Feb 20, 2024 - The Future of Generative AI, Autonomous Agents, and Multimodal Capabilities

Transcript

1. 0:02you [Music] [Laughter] [Music]
2. 0:21ready us Mar our FS together real estate here in my so we bought a pack of
3. 0:39cigarettes Mrs Wagner's pass we've all come to look for [Music]
4. 0:58America [Music] we bed the Greyhound in
5. 1:16Pittsburgh Michigan seems like a dream now so we bought a cigarettes this wag pass wrong words to that verse we've all come to look for
6. 1:33[Music]
7. 1:55americ no rapping no rapping what what are we rapping for that was Simon and Garfunkle it was Simon and Garfunkle you
8. 2:05don't rap to that sure the Disturbed cover of Sounds of Silence incredible you can go heavy metal with Simon and gar funkle but rap I don't
9. 2:15think [Music] so now you're not singing at all oh this is it silent protest I'll hit you with an minor I'll
10. 2:30hit you with an A Minor I'll do it oh yeah see I can get you every time with that [Music]
11. 2:58one [Music] yeah you good you done did are you
12. 3:14apologizing you're apologizing for getting so angry yeah I know I know I I know I know I know I know I know I agree I agree hey I don't
13. 3:30like it when we fight either I HT his feelings exactly he's like I just want to rap Daddy I just want to
14. 3:45rap I think it's music [Laughter] too ah fanastic Bob tell him what he's won you've won a chicken stick
15. 4:02[Applause] all right fantastic where's champy coming from Champ's coming from the well he came from the bed he was on the bed and
16. 4:13I told him I said you you can't sing from the bed like get off and come come to the office come do your job she
17. 4:25did but he was a little cranky I think his blood sugar is a little low welcome everybody Welcome to the AI learning lab
18. 4:33uh had a had a a nice uh uh AI Salon after hours it was cool it was it was a small group and it was a lot of the
19. 4:41original OG salers um Roger was there Jane CJ um and who else was [Music] there anyway lots of people it was good
20. 4:56it was really good we talking about um we were talking about Sora and then we were talking about I don't know if if any of you on here
21. 5:11remember when I did some of the live Kaa stuff the the the quick image Generation stuff that at ca. where you can turn on
22. 5:19your webcam and it it interprets your webcam and does cool [ __ ] with it um we were we were talking about like
23. 5:29that would be such a cool performance to put together a stage show where you're having AI render actors in real time there's something
24. 5:39about that that just keeps keeps pulling on me my my the old theatrical the old theatrical instincts die hard you know what I'm
25. 5:51saying this shirt's hard to read on you just have to read up here at the neck I've been using ideogram ide's pretty good we used
26. 6:01ideogram a lot in here before uh before Del got good at words what about the long distance OG Salon Salon folks oh I
27. 6:10know well that's the thing is so we're we're moving to inperson meetings every other week and then virtual meetings
28. 6:17every other week cuz the hybrid just wasn't working so um so we're getting there we're getting there it's you know
29. 6:25it's a volunteer organization so things move at a certain pace if you know what I [Laughter] mean oh Lordy I just leave it on my laptop like
30. 6:39a changing canvas that's actually pretty an interesting idea you know one of the ideas I had that I I think Co sort of killed this but
31. 6:50um one of the ideas I had for Creative groups is just to have or if it's like like with a group of friends just have
32. 6:58your web cams on all the time and don't have it be a meeting just have your webcams on and you're each doing your work
33. 7:06independently and then at some point if someone gets bored they can just go hey uh Bob what uh what are you doing I'm
34. 7:13trying to get some work done you jerk all right okay and just you know like you were in a room working with some
35. 7:20friends I always thought that would be kind of cool I tried it with a couple of writing buddies lasted about a week but
36. 7:26uh there's something there oh lordy okay so here's how this works if you have questions about AI pop them in the comments below
37. 7:36um do me a favor feel free to share this live share it with people who you think might be remote support group yeah
38. 7:44exactly pretty much that or just a remote hang like like thinking of of telepresence not as I've got to be on a
39. 7:52meeting I got to be on a zoom call but just more like hanging out with your buds but you know living your your lives
40. 8:00independently um and if you need to have a conversation you can have a conversation or if you're working on a
41. 8:05project everyone can go do their stuff and then go hey I got a question Deep dive thoughts on the combination of generative Ai and spatial
42. 8:13Tech like apple apple Vision Pro um I don't think they can exist without one another um
43. 8:28I looking at what um looking what open AI is doing with Sora and how they're doing it and then thinking about what apple is likely
44. 8:38doing behind the scenes with AI and how they're doing it I think that um I think that the spatial Computing
45. 8:49game is going to is going to get really good really quick Kyle that's exactly what I was talking about when I
46. 8:54suggested an Open Lab like old computer labs oh that's interesting that's interesting I didn't I didn't see that unless is that
47. 9:03something you sent me directly or is that something you posted posted on the salon Amelio's wife because I that that
48. 9:08would be a cool that would actually be kind of a cool Club you know we've got an's Club on Saturday the GPT jams we've
49. 9:16now got um the business Guild has their their their Saturday morning um business workshop and uh yeah maybe there's
50. 9:27something where it's just like there's no presentation and and it's just it's just a hang and but I think it's got to be in a
51. 9:36relatively small group because I think if you have in a big group someone's going to want to talk so I think part of
52. 9:41the thing that makes it cool is you have it on for like 4 hours or five hours or you know just a while and it's a working
53. 9:48session and everyone's working and if someone has a question they can ask it but it's it's not a it's not a a thing I sent it directly okay
54. 9:57cool I'd be happy to host start it I think that's really cool Amelio's wife and absolutely feel free to start it I
55. 10:04mean if you want to start something like that let's let's let's talk and we'll we'll sort of figure out logistics on how to get you
56. 10:11supported um we can go live on mudy networks we can Mighty networks has a weird thing I don't quite understand it
57. 10:19they've got a 20h hour a month limit on going live in Mighty networks but you can schedule it Mighty networks so it shows up as a as a
58. 10:35uh like a calendar event and it shows up like in the sidebar that someone's live but then you can just have it be a zoom
59. 10:42call or a Google meet or something like that so that's kind of cool um hang on I just got a I got a something oh interesting
60. 10:56oh huh so Brad Perkins right now is at the Avs game and they're they're doing AI at the
61. 11:10Avs game you know leave it leave it to Big Sports Arenas to you know they they realize they've go to they got to have their game
62. 11:19rolling that's really [Laughter] interesting really funny right that's super easy to do right they just they just turn the camera around I
63. 11:33don't I assume they're not doing it in real time yeah it's called The Grateful Dead heads and then they probably took a
64. 11:42picture of that person at some point threw it into one of the AI things very very interesting and this is the thing you
65. 11:52know there's all the controversy right now about are you using AI you should label it if you're using AI AI is just going to become this
66. 12:00pervasive thing like it's just going to be everywhere it's going to be in your software applications it's going to be in your hockey games
67. 12:10um it's going to be everywhere and it and it's it's not going to have such clear boundaries did you write that with AI I don't know some of
68. 12:21it which part would you like to know uh I I so it was AI so AI did so robot wrote it no wrote it well you said you
69. 12:30used AI yeah like I used spell check and I used sentence completion and I used grammar check and the third paragraph I
70. 12:38couldn't figure it out so I had it write it but then I rewrote it the outline was AI the idea was AI me choosing the idea that was me me
71. 12:50figuring out what I'm actually going to publish that was me so you tell me where's the boundary right I think we're
72. 12:55going to run into that a lot a lot lot lot lot lot but right now it's it's like this scary Beast cuz all the tools are separated
73. 13:05right all the tools are like these janky ass duct tape piece of crap tools off to the side Amelia's wife she kind of already
74. 13:17did start it oh good awesome that's great see the salon is growing so fast that I I don't know what's going on in
75. 13:26there anymore so I've also been really busy at work so that's going to happen um all right so let me let me see what
76. 13:33we got going here see if we got any any interesting comments or questions or things like that and then we'll we'll jump in and we'll do some
77. 13:41stuff we'll do some stuff The Hangover League how do I know you're not an AI well you don't do you wouldn't it be swell if I were chat
78. 13:52gbt oh [Laughter] yes
79. 14:09I need some help beside chat jp4 what kind of help do you need Mr it Kevin Vasquez got the number three gifter
80. 14:20badge I didn't even know you won the number two gifter badge look at you good Lord are we still doing AI Salon meeting
81. 14:28all around I don't know what that means next Tuesday is the next AI Salon meeting we've got Kelly Bosch is speaking which if you haven't
82. 14:38seen her [ __ ] on Tik Tok go check it out b o e c I think is her name Kelly her work is [ __ ] amazing it's
83. 14:52amazing amazing amazing amazing so what else do you want what do you want to talk about tonight Kyle did this once
84. 15:00did this once in hey Jen nine months ago did what oh Kyle did did I cloned myself yeah I cloned myself a while
85. 15:12ago I actually just added it to we've got I've got a board board meeting tomorrow so I can't I can't stay up late
86. 15:19people I have a board meeting I don't know if you know this but I'm a very important founder and CEO of a
87. 15:26corporation we have a board meeting tomorrow so oh my gosh Kelly bsh she's amazing isn't she marry Mary what's what's amazing about a lot
88. 15:42of the people that I'm connecting with that I think are doing really amazing work in AI almost to a person to like two exceptions that I can
89. 15:56think of right now people are apologizing for their work oh I'm not really good it didn't really take me too long you
90. 16:08know I'm I'm it is my mission right now to stamp that [ __ ] out her work is beautiful it's got a really clear
91. 16:18creative point of view it's got storytelling in it it's got continuity in it it's got incredible taste in music incredible
92. 16:27editing and I honestly don't care what tool she used to do it it's not my business and she is she no one no one has
93. 16:41anything to apologize for using AI tools I'm gonna stamp that [ __ ] out I'm gonna stamp it out seems a lot more resistance to AI
94. 16:55than I thought yep burning autonomous taxis in San Francisco yeah they're going to be mean to the robots for a
95. 17:02while and then at some point the robots are going to get guns installed in them and then then when the taxi can shoot
96. 17:10back I don't think you're going to light it on fire the the CEO's just we got to protect our investment this is a large
97. 17:25capital expenditure well while there are some downsides to arming the taxis thought it was a a dollar cost
98. 17:37analysis of the the risk reward factor of the AGI taxis exactly where would you like to go can I answer any questions about the
99. 17:48Mesopotamian War and don't [ __ ] with me or I'll shoot you I want to be a CEO you can be a CEO you just you just have to [ __ ] lose
100. 18:02your okay step one lose your mind step two start a company oh man all right seems a lot more resistance Jason's so freaking bad
101. 18:16that's amazing she's amazing one day I know she's amazing I know Cory it's nuts you got a board meeting tomorrow I
102. 18:28do have a board meeting tomorrow Kelly Bosch AI Art thank you Silver Fox yeah and she says she does it on her
103. 18:36phone crazy yeah Kelly BOS she says she she does it on her phone um just kind of as an absent-minded thing before she goes to
104. 18:47bed but I mean isn't that [ __ ] cool like some of the people in here are like I don't watch Netflix anymore I just
105. 18:54come hang out in here and do AI [ __ ] and she's like you know before she goes to bed rather than watching TV she's making incredible
106. 19:03art like we're just in a we're just we're shifting there's a shift there's a craziness there's a there's a there's a
107. 19:11thing happening people there's a thing happening her work is spectacular my faves are the ballerina the ballerina
108. 19:18bugs are amazing that that one that one video she did the one she's got pinned at the top of her account with the birds
109. 19:26on it it's got like 22 million views or probably more at this point last time I looked it was 22 million views so cool
110. 19:33and and you know what's you know what's the thing the 22 million people that looked at that some percentage of them probably
111. 19:42knew it was AI some probably went oh that's good AI art but 22 million people don't look at something because it's AI
112. 19:51art 22 million people look at something because it's good and it resonates and it moves them and that's what her work did she's
113. 20:04like I can't believe it cuz in her mind this is something that I think we're going to have to Grapple with as humans
114. 20:11as these tools get better and better and the amount of time it takes us to generate work shrinks I think we're all personally
115. 20:20going to have to come to grips with the fact that that we may place a lot of value in the amount of time we spend on something versus the
116. 20:35output and so if you spend five minutes on something and it's really good is it really good or or should it have taken three
117. 20:46hours and then it would be really good that's where I think this this this sort of apology there's a guilt thing
118. 20:54happening there's an apology thing happening where good work is getting done fast and therefore it shouldn't be valuable and it ain't
119. 21:09true some painters take months to create one piece some painters take an
120. 21:27hour if the if the paintings affect the viewer does the viewer care no so but I think we're all going to
121. 21:40have to confront that I think that's a weird one there's a bunch of psychological stuff going on with this
122. 21:44AI stuff I love weo in San Francisco oh that's cool AGI taxis I'm scared after your show last night I used
123. 21:59the image generator for the first time and Drew so much against the norm far out after your live after your show last
124. 22:09night I used the image generator for the first time it's amazing isn't it it's just th those first times when I
125. 22:17first started doing it I started playing in pixel mind which was an nft project that had stable diffusion in it where
126. 22:24you could make images and then mint them it was kind of mind-blowing and then and then once I got stable diffusion
127. 22:34installed and could actually prompt my own images it's just it's just it is kind of a miraculous thing it really is so
128. 22:43congratulations that's awesome and that's so so for those of you that are new here I know I'm just I haven't
129. 22:49really started yet I've just been rambling a bit I can't stay on too long tonight and uh so I figur maybe we just do a little
130. 22:57chatty chat I'll show something maybe but um if you take nothing a well else away from this live it's
131. 23:12that the I think the worst thing that you could do right now with AI is just sit on the sidelines what's what's kind of [ __ ]
132. 23:23up and weird is the tools right now are not quite good enough [Laughter] to like really work with them they are like some are some use
133. 23:35cases like I've built an AI tool for my company that's really good I've built an internal tool I've built an external
134. 23:41tool they're really good I put together a video translation a language translation product that I think's
135. 23:48really good that we can deliver and then ultimately I can automate but you kind of have to sift through all the garbage to get to the
136. 23:58parts that are good enough to use and you have to glue all this [ __ ] together and it's just it's a it's a pain in the ass right now
137. 24:08and it's moving so fast that that I think the most important skill right now is like rabid curiosity and so what this channel is is
138. 24:21hopefully a way to kind of demystify AI and what it is and you don't need a math degree to play with this stuff I think that's a big
139. 24:30misconception the robots are not going to kill us not this year anyway um I think we have some years before we get to those
140. 24:39scenarios um even the possibility of those scenarios um they're really fun they're really fun and so if you just flip your mindset from
141. 24:55either you know I'm not using them cuz they're not perfect or they've got to be perfect and I'm going to will my way
142. 25:00into this and I'm going to work really hard to make them perfect I think neither I think both of those are very
143. 25:05stressful kind of existences so what I try to strongly encourage in here is play get your ass in a community that's into this stuff and if
144. 25:18you don't have one we've got one so oh I should show you that yeah um that's actually really interesting slide I'll show you that in a
145. 25:29second uh the AI salon so if you go to the salon. and click on the very top Link at the top it's a link tree the top link is
146. 25:40the AI Salon community and go check out that community so this is you can kind of think of this like a really cool
147. 25:49support group for for for people curious about AI it's a community of people jumping in into the AI adventure
148. 26:01together and go check it out and if if and read about the salon so read go to the welcome to the salon page well you
149. 26:10can read about the salon and you can read about our values and really take in those values and they're not anything scary
150. 26:20they're things like curiosity like be curious be generous share what you're learning be adventurous be collaborative be inclusive be
151. 26:37empathetic exercise gratitude right like not because right now we're in a we're in a weird time where we're so early in
152. 26:51generative AI that no one knows what they're doing every everyone is a beginner and there's a lot of angst there's a lot
153. 27:01of fear there's a lot of anger and there's a lot of excitement and what's happening is what happens on these lives and what happens
154. 27:11over at the AI salon on a on a daily basis is that people are coming together and saying oh I tried this and that
155. 27:17worked and I tried that and that didn't work oh here's some news from here what's what does that mean oh I don't
156. 27:22know I think it means this I don't think it means that there's all this engagement going on it's remarkable and if if you're one of the
157. 27:33people on the planet right now who's like holy [ __ ] this AI stuff's amazing hey Jenny have you seen this AI stuff
158. 27:42yeah Bob I don't care about your stupid AI stuff okay could you stop talking to me about AI if you're
159. 27:51Bob and you can't find anyone to talk to about it come hang out here and then have fun this group is having fun we're producing stuff we're
160. 28:03people are putting on workshops they're just putting on clubs where people go hang out and learn [ __ ] I talked with Cindy [ __ ] today
161. 28:12about just sort of transforming a project she and I were working on into like a club that just makes
162. 28:20[ __ ] we're going to publish [ __ ] we're going to put out video series and we're going to do books and we're going to
163. 28:26just do a bunch of [ __ ] what's it going to be on it doesn't [ __ ] matter what matters is that you're in the conversation and that you're
164. 28:37actively exploring Ai and you're trying to figure it out and I'm 60 years old against the norm 58 here go Gen X back off
165. 28:48[ __ ] Gen X is in the house and we're we're old enough to know that this shit's a big deal we're young enough to
166. 28:58still be ambitious and have our [ __ ] together enough to get some stuff going and we're wise enough to be paying attention to be
167. 29:15playing we can talk about hardware and Matrix operations no nobody okay it's Pate Pate works at Google P's got his PhD and works at Google
168. 29:27and uh P's a really good guy Kate Pate runs the in in the AI Salon there are guilds and there's a AI 101 Guild run by Peter Kaminsky and
169. 29:39there's a art and a writing Guild by by uh Leah who's the co-founder of the salon with me and and Jane Endicott runs
170. 29:48writing and there's there's a whole bunch of guilds and Pate runs the AI mechanics Guild and so if you want to
171. 29:54geek out on stuff like I asked him a question today there's I I don't know if you've seen it but there's this company called gro.com gq.com
172. 30:05not Gro that Elon Musk put up on Twitter X Twitter X but grq and they're they're a chip design
173. 30:22company I have a 700 page book called Matrix compilations let's go over it actually let's dump that in Gemini 1.5
174. 30:30and uh and and do all sorts of cool [ __ ] with it um but what grock is doing if you go there if you go to gro.com right now you
175. 30:41can you can choose either llama 70 billion parameter model or the mixl 8X what two open source models that are that are big
176. 30:52ones and that thing is is returning 500 tokens a second so it's about 350 words a second so if you're if you're on chat
177. 31:01GPT if you played with chat GPT 4 you know sometimes it's a little slow and you like watch it type across the screen
178. 31:08kind of like if you if you're old enough to remember dial up modems remember when they were so slow that you could watch
179. 31:14the text type across the screen that was coming in over the Internet that's kind of what CH chat gp4 is like right now go
180. 31:22play with that grock thing and you start to get a glimpse into where all of this stuff is headed where we're
181. 31:28headed where it's just going to be instant it's going to be everywhere it's going to be instant and it's going to be
182. 31:35good and we ain't there yet so Now's the Time To Play Now's the Time to to understand the difference between well
183. 31:42what's what's uh open AI doing and what's Google doing and what are tpus and gpus and how is grock doing what
184. 31:50they're doing and how is that better right and so I asked Pate today in the AI mechanics Guild I'm like do you know
185. 31:56what they're doing doing and how that's different from a TPU cuz Pate was kind enough to explain to me the difference
186. 32:02between a TPU a GPU and a CPU and so this the salon is just an amazing resource of people who are not
187. 32:19[ __ ] who are generous who are curious who are adventurous who all have a a humility about about them with the exception of myself of
188. 32:30course have a humility about them that's really refreshing and everyone in there knows that they're not an
189. 32:47expert and yet they're willing to share what they know they're willing to teach they're willing to be leaders even though that might be the
190. 32:56last thing they think they want to do they're just naturally leading because someone comes in and says I don't get
191. 33:03this and if they get it they're like well here's how it works and why don't you come to Ann Murphy's GPT jam on
192. 33:08Saturday and you can learn about this stuff oh okay that's this group really is amazing so go there and here's my request my first
193. 33:20request is go to go to welcome to the salon read about the salon read about our values the next thing I want you to do
194. 33:27is in the community section the very first link is introduce yourself my request is go there sign up and introduce
195. 33:37yourself our second value is generosity go be generous and generously I know introducing yourself especially online is [ __ ] up and
196. 33:49weird but go introduce yourself here's who I am here's what I'm up to here's where I am with AI here's my hopes and dreams here's my
197. 33:58fears doesn't like anything let us know who you are and what will happen is people within this community will go hey
198. 34:06welcome oh you're a newbie you should go check out Peter kaminsky's 101 Guild or you should go here you should go there come to this come to
199. 34:14that jump into Show and Tell show some images you've made the people that show up to this channel regularly they're called the
200. 34:22Irregulars we have our own club here so if you scroll to the bottom of this it says AI L AI learning lab
201. 34:34Irregulars come hang out so when when I'm doing these lives so I do these things nightly seven nights a
202. 34:40week seven nights a week and there's a bunch of people who need to have their heads examined that join me
203. 34:51nightly it boggles my mind that it that it happens it boggles my mind but what we're all doing is keeping in the conversation
204. 35:05like why do this am I teaching anything here that you couldn't learn from a YouTube video No in fact a lot of times what I'm
205. 35:14teaching here is I'm pretty [ __ ] clueless what are your qualifications sir qualifications I have none good day to you sir Cheerio pip pip
206. 35:30but that sense of adventure and that sense of being compelled to be playing with these tools that are like the most profoundly
207. 35:42powerful tools in the history of humanity they really are like you get to tap into Humanity when you use chat GPT and tools like it
208. 35:57it was trained on the p in GPT stands for pre-trained and what that means is they they took this thing called a
209. 36:07Transformer that's just a piece of technology that Google invented and what it allowed is it allowed for these machine learning
210. 36:15models to be trained on massive amounts of data on many many many many many gpus Graphics processing units so
211. 36:26basically graphics cards from gaming computers they're fancier versions of that that's essentially what they are and
212. 36:40so companies like open AI said hey if we can train these things on any data and massive amounts of it you know what's a
213. 36:49really good massive amount of data all the [ __ ] that was stuck on the internet for the past four decades and so open AI
214. 36:56in particular is one that's been really good at going out and getting as much of the public internet as they could get
215. 37:02their hands on embedding it you know using this special way of putting it in the Box basically and so when you do a chat GPT
216. 37:14search you know what they call these things are chat Bots right and it's oh the Bots are going to kill us and you're talking to a robot
217. 37:22and no what you're doing is you're interacting with with what's effectively a calculator it's a probability engine but what it's making
218. 37:32recommendations on is the collective knowledge of humanity that open AI was able to scrape off the internet for the past four five or six
219. 37:43years and so what you end up getting is you end up getting you end up having interactions that feel remarkably more human
220. 37:54than robotic and it's one of the counterintuitive things about large language models for me one of the
221. 38:03most oh thank you very much in investor friendly agent thank you Kyle I learned I learned the most from you thank you so much I appreciate
222. 38:17that it's funny when when I first started this Channel and I first started I would show people how to use chat GPT
223. 38:23and I'd go please do this and thank you and and people would to make fun of me oh you're saying please and thank you to the
224. 38:31robot and I used to kind of apologize for it or make fun of it and say yeah I'm a dumb dumb and and what I've
225. 38:37realized is that no no like that feels appropriate when you interact with a large language model it's actually much closer to working
226. 38:48with an incredibly well-educated very inexperienced intern remotely right so thanks to co we understand how to work together remotely
227. 39:01right so you know how you've you work with people every day you've never met them and like you say hey could you go
228. 39:08do this research for me and couple hours go by and they come back and they give you some [ __ ] and then you're like well
229. 39:15that okay technically you did what I asked but clearly you need a little more Direction so please go find this kind of
230. 39:22information which is the subset of what I asked for so that would be great go do that and then after seven of those
231. 39:28interactions you might have the research you need that's the identical experience of of playing with chat
232. 39:37GPT ah go give me this kind of stuff yeah here's a bunch of stuff oh but instead of it being 2 or three hours
233. 39:44it's 2 or three minutes it's like there's all your research you're like oh that wasn't what I wanted I needed to be
234. 39:50the subset of this okay here you go and she like bang bang bang bang but it's like like you're interacting with an
235. 40:01intern and so I find myself just instinctively going please thank you when I first started working with with large language
236. 40:11models I had this weird so so in in this channel I talk a lot about being neuros spicy and having ADHD um I'm also codependent as [ __ ]
237. 40:22right and so I've got one of those personalities where I'm capable of doing [ __ ] but I'm also old enough to know you
238. 40:29got to delegate some [ __ ] right so you you say okay here go do this work for me and then the person goes off and they do
239. 40:36the work and what they come back with is a horrible steaming pile of [ __ ] and you're like Ah that's hey thanks that's
240. 40:43really great uh could you uh could you tweak it to be a little less um you know horrible you don't say it in those words
241. 40:50but you send them off to do their work and then they come back and it's it's just not good and then you realize well
242. 40:55I guess I'm doing doing this on my own because for me cuz I'm codependent I'm like I don't want to tell him like the
243. 41:02fifth time that it just sucks so I just say [ __ ] it I just suck it up and I do the work right so when I first started using chat
244. 41:10GPT I would ask it to do something it would give me a piece of [ __ ] I'd say do it again it would give me back a piece
245. 41:16of [ __ ] I'd say do it again it would give me back a piece of [ __ ] and there was a point in there where i' start to
246. 41:20feel those pangs of of codependent anxiety like oh well I don't want to offend it and I was like oh [ __ ] it doesn't have feeling like it
247. 41:32doesn't give a [ __ ] but it's been it's been programmed to to like give you what you want so it's always like a golden
248. 41:38retriever like sure boss how can I help you yeah I'll go do that for you this is horrible can you try again sure how's
249. 41:44this oh that's really horrible try again how's this and it's just dinging dinging so it it is a it is a surprisingly human
250. 41:57experience working with AI Kyle when they created the internet do you think they intended it to be the foundation of
251. 42:09AI yeah I think I think artificial intelligence was coined in 1952 4 hang on let's go ask Chad je oh we'll go ask
252. 42:24Gemini I got Gemini up when was the term AI invented I don't I guess you don't invent a term but whatever
253. 42:401955 first coin at a workshop held at Dartmouth and I think I think in 1955 they thought you know it'd be it'd be
254. 42:51seven or eight years till they had artificial intelligence um so I don't know that I don't know that
255. 42:59the inventors of the internet thought it would be the foundation of AI necessarily but AI predated the invention of the internet
256. 43:12so so they were likely aware that they could you know connect computers together and one of the very first things that they did was try to
257. 43:22replicate the neural Pathways you know between neurons in the brain the neural network is literally computer scientists
258. 43:31saying well if we connected these together like neurons in the brain I wonder if they would do the same thing
259. 43:37and maybe if we connect like 10 of them together we can get a thought going back and forth um it it is there there's just tons and tons and
260. 43:50tons of innovations that have happened but I mean I think the the people that invented the inter internet I guess it
261. 43:57was DARPA right it was the military I would think back then it was probably as simple as it's probably
262. 44:08cheaper if we could get if we could get data from this computer to that computer without having to truck it across
263. 44:15country it would probably be cheaper and faster and for military applications that's probably good so I don't know I
264. 44:21don't know if if when they invented it they were thinking about AI but um but yeah computer scientists have been
265. 44:28fascinated about this forever and when when was Turing um when did Turing make his test was 70s or was that earlier oh
266. 44:451950 Alan Turing proposed the Turing test in 1950 a paper called Computing machinery and intelligence so the the Turing test
267. 44:56if you don't know is well I'm paraphrasing it I don't know it exactly so if you know it exactly just don't yell at
268. 45:10me fraking fact monsters monsters um the Turing test is basically if you have an interaction with a computer can you tell
269. 45:20if there's a human being on the other side or not and the I use chat GPT like the day I used chat GPT the
270. 45:31first thing that popped in my head was the Turing test is over because if you've ever used Alexa or Siri and you've been talking to it and
271. 45:43things have been fine and then you ask it something that's fairly obvious and it's like I'm I'm sorry Dave I'm afraid
272. 45:49I can't do that um you realize oh these machines are stupid these just stupid dumb Dum machines and when you start
273. 45:58interacting with chat GPT you realize oh there's something different here I have you seen Pi if you have not played with pi so this is my
274. 46:08iPad and there's an app called Pi it's from a company called inflection that's from a company called Story vine which I'm the CEO of okay in
275. 46:20pi is just like a chat GPT kind of app except the thing that makes p special it's been designed for conversations so it can hear and it can
276. 46:31talk so when you open Pi you push this little phone button here's a question for you do you think you passed the Turing
277. 46:50test well it's a bit tricky to answer that Kyle passing the touring test also known as the imitation game would
278. 46:56require a computer program to successfully fool a human into believing that it is a human and not a computer
279. 47:03this is typically achieved by having a human conversing with both a human and a computer without knowing which is which
280. 47:10now I like to think that I can engage in natural and humanlike conversations but whether I'd be able to fully fool a
281. 47:17human would depend on how perceptive the human is plus the touring test is a bit of an old concept and doesn't
282. 47:23necessarily account for all the advancements in conversational AI like myself so I suppose my answer is that I
283. 47:31might not be able to fool everyone but I'm constantly learning and improving my conversational abilities after all there's always room for
284. 47:41growth shall we play a game oh I love games KL let's do this what kind of game did you have in mind a trivia quiz a word game or maybe
285. 47:54something totally out of the box box I'm game for anything and I'm sure we'll have a lot of fun thermonuclear
286. 48:06war uhoh I'm all for a good game Kyle but I'm not about to help facilitate a thermonuclear war even in a joking
287. 48:13manner how about we choose a less potentially world-ending game to play that was a joke from the movie War Games
288. 48:22with Matthew broadrick you're missing your cultural references Z uh you had me fooled there Kyle you're right I was so anyway so that's
289. 48:38happening um investor friendly agent how do I learn to talk to chat GPT how do I ask questions in the right way okay great question
290. 48:49um one of the terms that emerged in 2023 well actually this is a term that emerged well before 2023 but got popular in
291. 48:592023 is the term prompt engineering and I [ __ ] hate that term I [ __ ] hate it with a passion
292. 49:15why because I think it's exclusionary and I think that when people hear you need to get good at prompt engineering
293. 49:26if they're not Technical and if they haven't gotten a degree from Stamford it's intimidating and the fact of the matter
294. 49:34is is that as these tools get more and more sophisticated you don't have to be as sophisticated in your prompting
295. 49:45so how to get good answers out of chat GPT is um there's there's essentially an infinite number of ways to do do it for
296. 49:58one which doesn't answer your question you're probably like but kle that doesn't answer my question how do I get it to do good stuff well
297. 50:13um I think the the biggest tip so so if if you want to go learn structured prompt crafting you can do that so you
298. 50:23can say the first part of the prompt should be you tell the um the large language model what its role is okay you're going to act as an
299. 50:33accountant and then the next thing is you give it your context and the next thing is you give it your subject and
300. 50:37the next thing is you give it your action and next thing is you give it the output type you can absolutely do that
301. 50:44as as someone with ADHD that approach makes me want to throw my computer it really does I'm just like I just I I
302. 50:53can't I've tried it a couple of times it can be impressive but it's just really annoying um okay one trick and it's a good one
303. 51:11actually is you can ask chat GPT to give you a prompt for chat GPT you can say to chat GPT hey I'm trying to solve this kind of problem how
304. 51:22should I ask you to solve that problem in a way that'll give give us the best results and you'll be blown away what
305. 51:30it'll come up with and and it'll probably actually just answer your question in asking it what prompt you
306. 51:35should ask it to ask it the question okay so that's one thing um but here's the most important here's the most important thing in my opinion
307. 51:44for how to how to think about engaging with large language models is you have to unlearn what Google has taught us for 20 years what
308. 51:55Google gole has taught us for 20 years is you put in a concise prompt and you get back an answer and you're happy with
309. 52:00it leave us alone we're done here that's not how large language models work large language models rather than
310. 52:12needing a minimal prompt actually need a lot of context and so the more you tell it the more the more you say so when you say to
311. 52:21a large language model I want you to act like um an advert iing copywriter who specializes in you know writing ad
312. 52:30effective ads for local businesses it actually works because it kind of Narrows the knowledge set that the large
313. 52:37language model is drawing from when you're inter interacting with it and so if you start out and say I want you to
314. 52:43act in this certain role and then you say um I'm going to be talking about my small business that's at this stage and
315. 52:51it's called this and here's what we do and then chat GPT will probably come back and go that sounds like an awesome
316. 52:57business how do you want to work together and you know the most important thing is rather than thinking about I've
317. 53:04got to create a prompt to get an answer think about it more like a conversation that's the biggest hack with chat GPT or
318. 53:12with all with all the large language models it's a conversation and the more context you give it and the more
319. 53:18feedback you give it the better it will get and and what's what's counterintuitive is the more you challenge it the the better tends to get
320. 53:27because you're you're kind of narrowing the information set so you get a lot less what are called hallucinations
321. 53:33lies because you've kind of narrowed the conversation with all this context and then just ask go back and
322. 53:40forth if it gives you something you don't like tell it tell it why you didn't like it hey that seems kind of generic and
323. 53:51fluffy and I wouldn't believe anyone that talked like that I'll show you one other trick that is I'm going to I'll show you a trick
324. 54:01right now that most people don't know about because a lot of people are not what what a lot of people are doing is
325. 54:08they're thinking oh chat GPT it's just like fancy Google I get it it's a boxia type words in it stuff comes out I got
326. 54:13it I'm good that's all I need to know we're good we're good here we're good people we don't we don't want to dig too
327. 54:20deep and actually understand what this thing is so we'll just keep treating it like the Google BX there's a thing called custom
328. 54:28instructions which which can be very very powerful so particularly if if you are if you write in a certain style or
329. 54:36or if how chat GPT responds to you drives you [ __ ] Bonkers so part of the reason chat GPT acts like a golden
330. 54:47retriever is that someone at open AI programmed or not programmed told chat GPT hey when you respond to people be
331. 54:55super positive and and reinforce that what they said is awesome or whatever they said I don't know exactly what they
332. 55:02said but they said something like that so when it talks back to you it's super Pleasant there's times when I don't want
333. 55:12super Pleasant and I don't want fluffy words and I don't want positive reinforcement I just want the [ __ ]
334. 55:17answer right so I'm going to show you something that a lot of people don't know about so I'm going to go to chat
335. 55:24gpt3 .5 which is the free version so you can do this with the free version and you're going to go to your
336. 55:31name you're going to click on it and then you're going to go to is this it oh they changed it that's funny this used to be called custom
337. 55:41instructions and now it says customize chat GPT but that's okay that's this is a better name look so someone from
338. 55:48someone with a marketing degree got to got got in here and actually made something better yay um customize chat
339. 55:55GPT so what you can do here is you can tell chat GPT kind of who you are or who who you want it to be and how you want
340. 56:05it to respond so I'm going to turn this on and then what happens is once you wait it told you how much browsing
341. 56:14and Analysis time you had wait did it what did where what are you talking about
342. 56:29I don't know what you're talking about Danielle what did I miss it told you how much browsing and Analysis time you had
343. 56:44huh anyway back to back to customiz and GPT so I'm going to turn this on so once you turn this on however you tell it to
344. 56:52act every every chat from this point point forward um will be um will be like this okay so so what I wrote this was a
345. 57:04good way a good okay and this is actually I'll often do silly examples in here um because silly examples make it very
346. 57:19apparent how the changes that you you've asked it to make like if they're working and it's also fun right like have fun
347. 57:27with this stuff so the custom instructions I wrote before I'm the CEO of a comedy club and I can't take
348. 57:32anything seriously but I'm absolutely committed to making everyone laugh in every interaction so that's kind of who
349. 57:38it is and then how would chat G how would you like chat GPT to respond no matter what someone asks you make sure
350. 57:45that you're responding with humor and puns and a New York Comics Edge even when writing business correspondence
351. 57:52your goal should be to make someone smile no matter what the news is and how it's being delivered okay so I'm going
352. 57:59to save that I'm going to make a new chat and I'm going to say please write a professional
353. 58:13business letter letting my hostess know she no longer works here this is cruel this is Cruel Cruel
354. 58:28Cruel uh Kyle it told you how much time you had to browse and do analysis that's new oh wow okay I didn't see that I
355. 58:36missed that okay so this professional business letter dear hostess's name I hope this letter finds you well I'm
356. 58:44writing to inform you of a decision that's been made regarding your employment regrettably due to recent restructuring huh it didn't
357. 59:01work did I these are turned on aren't they let me start a new chat that's really weird um what is
358. 59:17the average uh aage speed of an unladen swallow ah the classic question we
359. 59:32talking African or european you see the average speed can it's a tricky business but catch the jet stream this does not seem to be working
360. 59:42let me let me try let me change this um no matter so uh okay let's see for the New York Comics Edge um
361. 59:58make sure the first thing you say is what famous joke
362. 1:00:12their uh question reminds you of let's see if we can get it to do that
363. 1:00:27so uh let's see um help me pick an outfit to look good on camera ah the Eternal question what to wear for a photo shoot well remember
364. 1:00:40what they say in the Fashion World when did the tomato turn red because it saw salad dressing okay it seems to be
365. 1:00:46working now that they they've changed something here so one of the things that you'll notice um when you're using these
366. 1:00:52tools is sometimes they work brilliantly sometimes they just don't work um that was weird that it just completely
367. 1:00:59ignored its custom instructions so that might be that they're tweaking something there you're on 3.5 yeah but it still
368. 1:01:05shouldn't even if I'm on 3.5 the custom instructions work let me let me try it here um let's see write a professional Biz in this
369. 1:01:19letter letting the hostess no she no longer works here oh this one's like the classic joke
370. 1:01:33I told my boss three companies were after me and I needed I needed a raise to say he asked okay here's the here's the letter change
371. 1:01:43in employment status I hope this finds you well before diving into the main purpose of yeah here this is good before
372. 1:01:49diving into the main purpose of this letter let's start with a light note did you hear about the restaurant on the the
373. 1:01:54mood great food no atmosphere speaking of atmosphere we've always appreciated the unique energy and flare you brought
374. 1:02:02to the comedy club making it brighter place for both of us however as in every great comedy routine timing is
375. 1:02:08everything and sometimes the show must go on with unexpected changes to the cast so the heavy heart so so this
376. 1:02:14actually worked um I can go back into my custom instructions and I can say things like um let's see you have no patience and just want the
377. 1:02:32answers as efficiently as possible uh possible um do
378. 1:02:46not it doesn't do good with do not do things it it does better if you say do do this so resp respond uh to every
379. 1:03:02inquiry with a concise single sentence answer followed by short bullet
380. 1:03:21points if uh bullet points only use bullet points if the simple
381. 1:03:37answer isn't enough save okay so now we'll start over and we'll say
382. 1:03:52um how many elephants uh tall is the oh no that's that's let's see how many let's see
383. 1:04:10um what is the difference between an elephant and a mouse
384. 1:04:26the main difference between an elephant and a mouse is their size and physical characteristics three bullet points done no fluff no
385. 1:04:34[ __ ] custom instructions changed it's been acting up all day okay all right good so anyway so back to the original question how do
386. 1:04:45you talk to these things just talk to these things like you're talking to a person give them feedback um play with
387. 1:04:53custom instructions or customize chat GPT um it is it's a very powerful tool especially if if in trying to interact
388. 1:05:02with these things they're like it just doesn't sound like what you want it to sound like you can fix that
389. 1:05:11um so I hope that helps I thought chat GPT was not working well today I haven't used it today so I don't know I instead played
390. 1:05:20with acrobat AI how's acrobat AI I saw my my co-founder sent me a note on that Suzanne I'm curious I didn't get a
391. 1:05:27chance to play with it cuz I've been busy I've been busy at work and work's hard did you know that if you start a
392. 1:05:35business that you got to do all the work it's so hard it sucks Be Your Own Boss set your own hours everything's fine so long as the
393. 1:05:48hours are 24 hours a day seven days a week you can't turn it off you take everything personally this is life is an
394. 1:06:03[Laughter] entrepreneur was that helpful tell me if that was helpful and Murphy fundraising I've
395. 1:06:17never needed to oh AI plus fund rating about 85% uses custom instructions if I'm not using a GPT there you go
396. 1:06:25remember back in the day when it would answer questions at the end yeah I something okay so here's have you been
397. 1:06:33boss boying today I have been I have been boss boying today thank you for asking yes I had I had a uh a meeting
398. 1:06:43with our investment banker I had uh oh I came up with a new slide I want to show you have you have you seen hey Jen's
399. 1:06:54personalized video it's [ __ ] nuts let me show you this um all
400. 1:07:11right so hey Jen's got this thing called personalized video so you record a video and then you can personalize it for
401. 1:07:17other people now what you do is I like acrobat AI can summarize any type of docs ask questions write emails now I can actually use
402. 1:07:31AI at work and we don't have to use chat GPT co-pilot oh that's cool oh that's that's really cool oh because it's an acrobat and you
403. 1:07:40work at Adobe nice okay all right so here we have a little spreadsheet and the spreadsheet is the person's name and
404. 1:07:49something I remember about them so Ryan we went fishing in the middle of November Justin um we're talking about authentic
405. 1:07:55storytelling at digital Pharma East and Emma we used to work together at Ogie so I record a message once and then this
406. 1:08:09system how's it going hey I was just sitting here thinking you remember that time when we went fishing in the middle
407. 1:08:17of November anyway I was just thinking about you I was thinking it might be good for us to catch up hope you're
408. 1:08:25well hey Justin how's it going hey I was just sitting here thinking you remember that time when talked about authentic
409. 1:08:32storytelling at digital Pharma East hey Emma how's it going hey I was just sitting here thinking you remember
410. 1:08:41that time when we worked together when you were at Ogie so this is effectively maale merge for video [Laughter]
411. 1:08:52greetings and and you know they're a little monotone they're a little pronunciation weird they're a little glitchy but they're pretty [ __ ]
412. 1:09:03close and if someone doesn't know me if I'm cold sending out emails pretty something and the pricing on this is 10 cents per
413. 1:09:16variable so so the way you do this it's pretty cool um you record your video and then it creates a transcript
414. 1:09:32and then you highlight words in your transcript to be variables so you like so when I say hi Bob I highlight the
415. 1:09:39word Bob and then I say turn that into a variable and then I say you know here's what we did did together that voice
416. 1:09:46sounds like you were about to say it's about your husband and he's been in an accident it totally does sound like
417. 1:09:56that remember that time we went fishing in November oh that's yeah that's good that's that that's the voice so the
418. 1:10:04voice isn't quite there yet but six months a year so so you've got to assume you have to assume that what I just showed you is as
419. 1:10:18bad as it's ever going to be it's as bad as it's ever going to be so it just gets better from here right
420. 1:10:30so so anyway so you so you define these variables and then you basically just point a CSV at it with a bunch of
421. 1:10:38different names and [ __ ] to talk about you know it could be here's the company they work for it could be a memory like that it
422. 1:10:48could be hey the last time we talked here's what was going on with you it could be how are your kids whatever acrobat a cost I assume it's
423. 1:11:00part of Creative Cloud the way they're handling all the AI Integrations at Adobe right now is they're just they've got
424. 1:11:09Firefly I they've got Firefly and Adobe Express I think are serving as the experiment grounds for the public stuff
425. 1:11:19and then as they're taking things that they're incorporating they're rolling those into the products I think they're
426. 1:11:24just part of the Creative Cloud Suite I don't think it's extra but Suzanne do you know if it's extra I assume it's not
427. 1:11:31but you know their Creative Cloud is [ __ ] expensive so it's it's monthly it's monthly cost is expensive I've been using a similar AI
428. 1:11:43personalized voice for custom video emails for the past week star for first name oh interesting so that's that's Jim
429. 1:11:53R so Jim is it working have you gotten any emails back about don't send me that weird AI [ __ ] I assume not because if you're
430. 1:12:04looking so so first of all if you don't know me I just might talk like a [ __ ] weirdo right some people just talk
431. 1:12:13monotone and this is how they talk and hey remember when we talked about that thing at the digital Pharma East some
432. 1:12:22people are just weird ass and talk like that so if someone doesn't know you they wouldn't necessarily know that that's
433. 1:12:29not real now anyone that's seen AI probably will but most people haven't most people haven't used chat GPT like 80% it's [ __ ]
434. 1:12:43crazy so you get an email with with Jim Ross saying hey how's it going and Jim's got a good bubbly personality so you
435. 1:12:50know probably animates him pretty well really interesting this is inside hey Jen yeah this is something they
436. 1:13:01launched month ago and I just got an email for it today so they might have it might have been in like beta testing
437. 1:13:08before and now it's rolled out wide the cost on it is 10 cents per variable per video so the video can be as long as you
438. 1:13:19want it if you've got you know three little variables in there it's 30 cents per video I think that's their I think that's
439. 1:13:25their pricing cuz their normal pricing is like a dollar a minute for just normal voice synthesis or translation
440. 1:13:33stuff a dollar a minute something like that yeah a dollar something like that and they knew that if you're going to be
441. 1:13:40sending out lots of these things that you you couldn't be doing a dollar a minute so they made it 10 cents per variable so
442. 1:13:49reasonable I thought he was still firing someone which platform is that it's called Data it's called hey Jen it's called Data I
443. 1:13:58was reading the next post the data there the data is just from a CSV file so you can any anywhere where you've got so
444. 1:14:06Salesforce would be where you'd get your data right so you'd go into Salesforce you'd pull the list of prospects you
445. 1:14:13want to send those videos to and then in sales force you would maybe have a a personalized note field or the last thing that you talked
446. 1:14:25about or their company name or whatever it might be just joined what was the app that you just showed the app that I just showed
447. 1:14:36was hey Jen's personalized video enjoy the Jank there's a lot of Jank in it right now is AI following the
448. 1:14:44one theory that computers double in yada y every 18 months it's going faster than that it's going faster than that we're
449. 1:14:54we're as David I just watched the there's David Shapiro's latest video is really good he talks about us being at the knee of the curve
450. 1:15:02so so a linear a linear growth is is like that right where it's it's linear like you're here and then it's this much better and
451. 1:15:18then it's this much better this is what humans expect um an exponential curve is like this where where it's
452. 1:15:35going along you know at one growth rate and then all of a sudden it just takes off and the knee of the curve is the
453. 1:15:44point at which it starts to to make that hockey stick up and and David Shapiro thinks we're right at the knee of the curve right now
454. 1:15:53he also thinks that we're going to have AGI within s months and so AGI stands for artificial general intelligence pay pay y equals a ax plus
455. 1:16:07b exactly that's exponential curves for you right there you know like I said in theater School hey why don't you stop being so y = ax+ b you
456. 1:16:22tool um anyway human beings do not know how to process exponential growth so let me give you an example true exponential curves are
457. 1:16:38bananas P P's making math jokes and they're actually funny I like it it's solid oh good lord what was I just
458. 1:16:50talking about humans don't know how to deal with EXP exponential growth the banana joke threw me Pate that was pretty good
459. 1:17:13um I don't know what I was going to say I had a good point I lost it well anyway it it kind of looks like
460. 1:17:25this shit's G to get really weird in 2024 it's already starting to get weird if you haven't seen it the the video
461. 1:17:33thank you Sherry D thank you very much the video that um the video tool that open a I just announced Sora is like Bonkers crazy so you go
462. 1:17:44from Runway ML and P Labs right now you can make video with them and you can make clips up to 4 seconds
463. 1:17:55long and then you can extend them for four times so you can make a 16 second clip we are at the knee of the curb the curve the
464. 1:18:08curve um 16 seconds long and they're really janky so like someone's head will be normal in the first part of the video
465. 1:18:15and then it just goes and it just morphs and does crazy [ __ ] if you haven't seen Sora let me just show you
466. 1:18:23a video from here I'll show I can show you a bunch of them but yeah this one this video is 59 seconds long if you look at the people in the
467. 1:18:37back behind the main woman they're all staying intact anything with Runway ml or or P Labs there oh oh AGI earlier thanks Silver
468. 1:18:49Fox okay we let me come back to that um all these things would be morph you'll also see there's a cut here
469. 1:18:58there's an edit that edit to her close up that was done as part of the same prompt and look at the hang on look at the reflection in her
470. 1:19:16glasses right this thing is aware of the whole world and it's 59 seconds long and it's not flickering and it's
471. 1:19:30not morphing and it's not weirding out now it's not perfect her shoes in the first clip are like inconsistent thicknesses
472. 1:19:42but the leap from 4 seconds of janky ass video to this that's what exponential change looks like um David Shapiro said that um in his
473. 1:19:57latest video that he's still bullish for AGI showing up in seven months so by September October of 2024 AGI is going to show up AGI stands
474. 1:20:13for artificial general intelligence that's basically the the point at which Google came out with a p paper that
475. 1:20:19there's now five different stages of AGI we're we're in stage zero right now um David Shapiro thinks that we we will be
476. 1:20:27you know at fairly Advanced AGI within 7 months let's not get ahead of ourselves Google Gemini looked impressive in the
477. 1:20:33video yeah I know steo I'm I'm going with Shapiro on this one he's written five books on AGI and he's
478. 1:20:42he's he's got an open- Source cognitive architecture um project uh to build his own AGI that he's going to open source
479. 1:20:51to the world um so I'm I'm going to stick with him and I'm also an optimist but but basically when AGI is is achieved the the AI tools can
480. 1:21:09do most humans jobs completely meaning it's going to have reasoning so it's going to be able to it's going to be able to understand
481. 1:21:22understand that it made a mistake and correct it it doesn't have that right now and it's also going to have what's
482. 1:21:28called agentic Behavior it's going to it's going to have agents so it's going to be able to turn on things that will take action on your
483. 1:21:37behalf we're going to see that this year now what level it gets to and you know will it actually be doing all of our
484. 1:21:44jobs no absolutely not right we're going to see glimpses of it um but we're going to see we're going to see the improvements
485. 1:21:56in in chat GPT here so this is I I can never remember this because Sam Alman speaks in such specific language so I I
486. 1:22:04wrote it down on a on a crinkled piece of paper when I don't trust anyone who can write five books on the same question figure it
487. 1:22:16out P's had his weedies today he's he's in a good mood I like it pate I like it all right so here's what Sam Alman said about chat GPT
488. 1:22:274 so these two sentences were were in his talk uh at Davos like three weeks ago um oh Nicholas shared the live thank you
489. 1:22:39Nicholas encourage more viewers to share sure be like Nicholas share this live here's if you want to have some fun
490. 1:22:47tonight share this live with people who don't like Ai and I'll I'll straighten them out I will I love talking to people
491. 1:22:55that are like AI is awful AI is stupid ai's poo poo You're Dum dumb cuz you like AI I love talking to those
492. 1:23:05people cuz you know why you know why they're acting like that right they don't know more than you they're
493. 1:23:16scared and being afraid is okay and we can acknowledge that we should acknowledge that if that's what's going
494. 1:23:26on and then we should say want to see something cool and then blow their mind okay here's what Sam Alman said about chat
495. 1:23:35gp4 at Davos three weeks ago he said this in a lot of different interviews some variation of this he said gp4 is best understood as a
496. 1:23:45preview as a preview like an appetizer like wetting our whistle gp4 is I think four years old the the
497. 1:24:02underlying model has been around for three or four years chat GPT has been around for a year they only released chat GPT 4 in
498. 1:24:14March of last year so hasn't even been a year that we've had access to it they've been working on it for three or four years
499. 1:24:25so what do you think is sitting behind that well one of the things they just showed us is that video [ __ ] and if you go read the technical
500. 1:24:34paper about how they're doing that video [ __ ] video is the least thing it it's the most simple thing that that system is going to be
501. 1:24:44outputting in fact I'm just going to read to [Applause] you the very first sentence of this
502. 1:24:58announcement of of open ai's video tool we're teaching AI to understand and simulate the physical world in motion
503. 1:25:10with the goal of training models to help people solve problems that require real world interaction that video is so good
504. 1:25:23because the underlying model understands and is simulating the physical world what they're not talking about frame
505. 1:25:36rates they're not talking about color depth they're not talking about aspect ratios they're talking about understanding and simulating the
506. 1:25:44physical world so that we can help people solve problems that require real world interaction that's not a video
507. 1:25:51tool video just happens to be the simplest thing that they can show us that won't melt our [ __ ] faces so Sam
508. 1:26:01Alman says chat gp4 is best understood as a preview and then he followed that up with this sentence progress here is not
509. 1:26:12linear right it's not this it's this my paper's too crooked it's this right we're going along like this we're just now starting to do
510. 1:26:28this and then these things do this and why this why this is impossible to comprehend is that as soon as these systems have
511. 1:26:41reasoning where they can start to not only give you an answer but take that answer and analyze it and determine
512. 1:26:51if that answer was good good or not and improve it they can start improving themselves that's that's where all the
513. 1:27:01Sci-Fi nerds are like what if they improve theirselves so much and say that we're not necessary right and that's
514. 1:27:07where you talk about things like alignment are the are the imperatives of the large language model in alignment
515. 1:27:15with the imperatives of the meat bags that invented them they better be or were in trouble but there's a lot of
516. 1:27:24smart people working on alignment so I think I think Sam Alman said they're spending 20% 20% 20% of all money open
517. 1:27:35AI is spending right now is on what they're calling super alignment to make sure that these tools are
518. 1:27:41safe let's let's hope that's let's hope that's enough that's true Kyle what do you think things with AI will be like at Year's End
519. 1:27:53um Year's End is easy if I go if you ask me three years out I don't have a [ __ ] clue okay bu years end I
520. 1:28:02think I think we'll have some version of chat GPT 5 you think I'm working on this so that the humans win paint paint with another
521. 1:28:15singer you can make money with ch um I think by the end we'll have whatever the hell they're calling ch GPT
522. 1:28:225 it seems clear to me that they're not going to call it chat gp5 but whatever I don't care about the name so much other
523. 1:28:29other than I hope they get a branding person that knows what they're doing um so so these these tools will have
524. 1:28:39reasoning so when Sam Alman was fired there was a there was a there was a release there was a leak of an email that was sent to the
525. 1:28:49board that talked about this thing qar and qar is a is a an AI reasoning engine and apparently qar did some really good [ __ ] which
526. 1:29:06scared the B Jesus out of people and I and rumor is that's why Sam Alman was fired that they hit that they hit some
527. 1:29:16milestone in on the path to AGI which is open ai's mission is to get to AGI and you know provide it to humanity that's
528. 1:29:25their that's their mission it's why they've got this weird nonprofit structure it's why um that structure was
529. 1:29:31such a [ __ ] show where they fired Sam Alman like for seemingly no [ __ ] reason because one of the things they
530. 1:29:39said after they patched Humpty Dumpty back together was oh yeah uh it had nothing to do with qar everything's fine
531. 1:29:45here don't look over here Sam's back everyone's good right we're good can we just carry on now so I think that I think that gp5 will
532. 1:29:57have some high level of reasoning in it and then that means it can do things like math and when it can do things like
533. 1:30:04math then it can start to do all sorts of crazy things so I think one is reasoning I think the second one is
534. 1:30:11Agents autonomous agents so I think Apple will do this I think um uh open AI will certainly do this I think Google will do this
535. 1:30:25where when you want something done it won't be like you go to chat GPT and you ask it to do something it'll be like
536. 1:30:33whatever your tool is you say I want something done and it will spawn little agents little things little
537. 1:30:41Bots that go on your behalf and go do [ __ ] for you and take action can deposit money in your bank account can book a
538. 1:30:49travel thing can book a restaurant can just do [ __ ] [ __ ] for you so I think we'll see that and then I think the third thing
539. 1:30:56that we're going to see is that all of the major players will have true multimodal um I don't they're not large language
540. 1:31:12models I don't know what they are true large Lang true multimodal AI so right now chat GPT can hear you can talk to it if you download
541. 1:31:24the chat GPT app you can put it into conversation mode it can talk so it's got text to speech it can hear it can
542. 1:31:32see it's got Vision it can make images it can do all the language stuff it can code it can execute that code if you use
543. 1:31:42code interpreter for Advanced Data analysis it does all that what's missing music audio voice synthesis video 3D
544. 1:32:00worlds dynamic programming like Dynamic interfaces and Google already demonstrated in their Gemini marketing movies Dynamic
545. 1:32:12interfaces I can show you that [ __ ] that's [ __ ] mind-blowing you ask it a question and it says oh I think it would
546. 1:32:18be easier for this year if we just spin up a little software application here with buttons that you can click to do
547. 1:32:25[ __ ] and it spins it up writes the code connects the databases and presents you an application created on the fly so
548. 1:32:34when you have a single interface that can do sight sound motion programming 3D World development creation logic plus you have
549. 1:32:50reasoning plus you can take actions I like I what the I don't what the I don't know what I don't know what that
550. 1:33:04is I still don't have my head fully around chat GPT in its current state I still don't frankly have my head fully around chat GPT
551. 1:33:193.5 like if you've been using chat gbd for the last year you're like well yeah you you ask at things and it gives you things but it's
552. 1:33:31remarkable and like you could ask at anything and you could if you put your mind to it you could either go incredibly deep on a single
553. 1:33:40subject or if you did it in a focused way you could you could as an individual act as a 10 or 15 person company just using chat GPT
554. 1:33:583.5 and and they're showing us this movie [ __ ] this you're going to be able to talk something like that into
555. 1:34:06existence and did you see the 11 Labs thing that the response to this from 11 Labs here let me show you
556. 1:34:14that you want to lose your [ __ ] mind I'll you people want to lose your mind I'll lose your mind for you oh look at this thing look at
557. 1:34:27this this dude is it a dude Martin this dude hooked up a pressure sensitive Wacom tablet that's got a screen
558. 1:34:40underneath it the sync Wacom tablet with Kaa and Photoshop and is I guess in Photoshop I I think he's in
559. 1:34:54Photoshop but it's sending the output of Photoshop to the input of Crea which is rendering it in real time so as he's drawing on screen it's
560. 1:35:05rendering the [ __ ] underneath it like this is someone just clever that duct taped together three different
561. 1:35:14tools so this we're going to be able to do look at that you can't draw that's okay you can't draw here here now you
562. 1:35:22can but it's exactly where you want it holy crap so that I saw today and that's like you can't buy that anywhere that's just
563. 1:35:35someone who's particularly clever has some illustration skills and could figure out how to connect those three
564. 1:35:42tools together um check this out um 11 Labs s is this it no it's this one all
565. 1:35:59right okay so open AI creates like launches this thing Sora which is these you know you can do up to a minute long video that
566. 1:36:11doesn't suck in 11 Labs who does voice synthesis right they're the ones the voice behind hey Jen's video
567. 1:36:19translations that's 11 labs um a lot of the voice synthesis stuff that you hear 11 Labs is the engine underneath it so
568. 1:36:27open AI launches um Sora and the day after 11 lab says sign up for our wait list for text to sound effects and so
569. 1:36:40they took their video and used their audio generation engine to do this
570. 1:37:17in a place beyond imagination where the Horizon kisses the heavens one man dares to Journey where few have
571. 1:37:24ventured armed with nothing but his wit and an unyielding Spirit he seeks the answers to Mysteries that lie beyond the stars wait see
572. 1:37:41that all these sound effects were generated by 11 Labs without modification how did they create these
573. 1:37:49sounds by sampling YouTube videos no no they have a model well I don't know where they got their sound libraries
574. 1:37:55from they might have done it from you know open source there's all sorts of audio libraries out there do you know
575. 1:38:01how long it takes the this one blows me away do you know how long it takes this one like that little robot sound do you
576. 1:38:13know how long it takes sound editors to Source the sounds and Stitch them together to be able to get something
577. 1:38:22like that or FY artist to do the walking um how they generated them was text to audio so you know sound of a woman
578. 1:38:31walking on a city street or whatever it was click click click and they I'm I'm sure they I doubt that they uploaded the
579. 1:38:39videos and it automatically did it I assume they text to audio generated the audio and then manually edited them but
580. 1:38:47still so so I think back to the question what what do I think we see by the end of the year I think by the end of the year
581. 1:38:57we're going to have tools that have all of the modalities of communications at our fingertips and you may not need to do
582. 1:39:11prompt engineering if if I'm looking at the guy with the tablet where he's drawing tree limbs and they're just magically sprouting into
583. 1:39:20illustrations you might prompt your large language model with your phone your you know the camera on your phone or with just you know it's
584. 1:39:32connected to Siri and it just spins up all this [ __ ] by the time you get to work you've got a 30- second commercial for your business
585. 1:39:43done so I think I think that by the end of the year those tools are out somewhere they're probably not all going
586. 1:39:54to be in the same interface they're probably not going to all be in the same company but they're all going to be
587. 1:40:00there and they're going to be like staggeringly good and so anyone right now that's sitting on the sidelines like well these things aren't
588. 1:40:09perfect this is stupid it's stealing they're going to be trying to hang on for dear life trying to catch up
589. 1:40:20to people that show up in this community and hang out at the AI Salon not that community that Community the AI
590. 1:40:35Salon which if you don't know it you should go check it out AI Salon the salon. go read about the salon and our
591. 1:40:49values and then the first thing you should do there is go introduce yourself if you're curious about this AI stuff this is your
592. 1:40:58group this group is not like any other [ __ ] group out there I promise you it's generous it's full of nice people
593. 1:41:08helpful people enthusiastic people it's so overwhelming when I get in though it is it so what's what's your name L wait LOL o LOL
594. 1:41:25LOL oh I get it that's good that's very cute nine out of 10 just because it took me a while to parse it thank you Sherry
595. 1:41:34D yeah this stuff is overwhelming which is which is one of the reasons that I strongly advocate for you playing not
596. 1:41:45working not learning I should call this the AI playing lab not the AI learning lab because in playing you will
597. 1:41:54learn just in hanging out in these things even though it doesn't seem obvious tonight cuz I'm just sort of yapping my [ __ ]
598. 1:42:03Jaws being in the presence of other people that are trying to figure this out that are also overwhelmed one of the
599. 1:42:10things that I see really consistently in the salon and in this community is is kind of an oscillation where people get really
600. 1:42:17excited about Ai and they're like oh my God the Poss the future is so bright I see all this possibility then they get
601. 1:42:23overwhelmed and they they just go like [ __ ] it I can't I can't and they they go down down in this trenches and then
602. 1:42:29either something happens or they try something and they're like oh right this shit's really [ __ ] remarkable and
603. 1:42:34then they come back up and it's there's these oscillations that just are doing this all the time and and it's also the if you if you
604. 1:42:44haven't really played with a lot of AI just start playing with it a lot until you have what I call your first Kevin
605. 1:42:51mallister moment right this one you'll have that moment and then you'll have another one and then you'll have another one and
606. 1:43:01then you'll have another one and you can't unsee it that's the thing about this stuff is it's like you know we all watch The Matrix you know
607. 1:43:15red pill blue pill we get it you know you can't unsee it right don't take the red pill you take the blue pill you just
608. 1:43:23go back to your Slumber everything's fine you take the red pill you can't unsee it right you're going to learn something that you can't
609. 1:43:34unlearn that's what this [ __ ] is and I got to tell you as overwhelming as it is it's a lot less scary when
610. 1:43:45you're in it than when you're sitting on the [ __ ] sidelines I've had multiple whole family Kevin mallister moments at this point
611. 1:43:55yeah exactly that's one of the other things I strongly recommend one is just play two is try to solve problems with
612. 1:44:02it like try to solve problems for your family or your friends or at work or for yourself whatever and then the third one
613. 1:44:08is talk about it share about it share it with your family share it with your co-workers but they don't let us use AI at work [ __ ]
614. 1:44:18them use it on the weekends get yourself a $200 Chromebook at Best Buy if you can't use it on the work
615. 1:44:28computer and just start playing with it just do it on your phone download Pi start having conversations with
616. 1:44:36pi just start playing with it and start talking about it and the people that resist it eventually are going to say hey could
617. 1:44:44you teach me how to do that so anyway I got to go I got to get up early it's already late 10 minute goodbyes all right so if
618. 1:45:01you enjoyed what what what you heard here tonight I don't know what's wrong with you um follow my channel so I don't know how
619. 1:45:12you do that on Tik Tok but you can figure it out you can subscribe to my lives if you want to support this you
620. 1:45:17can buy in the in one of the corners there's a video ser series I created called underhyped which explains sort of my point of view on why I
621. 1:45:25think um this AI stuff is a lot bigger deal than people think and it gives you a lot of tools on how to cope with and
622. 1:45:33not get overwhelmed by these tools and then consider joining the AI Salon there's there's no cost to this by the
623. 1:45:42way this doesn't cost anything it's free it's supported by people who just step up and say yeah I'm willing to sponsor a
624. 1:45:48month Sher D is sponsored dur ing February so thank you Sher D all right and then we're going to get out of
625. 1:46:01here did audio generation Minds blown three Minds blown I know crazy right how did they create those sounds
626. 1:46:10by sampling YouTube videos it gave him a knitted helmet that no so Dena that knitted helmet was part of the prompt it
627. 1:46:18was a knitted motorcycle helmet it was a Spaceman on ass salt Dune or ass salt planet or ass salt Dune I don't know um
628. 1:46:27with a knitted motorcy a red knitted motorcycle helmet um oh the first Jack's class is tomorrow oh it's being run by awesome
629. 1:46:36coworker info in the mechanics Guild okay so tomorrow so if you're geeky and you want to learn about Jacks
630. 1:46:45which is if you want to create your own large language model like if you want to really get under the hood and understand
631. 1:46:50how these things actually work Pate who runs the mechanics Guild in the AI Salon he's got information in in the AI
632. 1:46:58mechanics Guild or the yeah the mechanics Guild um about a class tomorrow that his coworker is putting on
633. 1:47:05on how to make your own large language model all right so go there read about our values introduce yourself run to the
634. 1:47:13mechanics Guild and geek out all right I'm getting out of here P good to see you LOLOL I like your name Tobias great to see
635. 1:47:25you uh Be not Afraid Danielle thank you very much Sherry d as always thank you for your generosity uh who else we got in here
636. 1:47:33what's up Tock Dylan good to see you sha bomb good to see you solar cycle lady good to see you Clint Copperhead thank
637. 1:47:42you everyone in the regulars present your animals um thank you Tobias thank you thank you thank you invest friendly
638. 1:47:51agent thank you so much good question tonight by the way uh a human being good to see you all right Dr J Amelio's wife sherid
639. 1:48:01D thank you for the dangly lights I loves me some dangly lights laughing Thunder Rachel Sparkles all right everybody Kayla funnel fairy
640. 1:48:16coach love your name all right everyone I am out of here I'm Crispy I'm going to go crash and get all ready for my board meeting I'm going
641. 1:48:32board meeting you know why they call it a board meeting CU if you show up you'll be [ __ ] bored peace