think it's called now you
785. 2:08:48know one of those big companies or one of the big newspaper companies that buy up all the local markets and then the local
786. 2:08:54the local news economy literally just goes away so so you only have national news being masqueraded as as local news
787. 2:09:04I think we could see independent journalism come back with newspapers that are generated from Ai and with news broadcasts that are
788. 2:09:14generated from AI in an economically feasible way that that brings local journalism back and I think that would be good there's a
789. 2:09:26business to start although I gave that one out two nights ago all right I wanted to make jokes from thousands of
790. 2:09:35joke structures that I can input how to force it to use the structure you can just so go to claude.ai go there or go to poe.com and pay the
791. 2:09:47subscription but you can do it for free at claude.ai claude.ai has a context window of a hundred thousand tokens that's 75 000 words so put
792. 2:09:59categorize make a spreadsheet of your thousands of jokes in your different categories upload that or copy and paste it into
793. 2:10:09the prompt and say based on all of these jokes in these categories I want you to ask me for a topic and a premise
794. 2:10:21like a subject and a premise and or or a a premise and a and a joke category right I want a knock knock joke
795. 2:10:30I want a story I want to this I want to that and you've got them all categorized up above so ask me you know for a premise and
796. 2:10:39then I want you to give me a joke per category whenever I give you an answer and then and then I want you to and I
797. 2:10:50want you to ask me which of the jokes you wrote is the funniest and then I want you to give me 10 more like that you could do that right now
798. 2:10:58you could just like I would go right now to claude.ai and do that Mr AI a i o oh fantastic Name by the way kudos to the Mr AI a i o come on
799. 2:11:10huh do you see people how it's done do you see how it's done Elon never have a generative AI project unless he buys it we'll see you know
800. 2:11:22we'll see I mean he's not a he's not a dumb man he's he's actually a smart guy I'm up to over 9 000 likes whoa [ __ ] tap
801. 2:11:39that like button let's give me over ten thousand thank you for the for the heads up I don't pay attention to any of this
802. 2:11:44[ __ ] I don't know where are you seeing this total views diamonds good connection I don't know oh here's my duration I've
803. 2:11:55been at this for two hours and 12 minutes right as my butt's getting numb it's like clockwork guess who used chat GPT to work on their
804. 2:12:05resume and job app today fantastic yes like chess GPS great for that kid's bedtime thank you thanks Sean appreciate it
805. 2:12:16all right lol Kyle pause when you switch from front to back camera yep we miss what you say sorry about that what Bill Gates Affliction with
806. 2:12:25Microsoft now what's Bill Gates Affliction with Microsoft now I don't know what you mean by that did something happen that I missed
807. 2:12:33or like what's his oh what's his affiliation is that what you're asking uh made more sense to chat GPT than backwards
808. 2:12:43have a good night apple apple it's a cult it is a cult I'm part of that cult I'm an apple Fanboy from way back
809. 2:12:50gotta go pleasure as always thanks silver fox s b okay there's all that stuff I'm gonna wind this puppy down
810. 2:13:00Claude is great but they get but they better get their act together on scale it was super slow today I know I mean listen this is
811. 2:13:10you know Elon can be pissed at Sam Altman all he wants for taking that thing for profit but the fact that open AI partnered with
812. 2:13:22Microsoft and Microsoft could throw their entire Cloud infrastructure at making this thing work like if think about if you threw together
813. 2:13:37a little piece of software in two weeks let's say you did it in a month it took you a month and you launched that piece of software and that
814. 2:13:47piece of software even went to a million users in two months could you handle it no if it went to a hundred million users in six weeks like
815. 2:13:57chat GPT did like that thing shouldn't be functioning right why is it functioning because they cut a deal with Microsoft and part of
816. 2:14:06that deal was we're going to launch this thing and if if by some stretch of the imagination it [ __ ] takes off
817. 2:14:13we want to be able to scale up on your Azure Cloud infrastructure like that that's why that thing's still alive so
818. 2:14:20yeah I totally agree that claude's gonna have to get their [ __ ] together and if you look at the um the the investment uh you know what
819. 2:14:30the what the folks at inflection are talking about what they're going to spend the bulk of their 1.3 billion dollars on it's building
820. 2:14:38a massive data center to be able to scale so I think you're right I think anthropic's got to figure that [ __ ] out pretty quick
821. 2:14:47like they could have could have here your answers but if they're too slow and don't work like now that we've got thank you
822. 2:14:58at least three kind of major competitors and pretty soon we'll probably have five or six like we as humans we can only like we
823. 2:15:10can do only do things in three we can only do things in three so let's say there's six major competitors with these large language
824. 2:15:17models Plus open source only three emerge out of that and that's gonna it's gonna the the winnowing is gonna happen quick so who survives
825. 2:15:28it's gonna be someone who's got some [ __ ] infrastructure that's for sure um easily how long before AI webcoast news anchors
826. 2:15:38we did that the Thousand jokes when we did that that's a good idea regarding local journalism Revival I
827. 2:15:45think it's actually a really important thing that could be a good business that could be absolutely solved with this AI stuff
828. 2:15:53probably pretty simply I was even thinking about I think I might do a zapier automation um to um to mock it up because I think you
829. 2:16:02could probably pull local public records and with a zapier automation create a daily newspaper a daily local newspaper for your neighborhood
830. 2:16:13why not prompting wouldn't be that tough it's just sourcing the public records and automating it not that tough now what would be tough
831. 2:16:25is how do you turn that into a business but that would be an interesting exercise to play with thanks for answering my questions you're
832. 2:16:32welcome joke sir joke sir Jack sir the mighty jokes are the mighty oh jokes are the mighty over ten thousand likes now peace thank
833. 2:16:40you people yes sir affiliation I'm assuming he's still on the board uh Bill Gates affiliation with Microsoft I think he's
834. 2:16:49still on the board but he's definitely out of day to day I and he doesn't it's funny he doesn't he doesn't seem to have
835. 2:16:55a horse in the game although I think go look at the investor um the the investor list for inflection I think Bill Gates and Microsoft both
836. 2:17:07invested in uh in inflection thanks Kyle you and your hair made another fun evening I know oh it's getting worse if you see it's [ __ ]
837. 2:17:17losing it it's I'm ready to take it's like a helicopter that's like if you said to Mid journey I want you to show spinning helicopter
838. 2:17:25blades I think that's what it would make all right um I'll leave that one as the final comment thank you all so much for hanging out
839. 2:17:39peace good night Chris peace out everybody I'm gonna go do something with this hair act like Kyle's here peace everyone