aiLL 7/26/23 - The Transformative Power of AI

Transcript

1. 0:01you want to come inside come on Champion come inside no come on come inside lay down lay down you big dum dum
2. 0:19all right welcome welcome welcome
3. 0:42yo felonious you there how you doing it's happening ah wait hang on how do I get
4. 0:56yeah it was shaking hello so I got a an appointment for Friday they're there Thursday and Friday the last thing I sent you was a link
5. 1:07I think I can I still have it on my clipboard oh cool for the uh for the appointment yeah you can find out where they are
6. 1:18they're in different cities I can look I can't look while I'm on here I'll look when I go down and I'll let you know yeah just let me know
7. 1:31because I'll be in here so I'll be able to do it but if they come to Denver that'd be nice cool so what do you do
8. 1:42uh I am the director at the Salvation Army of a family homeless shelter and a drug rehab two different agencies on
9. 1:55the same property I get my steps in every day [Laughter] uh so 30 families in the shelter and a hundred well it's 40 now uh
10. 2:08addicts recovering from substance use it's good work I used to work I used to work in technology sales and a couple
11. 2:19years back decided to do something a little more uh fulfilling it's nice also going to also in grad school full-time for a therapist license
12. 2:31oh wow oh that's super cool yeah there's a guy in the AI Salon here in Denver that works for a non-profit as well and
13. 2:39is doing some cool stuff with AI so that's pretty cool yeah it's really helping in grad school it's helping with licensing issues at work um
14. 2:51and uh hopefully I can show my mom how to use it yeah it's awesome man I don't know how she's going to use it but
15. 3:00um I really appreciate your live I try to make it every day yeah I appreciate it man and I scuba dive oh nice nice I guess that's what the
16. 3:10picture is is that underwater somewhere nice that's in the Maldives in November on my site on my profile there's a video
17. 3:18I took in November of a whale shark the one thing I wanted to always see those are [ __ ] huge brilliant
18. 3:29brilliant so doctoral graceful and uh I was in the water with it in the middle of the night that's so cool thanks for what you do
19. 3:40hey yeah I appreciate it good talking to you likewise all right cheers all right welcome everybody welcome
20. 3:52welcome welcome this AI learning lab great chatting with you there Thelonious uh tell me tell me if they uh if they
21. 4:00zap your uh retina with the with the laser when you get armed you can be the guinea pig yeah apparently the uh the
22. 4:07world coin the line for getting the world coin the world ID I think they call it in New York City the lines are
23. 4:14incredibly long so if anyone's freaked out um they're not in New York at least at least the crypto Bros are lining up for it
24. 4:30this day I learning lab uh I'll talk about these URLs here in a bit um my name is Kyle Shannon I'm a long
25. 4:37time entrepreneur started a bunch of companies um I was an entrepreneur in the early days of the web and what's going on with generative AI
26. 4:45right now is incredible feels incredibly familiar to me it feels like ah we're doing one of these again um although it feels like this
27. 4:53generative AI stuff is coming a lot fast a lot faster and a lot more profound uh implications I think I think and uh
28. 5:02if you have uh questions comments about AI pop them down below and then let me uh flip the camera here [Music] um all right so
29. 5:18so if you're new to generative AI if you have not played with chat gbt that's a large part of what this channel is about is I just curious
30. 5:29whether you want it or not to give you some some context here there's there's seven sites here that are all different flavors of Chachi PT
31. 5:39um you know competitors you know different sites where you can use chat GPT um this used to be three three sites
32. 5:46long and as of a week ago it was seven sites long and um Elon musk's company's not on there but that'll be there at some point Apple's
33. 5:56not on there that'll be there at some point there's probably two or three others that are not on there that should be on there um
34. 6:04chat.openai.com that's the official chat gbt website if you haven't played with it go there and start playing with it
35. 6:10bing is also chat GPT that's Microsoft's uh search engine that has incorporated gpt4 as their base chat model and it's
36. 6:20connected to the internet and you can do image analysis and you can make images with it then you've got
37. 6:27Claude AI which is uh a competitor to chat GPT that's actually really good it's got a hundred thousand token context window you've got
38. 6:36bard.google.com that was laughably bad just a week ago and then they swapped out the model with this new multi-modal
39. 6:45beast called Gemini that seems like it's actually pretty good um and then labs.perplexity.ai that's metas you can
40. 6:54go play with the meta llama 2 models there and then Pi is super conversational and Poe gives you lots of
41. 7:01these things you can play with all in the same interface so that's all of those things then oh we don't stop there we we give value all over
42. 7:13[Laughter] um so so the top two sites there are getting started resources so prompts.chat will teach you how to talk
43. 7:23with those large language models openart.ai promptbook will teach you about prompting image generation things
44. 7:31it's written for stable diffusion but it's such good information you can really it really applies to all of these things
45. 7:38futurepedia.io is a big directory of AI sites then you got mid Journey Adobe Firefly and Leonardo slash stable diffusion uh which are all uh
46. 7:50really good uh image generation tools and then you got some things down below all right so pop any questions you have
47. 7:57down in the comments below uh do me a favor share the uh share the live if you would if you're willing to
48. 8:03let people know you're a weirdo and you like this AI [ __ ] and we'll we'll get some more people in here
49. 8:11and we're just getting started what's tonight Wednesday night middle of the week middle of the week midweek midweek
50. 8:18um the other thing if you're interested in joining um a group of like-minded AI adventurers um the AI Salon is something I started
51. 8:28back in December we meet bi-weekly we meet in person here in Denver but then have an online uh component and our next meeting is is
52. 8:38next Tuesday and we have a Discord server where you can play with mid-journey and ask questions and things like that so feel
53. 8:45free to check that out and storybinding.com that's my company I'm the founder and CEO of that so with that
54. 8:51let's get rolling let's get rolling rolling rolling rolling rolling rolling all right got on my Fender t-shirt
55. 9:04any guitar players in the house I had electric guitars for a while and then I realized that the electric guitar
56. 9:10is actually a different instrument than the acoustic guitar and I just wasn't as good at it but I just
57. 9:17didn't get it I didn't get it because I guess I learned on an acoustic and just never made the switch all right let's get to some questions
58. 9:25questions all right Miami New York is like not in Denver now checking yeah I'll do that we'll get there well we probably won't
59. 9:38so so do we want to verify any humans in that Cow Town the base of the mountain nah we're good [Laughter]
60. 9:49uh yep the the world app has got all that stuff I haven't tried Claude yet but I heard it's great yeah Heather the
61. 9:56the Claude what's what's really nice about Claude is because it's got a hundred thousand token context window you can either put
62. 10:04in really large documents or you can put in multiple documents or you can just have long conversations with it and it
63. 10:10doesn't lose its memory you know how you know how in chat GPT you're talking and all of a sudden it's like do you not
64. 10:16remember the beginning of this conversation and it's like nope I don't I'm I'm a little dumb dumb after about three thousand words
65. 10:26so claude's much better for that um a couple of places where it's it's not you know it's not up to Snuff with the
66. 10:33writing of gpt4 but for the most part it's pretty good uh what is image analysis how might we use it oh let's go play with that I'll
67. 10:42go I'll show you that it's not well it well there's a difference between image we can do image analysis using Code interpreter
68. 10:53um but both Bard and Bing are multimodal meaning they can read images so you could upload an image and have it
69. 11:02OCR it you could upload an image and have it tell you what's in it or write an ad for that image so let's go let's go try that
70. 11:12oh this is the new oh by the way there's a new um stability AI released sdxl 1.0 so they're they're new image model
71. 11:27[Music] um so that is an image generated this model is pretty good this is my my uh 70s muscle car in an abandoned
72. 11:41Factory that's indeed a 70s muscle car these things are always like these Weird Hybrids between like all the different
73. 11:48manufacturers it's funny that one's sort of a Camaro front end with kind of a javelin in the middle and then
74. 11:56sort of Monte Carlo lines it's weird anyway at least it's 70s looks pretty good um you can go play with this at clipdrop.co
75. 12:08there's also another site I forget what it was you can go play with it at but anyway let's actually let's grab one of these
76. 12:17we'll grab one that uh that one looks like a mess let's grab this one all right so we're gonna save this as muscle
77. 12:34and then we're gonna hop over to Bing and so so when you're Bing if you go to bing.com it's a search engine and if you
78. 12:43click on the chat button chat button it brings you to this which is this is basically chat gbt but the big flavor of it
79. 12:52um one nice thing if you haven't played with Bing is you've got more creative more balanced and more precise so this
80. 12:57is if you've ever played on the open AI playground this is the equivalent of temperature right so more creative would
81. 13:04be higher temperature more precisely lower temperature I have no idea what what actual settings are sitting beneath
82. 13:10those but we're not supposed to know that because we're just lowly end users uh which is probably a good thing but
83. 13:20down at the bottom here you know I could I one thing you can do on Bing if you didn't know you can make images so I can do um let's see um
84. 13:31chubby black and white mutt uh dancing we'll say cartoon
85. 13:48of chubby in uh let's see Ian Pixar Style all right and so it's gonna go off and make us an image although because I said
86. 14:00Pixar style it might go we're not allowed to do that oh no it looks like it's gonna do it one of the things about Bing's chat GPT
87. 14:08is the safety guard rails on it can be really [ __ ] annoying there's sometimes just a little too too big but they're just generated that
88. 14:18image but now check this out so I can say add an image I can upload from the device and let's go we'll go grab that green muscle car
89. 14:27image I just did this one we're gonna pop that in there and now we're going to say please describe the scene
90. 15:03uh let's see let me let me I'm going to do a new topic it got confused because I asked it to make an image and then I
91. 15:10asked it to describe the image but it didn't know that I was asking it to describe the uploaded image okay so I'll say please uh describe
92. 15:22what you see here [Music]
93. 15:39images vintage card an abandoned warehouse the car appears to be a 70s era muscle car with a black stripe on the hood and a
94. 15:47black vinyl roof the warehouse large Windows broken panes peeling paint on the walls the overall mood is desolate
95. 15:53and abandoned is there anything specific you'd like to know about the image please right me five different advertising
96. 16:12copy lines for this image for a Car Wash and detailing service period the ads
97. 16:27should be clever humorous and memorable I think oh did I not have this in creative mode I think I'm in
98. 16:41balanced mode but whatever we're not in creative mode actually it'll be interesting maybe we'll flip over to
99. 16:46creative mode and see if it's any better don't let your car end up like this one keep it clean with our Car Wash and
100. 16:52detailing services even this vintage beauty deserves a good car wash abandoned warehouse or not every car
101. 17:01deserves a good deserves to shine let us help you with that don't let the dirt and grime take over so let's let's go new topic
102. 17:10yeah that was more precise so let's do it we'll do it more creative now so we'll pop that image back in there [Music] right me
103. 17:28a set of five add copy lines for this image comma that reference the elements in the image but after a Car Wash and detailing
104. 17:45service period make the ad clever funny and memorable all right that's somewhere in the neighbor of what we said before let's
105. 17:56see what it does in more creative mode Let's see if Bing disappoints us and writes like crappy Microsoft copy or if it writes something
106. 18:06actually good here are some possible headlines don't let your car end up like this don't let your card up like this
107. 18:13Warehouse bring it to a thorough service your car deserves better than this that's not bad this is what happens when
108. 18:21you neglect your car okay I'm gonna say um these are all too long make them three to seven word ads that are incredibly Punchy and memorable
109. 18:40watch detail repeat don't be this guy that's pretty good save your car call US Car Care We Care Warehouse bad wash good
110. 18:47that's good Warehouse bad watch good and that one Warehouse badwash good I would actually that's okay all right Microsoft
111. 18:56doesn't suck so you could do that with images you could upload a PDF and have an OCR it or or tell you what's in it
112. 19:03potentially even pull data out of a chart or a graph potentially I don't know um but yeah it can see images and and
113. 19:12Google bard does the same thing let's go look at Bard let's do the same thing with bard so we're gonna go grab the old muscle
114. 19:18car there it is I'm gonna pop that up in Jolly Old Bard and let me go just I'll just go copy my uh uh um [Music]
115. 19:33copy so the exact same prompt this is now in barred so the first set that it did was those scalebard does
116. 19:48I was pleasantly surprised with something I did with Bard the other night on a live here are five ad copies our car wash is
117. 19:56so good it'll make your car look like it just drove out in a of an abandoned building we'll take your car from looking like a relic
118. 20:03to looking like a 69. that's actually really interesting it kind of it kind of got the era of the car there's kind of the cultural reference
119. 20:14there huh that's interesting our detailing service is so thorough we'll even find the keys that were lost in the trunk 10
120. 20:22years ago that's not bad don't let your car become a permanent resident of abandoned building our car wash is so good it'll make you want to
121. 20:31start a band name and name it the abandoned muscle cars that's actually good let's and let's do the same thing
122. 20:37we did with with Bing where we told it those are too long to give us short ones same thing [Music]
123. 20:52abandoned car wash it out dirt Will Blast it off make your car look like new again we'll find your keys your car
124. 21:00deserves a watch I you know that's not fat your car deserves a wash it's just sitting in a [ __ ] Factory
125. 21:10deserves to watch it's a little bit of Shame that's that's that's good I like it it's shame marketing [Laughter]
126. 21:19ah welcome everybody this is the AI learning lab we are learning about in that case we were learning about both
127. 21:26Bing and Bard can read read images um and see what's in them and that's there's some interesting stuff going on
128. 21:35there all right let's see you probably get asked this all the time but what's the best one for code generation I don't
129. 21:44I'm not a CO I'm not an engineer so um I would say probably the primary place to start is GitHub uh co-pilot right because it's for
130. 21:57coding it's it's it's basically AI pair programming but if you um haven't tried just you know pair program pair programming with gpt4
131. 22:10it's pretty insane um I'll show you two things that are that are kind of interesting there we go um so so this is regular old chat GPT
132. 22:29um I'm just gonna go to default gpt4 and I'll say um write me a python application that calculates fuel efficiency
133. 22:43of a car driving across country and stopping for gas every hour
134. 22:58and sure here's a simple Python program that'll do that and it writes the python code for you and then you can go off and
135. 23:07run the code and see if it works and and you can just go back and forth I mean the the other thing that that these
136. 23:13things are really good at they're good at reading code they're good at documenting code so maybe you have a
137. 23:18piece of code you can't find a bug in you can just dump it in there it will find the bug and fix it and probably
138. 23:25explain to you what's going on you can take a chunk of code have it explained to you what that code is you could do
139. 23:32that with trading Bots for example a lot of people come in here and ask me about trading which I know nothing about but it can do that
140. 23:41um and then there's there's this mode in chat GPT now called code interpreter and this is even more insane so so what we're going to do here
141. 23:51is we're going to go grab this is a ZIP file I don't know what's in it I do well I can't remember what it
142. 23:58was it was something we downloaded last night from kaggle so if you want to play with data so what code interpreter does
143. 24:04is you can upload a file to it I think up to 100 megabytes and then you just ask it [ __ ] about the file and it goes
144. 24:13and figures it out for you so we're just going to say what's what is this and why might it be interesting
145. 24:37this appears to be a zip file I would need to unzip it do you want me to do that yes you don't need to ask my permission to unzip a zip file
146. 24:48so now it pops up that sort of plug-in looking thing and if we open this what's it doing it actually wrote python code and
147. 24:57executed it it imported the zip file Library and wrote the code and unzipped the file and so now it says the zip file contains
148. 25:08a single file named backloggedgames.csv would you like me to load and preview this so we'll go yes and so again
149. 25:21it's opening pandas I think it's dumping that into a what do they call it a data frame or something within the virtual
150. 25:28machine so it can it now has all of this data in its memory basically so now it's analyzing what's in it
151. 25:42the data set might be interesting for a variety of reasons could be used to analyze Trends in the video game
152. 25:47industry to understand player behavior and preferences what would you like to do next with the data please visualize the data in the most
153. 26:01uh [Music] the most beautiful way possible and that captures what you think
154. 26:18are the most interesting trends in the data so this is the kind of stuff that you would ask a data analyst right this is
155. 26:28the kind of stuff that if you were tasked with finding out if your marketing was working or if your sales numbers why the the sales were not
156. 26:36working um you would either have to know how to do this write your own python or you would have to you know have someone at the
157. 26:44company that knows how to do this that studied data analytics and now just as a business person you can go
158. 26:51yeah just tell me what's going on here and so it writes the code it writes it creates charts those charts are not all
159. 26:59that interesting because the scale's all [ __ ] up um more uh I'll say please do a more in-depth analysis
160. 27:17so like I don't know [ __ ] about data analysis so I'm like I'm like a bad one to be demoing this but if you knew
161. 27:23anything at all about it I mean even without knowing anything about it you can see that this is pretty remarkable what's going on here
162. 27:30and so here so the let's start with the first so now it did that's an interesting table a correlation Matrix of of game attributes so so
163. 27:44you know um comparing the different game attributes let's see what else now let's visualize the relationship between rating and plays
164. 27:58and between reading and reviews let's so it's going to create a scatter plot so so there's your Scatter Plots between ratings and reviews
165. 28:10like I mean this is [ __ ] remarkable I mean right this is insanity it's insanity so what can you do with
166. 28:21this I don't know yet you know I asked someone asked before when I said Bing and and Bard can analyze images They
167. 28:27Don't Really analyze images but this can anal analyze images okay there's a Christopher walk-in that we made in
168. 28:35mid-journey the other night so let's open this up and let's say what okay find the top 15 average colors in this image comma
169. 28:50create a bar chart of the percentage of each color where the bar is that color comma
170. 29:06and the names of the colors are creative crayon colors so let's see what this is doing here so it pulled in
171. 29:24a couple of libraries web colors numpy it's still it's doing whatever it needs to do to average the colors
172. 29:34using k-means k-means clusters clusters equals 15. so I asked for 15 average colors let's see I apologize for the
173. 29:45inconvenience but appears that web colors Library is not available in this environment an alternative yes so do you want me to use an
174. 29:56alternative approach so check that out it knew what it wanted to do it knew it wanted to use a particular library and
175. 30:04then within code interpreter right now that library is not available so it said I can find another way to do it so this
176. 30:13says I apologize it seems like there was an error colors hex wasn't defined through the previous error let's correct
177. 30:19that by doing k-means clustering again and then convert RGB to hex and finally plot the dominant colors um
178. 30:31still not happy with it so now it's trying again so it's troubleshooting its own work it's like it's just [ __ ] crazy
179. 30:41see if we can get it to actually find the colors and right now code interpreter just uh uh it looks like it might have hung code
180. 30:54interpreter just does python but you know when when you hear people talk about autonomous agents or Auto GPT things
181. 31:02like that this is the kind of things it's doing where you give it a goal and then it goes off and does its thing uh let me try continue now
182. 31:15it's not it's still thinking it'll probably time out here in a minute but anyway isn't that wild I mean oh there we go
183. 31:28so there's the colors right so it it was it was you know a face so so there there was the those dark Moody colors
184. 31:38all right so so you get so so I'm gonna say um no wait that's great oh wait that's not what I wanted no
185. 31:54that's great you gave me the bar chart comma but I also wanted you to rename the colors in Creative crayon colors that are appropriate
186. 32:05for each color period and why don't you put it in a pie chart instead label each segment of the pie with the new creative color name
187. 32:26so now it's got to read the the hex values of the color and then somehow reference that to figure out is it a
188. 32:33flesh tone is it a green is it a blue or maybe it knows that just because of where it is and let's go let's go look
189. 32:39at them so there's our chat pie chart so Emperor's Emerald Sovereign Silver let's see princesses Periwinkle Princess
190. 32:51Peach Queens quartz monarchs maroon so this those seem Majestic mauve Regal Rose some of these are too dark to see Noble Navy Royal Red
191. 33:04so so you're taking some of the best of what chat GPT does you're taking this idea of autonomous goal goal-oriented you know problem solving
192. 33:16and you know please note these names are creative and do not necessarily reflect the exact color they represent they're intended to
193. 33:24be a touch of royalty so yeah so that's code interpreter crazy it's nuts all right you wanna see something else fun while we're here all
194. 33:35right so there's a new feature so if you've got in order to see all these these fancy new features in chat gbt you have
195. 33:43to have chat GPT plus then even if you have it you may not see them because you have to explicitly turn them on so you go to your settings
196. 33:50and then there's a beta feature section so you turn on beta features and then you actually have to enable them so turn them on
197. 33:57but one of them here is called custom instructions and if you've ever played on the openai playground on the playground
198. 34:06um you've probably seen system instructions this is very similar to that I think it's identical to that
199. 34:12so you've got about you and then what you want the the chat bot to respond with and so this is one I'll just turn
200. 34:21this back on from last night this one was pretty funny um let me read it to you uh I'm a master thespian and pompous ass
201. 34:29I have a large vocabulary and I Relish in sounding smarter and more dramatic than people around me no matter what
202. 34:36anyone asks so this is um how would you like chat GPT to respond no matter what anyone asks asks start out with a
203. 34:43pretentious statement that draws from the plot of a classic Broadway play follow up that declarative statement
204. 34:49with an article formatted using markdown complete with title subtitles and bulleted lists no matter how trivial the
205. 34:56question was that was asked um compare um the writing of the article to the
206. 35:12New Yorker whenever possible all right all right so there's our thing so now no matter what I do whether I do GPT 3.5 or
207. 35:244 or code interpreter and sometimes this is confusing if you forget to turn it off I'll go to code interpreter to
208. 35:30analyze some stuff and it'll just be the stupid pompous ass actor dude uh so I'll just type in here um let's see uh we'll do something
209. 35:41non-theater related so we'll do um why does fresh asphalt smells so good ah my inquisitive interlocutor
210. 35:58the Allure of a delightful Aroma that emanates from fresh asphalt can be likened to a poignant actual production of life in the
211. 36:09melodious Overture of this olfactory experience we find ourselves transported to a scene reminiscent of the opening
212. 36:16acts of the streets of New York a classic Broadway play and then it uses um markdown so notice it's got bold and
213. 36:24bullet points and things like that and then it wrote the article The New Yorker connection so while my example here is silly
214. 36:36um let's say you always wanted uh chat GPT to respond with comparative tables or you always wanted it to respond in
215. 36:44less wordy ways right right now Chachi PT is a little wordy asking for stuff it's like I'm so glad you asked Bob let's dig into that
216. 36:54just want the answer you could actually put that into these custom instructions so a solid interesting uh new feature that came out
217. 37:05three days ago two days ago something like that a lot of this stuff is new new new new welcome everyone if you're new my name
218. 37:12is Kyle Shannon this is the AI learning lab these sites here are all different flavors of Chachi BT if you have not
219. 37:19played with Chachi BT that's a requirement of this channel you can hang out here we hang out we cuss we have a
220. 37:25good time I drink iced coffee like why wouldn't you want to watch that old man on a tick tock old man on a tick
221. 37:33tock what'd you do Wednesday night well I don't want to talk about it um ask any questions you have below
222. 37:43here's the deal I'm 96 messages behind right now so if I'm not answering your question it's not because I'm not live
223. 37:49and it's not because I'm ignoring you it's just because I'm going to answer all the questions and it takes me a
224. 37:56while and that's the deal and if you don't like it then you should go you should go now you should leave what's
225. 38:01going on with the hair here's all right tonight looks like looks like speaking of Broadway it looks like some uh what am I in Greece
226. 38:15all right Izzy Lux hola hello is Adobe Firefly a separately buyable product I actually don't know the answer to that that's a
227. 38:28really good question can because it's only beta right now so I don't know can you subscribe to Adobe
228. 38:44Firefly on its own foreign you need to have a Creative Cloud subscription that includes Photoshop
229. 39:00nope you cannot sorry that's drag actually hey Tobias shaken that's a very animated wait what oh wait I oh tick tock tick tock show me the
230. 39:14bottom I don't like it take that losing cranky someone said something about linktree and don't ever trust anyone with link to
231. 39:24Troy Decker Kyle for president yeah that'd be good [Laughter] um oh man there's been a Cyber attack Kyle
232. 39:36what do we do I don't know [ __ ] deal with it maybe I drive my my Engineers crazy they're they're like well you know what's what's
233. 39:45the product requirement here I'm like product requirements actually quite simple I wanted to [ __ ] work laughs
234. 39:59ah it's comedy yeah we're gonna do comedy all night maybe I'll just talk like this all night no that even bugs me all right sorry I'm
235. 40:12jumping all the way to the bottom I'm trying to figure out where it was oh here we go okay Tobias hey oh you're you're very
236. 40:19animated and funny thank you welcome to movie phone all right where should newbies start newbies should start
237. 40:29well actually there but let me go to one other uh oh did I lose my yeah I did okay newbie should start there so if if you're just brand new
238. 40:41go to chat.openai.com I'm trying to think if there's any others I want to send you to there or bing.com
239. 40:50you can go to any of them but but chat chat.openingi.com that's the official chat EBT website they're the ones that
240. 40:57started this whole [ __ ] mess right ai's been around for decades people have been slowly inserting it
241. 41:04into this we've been getting our cars to drive with it then November 30th 2022 Sam Altman and team launched chat GPT and all this [ __ ]
242. 41:14hits the fan so that's what's going that's why this is all insane right now is because of that so go there that's
243. 41:20the original you know that's the original burger get the fries well done and then when you're there since you're brand new to this
244. 41:32go to this thing prompts.chat that's a web address so just point your web browser to prompts.chat and that's going
245. 41:39to take you to a thing called awesome chat GPT prompts or something like that and the first part of the document is
246. 41:47just a description of what these things are and how they work what these large language models are and how they work
247. 41:52and then the second half or the second like most of the document is all these prompts that will um tell chat jpt to act like a certain
248. 42:03Persona so act like a screenwriter act like an accountant act like a therapist act like an expert in history act like
249. 42:09whatever it is and what that document will do is give you a real good sense of how [ __ ] insanely deep these things
250. 42:21are so so when you go to them what you're going to see is just a it looks like a Google search bar but unlike Google
251. 42:29that's just indexed the web these things have been trained on all of the words on all of the public internet and it it has
252. 42:40mathematically distilled them into some sort of friggin miraculous you know what they call Vector space doesn't
253. 42:49really matter there's basically just pairs of words and meanings and when you type in your prompt it it answers you
254. 42:56directly and it can code it can write poems it can write songs it can answer questions it can analyze something you've wrote and
255. 43:05pull the outline out of it it can tell you where you've got grammatical errors it's just it so so here's the deal you owe it to
256. 43:15yourself if you're new to this stuff to play with these tools long enough for you to have what I call your Kevin
257. 43:20McAllister moment remember home alone this moment just play with them until you have that and then the fun starts
258. 43:34and then the next thing after that is like take something that you know really well and see if it can do that
259. 43:40like maybe you're a copywriter or maybe you're a project manager maybe you're an account person or maybe you're a salesperson
260. 43:48have it analyzed someone's LinkedIn profile just copy and paste their LinkedIn profile pop it in there and say how
261. 43:54would you talk to this guy if I were if you were a salesperson how would you approach them thanks crypto Joe
262. 44:00I liked me some ice cream I like the soft serve ice cream have you ever had the soft serve ice cream sauce serves
263. 44:06really tasty the the harder ice cream I like it but it's too hard and then all of a sudden it's melting all over me the
264. 44:13soft serve I can just lick it it's just delicious right from the beginning that's why I like the soft serve okay
265. 44:27one of the things that gives me joy in life is knowing that someone is flicking through Tick Tock just sees that the same person ah
266. 44:45all right so hopefully that helps what's your name uh Izzy Lux yeah that's where you should start start there
267. 44:52just start with the words words words words words Shakespeare once said words words words words language is everything
268. 45:01language is everything it's what gives us meaning and these are called large language models and they are
269. 45:18hmm I feel like Humanity reflectors so like they've been trained on all the [ __ ] we've put on the internet for the past
270. 45:27three and a half four decades and their ability amplifiers so if you're good at something these things
271. 45:34can make you really good at it if you're not good at something these things can make you good at it so instead of having
272. 45:40to put in your 10 000 hours the Malcolm Gladwell ten thousand hours thing to get good at a something
273. 45:46with this maybe you put in 100 hours and you're [ __ ] brilliant at it so if you have a hundred X Improvement in what you can execute
274. 46:00versus what you could do without this tool that's the kind of power these things are in in in San Francisco a lot of the uh
275. 46:11the tech shops that that have gone to full you know AI assisted pair programming for their Engineers they're
276. 46:17seeing 30 to 40 percent of productivity 30 to 40 percent and I bet if they get better at it it might go up and as these tools get
277. 46:28better it's going to go up the other thing about these tools is right now they're kind of off in the AI
278. 46:35ghetto right we've got our what we do for work over here and then there's the say I crap over here and we have ai
279. 46:42detectors we don't want anyone to use AI we're going to put AI watermarks in here and we're these things are all separate
280. 46:48but they're very very quickly being integrated into all the tools that we already use like for for example right
281. 46:54now I don't know if this is there for everyone but it showed up for me three days ago when I go to do a new post on
282. 46:59LinkedIn it now says would you like AI to help with that and so Microsoft is going to be rolling chat GPT into all of office 365.
283. 47:10um Google's going to be rolling it across all of their Suite um meta has open sourced their um large language model llama2 they open
284. 47:23sourced it with a commercial license this week so so the entire developer Community like all the GitHub and hugging face
285. 47:31repositories all the trending things are all llama 2 projects because it's now commercially licensable and it wasn't
286. 47:37the Llama one was was only for research so this stuff's coming and it's coming fast and it's deeply profoundly powerful
287. 47:46so uh yeah so that's why this Channel's here like I want people to get let's listen I don't care if you try this
288. 47:53stuff and then you go I [ __ ] hate it I'm never gonna use it again cool I got total respect for that love that
289. 48:00but if you're sitting on the sidelines and like yeah I heard it's not perfect I heard it's stealing from people
290. 48:10you're gonna get blindsided by it and and the the more you can get your head around it the better position you're
291. 48:18going to be in as these changes start to come um Tyron truth nice name I'm gonna go 8.5 out of 10 because it took me a while to
292. 48:30pronounce it but I like it um so hello Heather can any of these llms read actual uploaded documents or
293. 48:38does it have to be text no they can well hmm I don't know what you mean by uploaded documents like documents with images and
294. 48:47things in them I would try I don't think so I think right now they're all a little weird I would say the one that you have the
295. 49:01best shot at with that is code interpreter with chat GPT where it could probably write a python script where it
296. 49:08does some sort of OCR or some sort of image recognition on the images and then pulls out the text um but like Bing and Bard can both see
297. 49:19images and they and they can OCR images but if you've got a mix of text and images I don't think they can do that and then
298. 49:31code interpreters mostly like data Cloud can only see data and Poe I think po can only see data so I I don't know that's a really good
299. 49:45question that's that's worth an experiment and in fact Heather I mean I I think for anyone on the channel
300. 49:52right now I think a really interesting exercise even if even if you've been using these things for a while
301. 49:59these things are evolving so quickly I think a good exercise would be to take two or three different kinds of quotes
302. 50:07like maybe maybe you've got a creative writing quote and then you've got something where you make social media
303. 50:12content and then there's another one where you do I don't know project management stuff whatever take three
304. 50:18different quotes and run the exact same quote through all of these different tools and it's beautiful um
305. 50:28it's a it's an interesting experiment we did it last week with uh writing jokes someone asked can these large language
306. 50:36models write jokes and I'm like not really and then we we confirmed that but gpt4 actually wrote a joke that was a joke
307. 50:44you know would it have survived an open mic night no it was a joke the other ones were the other ones were pretty bad they were
308. 50:53joke structures but they weren't funny all right I asked Bing image to create a picture of a certain president sitting at a table
309. 51:04there you go I'll tell you the 2024 election is going to be somewhere between comedic and uh [ __ ] terrifying with all the um
310. 51:16misinformation and image stuff that's going to happen and video stuff I think Bing is using the bong yeah Bing
311. 51:27I want to like think more than I do right it doesn't have a good image generation tool behind it I I really
312. 51:34wish open AI would come out with Dolly three like something that's in the neighborhood of stable diffusion and
313. 51:39mid-journey because Dolly's just shitty right now and then Microsoft has Bing chat so safety handcuffed that it's just it's just bad
314. 51:50but it'll it'll get better I don't like I'm not taking the other thing about all of these tools don't feel like you have
315. 51:56to master one of them and you have to dive deep into it they're all changing so fast like they're evolving so fast that whatever
316. 52:06is required for you to get good work out of these today you're not gonna have to do that six months from now but
317. 52:14getting AI literate right now I think will pay dividends I think you I think we all have about a year that if we can
318. 52:20be AI literate and on this stuff and and getting our heads around it and talking about it and communicating it is like and then
319. 52:30like an unimaginable about amount of opportunity uh if you ignore it there's going to be an unimaginable unimaginable wait
320. 52:41I'm like Mitch McConnell did you see him he just stop talking today my brain just shut down unimaginable yes that
321. 52:49um amount of pain if you just ignore this stuff because I think it's going to transform nearly every position
322. 52:55certainly every position that is mediated with a computer so whether it's creative or whether it's knowledge work
323. 53:02or whether it's consulting or whether it's just even basic simple [ __ ] like Community organizing um everything's going to be impacted by
324. 53:11this stuff copy and paste the conversation so it remembers what you wrote you can do that Maga um up to a certain point so with um
325. 53:23chat GPT it's got about a 3000 word context window so if what you copy and paste is too long it will forget the [ __ ] at the top right
326. 53:33Claude has a hundred thousand token context windows so that's like 75 000 words so if you want a conversation with
327. 53:42one of these chat Bots where it doesn't forget what you talked about at the beginning of the conversation use Claude
328. 53:48or if you want to take the Great American novel that you wrote when you were 17 and actually rewrite it so that
329. 53:55it doesn't suck you could paste the whole thing in the cloud and tell it to help you rewrite it so claude's pretty [ __ ] amazing and
330. 54:03it's free and it's if you want to know who who's behind it the people that made gpt3 the the sort of thing that started it all
331. 54:12quit open Ai and went and started anthropic so it's a bunch of ex-open AI people that created it so it's it's an
332. 54:21interesting uh competitive this whole competitive this whole competitive scene right now is [ __ ] fascinating
333. 54:27yesterday Apple announced yeah coming it's gonna [ __ ] rock you know it's gonna rock so Apple's gonna be
334. 54:36added to this list and then elon's got once once he figures out how to get the Twitter logo off the building in San Francisco
335. 54:44and he renames that thing X and once he gets through the uh trademark issues with Microsoft Xbox and uh Mark
336. 54:52Zuckerberg with all of his trademarks for this thing x uh once he gets through all that and he gets his AI company up and
337. 55:00running that'll probably be on this list um I would not be at all surprised if um Amazon doesn't do something Snapchat's
338. 55:11got one which I haven't even played with Snapchat um yeah there's gonna be a bunch of them so anyway it's crazy times crazy times it's
339. 55:21so crazy it's a private did you see did you see is this image generator paid or free I don't know which one
340. 55:37um just like a cocktail [ __ ] dancing monkey you put a virtual hat on me I'm like hey we could we could turn this into one
341. 55:49of those channels where I just I just act like an NPC heart heart heart heart no all right fine fine
342. 56:05if you don't want me to make 700 a day reacting to cowboy hats that's fine we'll just talk about AI then fine I'll be broke I'll stay broke
343. 56:17um the image generator I was using well let me go see there was
344. 56:34so if you go to clipdrop.co that's stable diffusion and it's their it's their um it's like their showcase so they're
345. 56:46different things so I think clipdrop.co is free and you can use stable diffusion XL the other thing that I was using was
346. 56:55Bing's image generation tool that is free leonardo.ai is built on top of stable diffusion and it's not free but they give you like 150 50
347. 57:08credits a day so you can use it some amount of the day for free sorry I don't remember which one I was
348. 57:17using it was with too many messages ago I'm very far behind on the messages I don't know why people are still
349. 57:23impatient all right what was the AI that could generate PowerPoints tome tome.app t-o-m-e dot app but also go to futurepedia.io
350. 57:39and that's just a big old nasty [ __ ] list of like 4 000 AI tools there there I would bet I haven't been there because
351. 57:50I can't it makes my head hurt I would be surprised if there weren't a hundred 150 PowerPoint generating tools
352. 57:58out there so there's probably a lot of them here's the problem of the 150 you may find 145 are gonna suck really bad
353. 58:07and there's not going to be a good way for you to know what's the difference until you actually go use them so what I
354. 58:13would do personally and I mean you know maybe this is because I'm comfortable with PowerPoint I would use gpt4 I would
355. 58:20go to chat gbt gpt4 and tell it here's the presentation I want to write I just put in my my key you know key Concepts
356. 58:29and then have it [ __ ] write it and then just just copy and paste that into uh into a power pointer keynote outline I'm
357. 58:39going to Max I use keynote but we're switching oh my God I can't believe we're doing this we're switching my
358. 58:43company from Google to Microsoft it's been [ __ ] 11 years we've been on Google products we're switching it to
359. 58:51Microsoft which is I think it's stupid but they're rolling GPT across all of what they do we do a lot of work in Pharma so
360. 59:06like a lot of our farmers are blocking Google meet and uh I think a lot of our applications may need to run in on Azure service so
361. 59:18that's going to be painful ask me how I am a month from now because it's going to take us 30 days to make the transition oh
362. 59:29it's cool you're talking with each other I love that fantastic let me I'm going to check in on what you're talking to each other about here
363. 59:37let's see what do you mean scan text you told me to [ __ ] off it said if I persisted I'd lose my access what was
364. 59:45that what what service told you to [ __ ] off that's interesting no way for real okay good to know I'll need to know yes
365. 59:52I was shocked I should have screenshotted it yeah what what was the service that told you to [ __ ] off that's pretty interesting
366. 1:00:00was it it wasn't perplexity labs.perplexity because that's that's llama and it probably doesn't have any safety guard rails on it
367. 1:00:08those things will tell you to [ __ ] off all right all right I also use please when prompting the AI might end up being your boss of the
368. 1:00:20future it's funny like I don't like I don't do it I I don't think about it as you know our robot overlords I just find
369. 1:00:27myself it's really polite to me and I just find myself going oh thanks that was great like like my experience in working with these
370. 1:00:35feels very similar to working with the remote intern who needs a lot of action gets a lot of [ __ ] wrong but is [ __ ] brilliant
371. 1:00:44that's what these things are like and they they tend to work best when you give them lots of context you give them
372. 1:00:50lots of Direction you explain why you're doing things you say things like please and thank you and oh that's great you
373. 1:00:56know try it again but make it suck less I hadn't I had an acting teacher to do that once I was in an acting class
374. 1:01:05oh my God this is called this this woman gets up and she does her scene and the scene ends and he goes ah yes yes yes um
375. 1:01:19why don't we try that again and this time uh why don't you try it as if uh you had Talent it's like the whole point
376. 1:01:30it's a rough business the old acting business you know because all this fun and games hanging out at the Red Carpet oh no oh
377. 1:01:42let's try that again this time as if you had Talent what's the website address again for what I don't know which one I was on but
378. 1:01:53I'll tell you more Talent than Hollywood come here first right now can't find a sag actor I got a tick tocker for you [Laughter]
379. 1:02:06I'm SAG eligible I still get I get I did one film I did New Jack City in the 80s I still get checks from that from sag and after
380. 1:02:15Kyle for president I think that would be a very bad idea be entertaining Maybe I've been comparing Bard versus GPT and
381. 1:02:24GPT is a far better writer it is gpt4 is amazing there's nothing close to it um ask both to write a song and compare
382. 1:02:31yeah no but but listen Bard was laughably bad for months and then the new Gemini model that sits underneath it is
383. 1:02:42um it's also interesting looking at the the Llama models that that you can play with at perplexity they're not bad but
384. 1:02:50they haven't been fine-tuned like one of the reasons chat gbt is so good is they fine-tuned the [ __ ] out of it so you got
385. 1:02:56your your base large language model down here and then you got all of the sort of you know power prompting they've got
386. 1:03:02sitting up in the application layer um and then and then apparently gpt4 is like they instead of it being one
387. 1:03:09massive model they broke it into like 12 experts and then there's 120 layers so when you do a prompt it's doing all sorts of
388. 1:03:17you know segmentation of which tool it uses when and which expert it uses when so there there's whatever they're doing
389. 1:03:24with gpt4s is just of of a different Beast than what everyone else is doing right now hello from Danville California what's up Johnny Bean
390. 1:03:34let's shake it welcome welcome to the AI learning lab this is the lab you might go but Kyle kind of looks like
391. 1:03:41a shitty home office you'd be right that's fantastic Johnny [Laughter] I need to get out oh all right
392. 1:03:56it's gonna be fine it's gonna be good crypto Joe I like being the least for code that doesn't surprise me although um I hear that uh that uh
393. 1:04:08co-pilot for uh GitHub is pretty good right because that that was specifically tuned to do code right so it's Chachi PPT but but tuned
394. 1:04:21for coding don't trust anyone from linktree don't trust anyone from linktree is that because you went to my thing
395. 1:04:31there and there's a link tree or because you went to my salon and there's a link tree there I must be evil
396. 1:04:40he's evil he doesn't answer the question it's got a link tree don't click on anything he'll drain your
397. 1:04:48wallet he'll drain your wallet don't do it for the love of God don't click on the link tree don't click on the link tree
398. 1:04:59what was the last thing he said I don't know he said something about I'm going to click on the link tree and I said no no all right
399. 1:05:13glad we could help this is exactly what I needed to know I know link trees right exactly I don't know what you're talking about
400. 1:05:21Heather but glad whatever I said was what you needed to know oh my god wow again not sure what that's in reference
401. 1:05:28to possibly my hair oh my it's gone South the glue has broken it's it's it's taken on a life of its own this and you
402. 1:05:37know what are we gonna do We're not gonna do anything we're just gonna let it entertain itself all right I've tried
403. 1:05:43to use this to code macros uh with VBA and it hasn't really been able to correctly any tip I don't know what this
404. 1:05:51is do you mean chat gbt I would say um go to gpt4 um for for gbt4 is better than GPT 3.5 at coding
405. 1:06:05and don't go to Bing's gpt4 go to that one um I would also say try Bard because Bard is better at code than it is at words and it might be
406. 1:06:18better with macros um I don't know enough about VBA and macros in Excel to be able to help you I'm relatively clueless on that other
407. 1:06:27than I know that you can do them I know that if you know what you're doing you can do some pretty remarkable [ __ ] and I
408. 1:06:34know that when I've been stuck like in in Google Sheets trying to get it to do something and I don't know what the
409. 1:06:39formula is and my ADD will not allow me to look through a long list of formulas yeah since that from happening whatsoever
410. 1:06:49um it helps it's gotten me out of a bunch of jams um but if you're doing some more sophisticated stuff I would say just
411. 1:06:57it there's so many variables here so so if you're if you're new to prompting Mika or Micah Micah Mika Micah um uh Wyatt um
412. 1:07:14if you haven't done things like tell it to act like uh an Excel macro and VBA specialist and and kind of you know sort
413. 1:07:24of focus the conversation down and have it act like that kind of specialist that will actually make a difference because
414. 1:07:30the way the large language models work if you just give it a generic you know give me a macro that does this it's kind
415. 1:07:37of looking at all of what it knows out on the internet which includes the right stuff where people have solved that
416. 1:07:43correctly and all of the people who have not solved that correctly and put you know bad information out there so if you
417. 1:07:50tell it to act like a VBA specialist it will kind of narrow the the semantic you know ooze that it's that it's
418. 1:07:58searching in and that may help the other thing is um do what's called prompt stacking um where instead of trying to write one
419. 1:08:07perfect prompt and get it to give you the perfect answer you could say um I'm trying to write some macros uh with VBA
420. 1:08:18um is that something you understand yes that's something I understand great the area of vbas I want to do you know
421. 1:08:25centers around automating pivot table filtering I don't [ __ ] know whatever it is right and then and then you know
422. 1:08:32is that something you can help with and just keep layering the prompts so what you're doing is giving giving it
423. 1:08:37increasing thanks thanks Bear for the uh the Roses I appreciate that um it's giving an increase in context
424. 1:08:43where you're getting more and more specific so it should hallucinate us um the other thing to try if you've got access to um
425. 1:08:55to GPT plus is try code interpreter see if see if grab some VBA out of excel pop it into into a file throw that into code
426. 1:09:04interpreter and have it help you rewrite it there because it's right that's writing and executing python code so if
427. 1:09:12your VBA has to have any formulas in it maybe the the python would would execute that better than just chat gbt I don't know um
428. 1:09:22this is a all we're doing now is mansplaining which is mansplaining we're just saying words we don't know anything
429. 1:09:28we don't know any facts we don't know any series we don't know nothing but we're doing it confidently we're doing
430. 1:09:33you know what could help you Mika Micah Micah Wyatt Mike Mika Michael Michael Wyatt is you can do all this but I don't
431. 1:09:40know nothing so it's that's just mansplaining that's I'm a mansplainer that's a mansplaining does the AI man's planing lab
432. 1:09:56I sense that Bart is less emotionally intelligent yeah Chris you know why because it's all Engineers over there
433. 1:10:04everyone tries to figure out why is Apple so good because they understand the human side of things it ain't just about the engineering
434. 1:10:15Google GPT Ai and go to the one opening uh someone's helping someone else that's cool data Is My Jam if data is your jam
435. 1:10:23man code interpreter you should be all over code interpreter Heather all over it all over there me
436. 1:10:37excellent pedal crafts wow GPT writes children's stories and you can enjoy the stories there's actually a plug-in for a
437. 1:10:44chat GPT called stories that writes you a children's book and illustrates it the illustrations like the the prompt
438. 1:10:52work that they did on the illustrations is pretty shitty hopefully they'll make it better but there's a plug-in that does that too
439. 1:10:59it's yeah the the children's book thing is really cool insane I don't know what I was doing but I was probably doing
440. 1:11:05code interpreter and people are saying that's insane how fun to be able to generate a new bedtime story yeah
441. 1:11:10personalized bedtime stories so this is I just wrote a an article on LinkedIn called the uh the dawn the dawn of hyper personalization
442. 1:11:20um I think we're entering an era where you know your your child wants a story for the night you don't have to make it
443. 1:11:25up you just generate it and have it starring your kid and your kid could be a different animal every night you know
444. 1:11:34um and and very very soon we're going to be able to say I want to watch Top Gun but I want Jim Carrey you know as as uh the Cable Guy
445. 1:11:45starring in Top Gun and it'll just start playing like that that kind of [ __ ] is coming like rather than having Spotify
446. 1:11:52playlists you're gonna have a Spotify playlist where it's all original music mixed with your favorite music but the
447. 1:11:58the original music is music that is like knows what you like and every single song is like a hit for you foreign
448. 1:12:12is this chat gpt4 or a plug-in I think what I was showing was just regular chat gbd4 but then I did code interpreter code interpreter is
449. 1:12:24a plug-in it's just a really [ __ ] fancy one that's good um you prompt your idea of what is the story about it starts to write and makes
450. 1:12:34edits along the way um another thing to play with if if you want to if you want to do the the uh Marvel at this thing thanks Groove
451. 1:12:44attack uh Marvel at this thing writing narratives is go find a project called uh GPT author that implemented on a Google collab
452. 1:12:56notebook so you have to be a little geeky to to get it up and running but once it's up and running um it will write you a novel
453. 1:13:05you give it your open AI key and you give it your stable diffusion key and then you give it a topic you give it
454. 1:13:14a writing style and you give it a number of chapters and then it you just watch it it writes the outline and it says okay outline complete
455. 1:13:21writing chapter one writing chapter two writing chapter three and just goes through writes all your chapters then it goes to stable
456. 1:13:29diffusion and creates cover art for your book based on what it wrote and then it saves that as an ebook and you can open it in
457. 1:13:39any ebook reader um there's also a thing called showrunner AI that you give it a prompt and it will write you or it will create
458. 1:13:48for you an episode of South Park with voices and animations and music and story um and both of those are kind of crappy
459. 1:13:59right now like the the book that it writes is kind of crappy but it kind of just wrote a book for you like in five minutes
460. 1:14:07and you know the South Park thing is a similar sort of thing so don't look at those projects for what they are look at
461. 1:14:14them for what they represent right this idea of a truly hyper personalized future let's go let me show you another one that I I
462. 1:14:27think this one is just stupid fun so this is blockade labs this is called skybox.blockaidlabs.com so here's a
463. 1:14:42cool little 3D scape right but we can do let's see we can draw a door
464. 1:14:57Brown [Music] and we'll do a little window and we'll do a little window
465. 1:15:13and then we'll do um door leading out to The Stairway to Heaven
466. 1:15:28in a rock and roll universe and then I'm gonna go dreamlike and I'm gonna go generate
467. 1:15:43and now it's going to do that so it's going to make me some Doorway to a Stairway to Heaven we'll see if it got any of the reference
468. 1:15:55to the rock and roll universe is there gonna be a guitar here there better be a [ __ ] guitar here we go
469. 1:16:11there's the door there is no Stairway to Heaven this is a massive disappointment there's a stairway not to heaven where's
470. 1:16:18the rock and roll here oh what's this oh that's cool oh there's a little portal that's pretty cool but anyway so and we could you know we
471. 1:16:30could do like I could do Neo Tokyo at midnight in the rain and do I think there's a cyberpunk style
472. 1:16:47uh generate and so now it's going to go do that so now what this is doing this is actually kind of a simple thing you can't really
473. 1:16:59enter this world but someone has now figured out how you can hack this so you kind of combine it with Unreal Engine
474. 1:17:07and there's a way you can take images from this tool put them in Unreal Engine drop in 3D objects and then actually
475. 1:17:16start interacting in the world so we are not that far from the ability to that's pretty cool we're not that far from the ability to
476. 1:17:27say hey I want to play this kind of game in this kind of style with this kind of game mechanics and these kind of objects
477. 1:17:34and these kind of weapons and these kind of characters with these three friends and I want to play it right now
478. 1:17:41and we'll just start playing it and you know the initial versions of these games are going to be like you know the real
479. 1:17:48game developers are going to be like yeah but it's not Triple A it's not it's not a real game it's that's janky
480. 1:17:55gameplay and then a year after that the janky gameplay will be gone and then there's Tripoli game developers are either going to be you know
481. 1:18:04dealing with this reality or not and if not they're just gonna watch the demand for bespoke handcrafted games go down as
482. 1:18:13the demand for hyper personalized games goes up right because if I can play that kind of game that I like I love
483. 1:18:21um trigger Twitchy games and I like racing games I can't stand strategy like my 80d I can't do it so I got particular
484. 1:18:29things I like but I like certain Styles but they always have the gameplay that I don't like well what if I could do a
485. 1:18:35style that I like that's normally reserved for you know MMOs and [ __ ] like that and get all the adrenaline [ __ ] that I need
486. 1:18:48it you know instead of the uh the strategy stuff well I can I I will be able to do that right and that's
487. 1:18:56that's crazy that's just [ __ ] nuts it's nuts just cuckoo it's just cuckoo I don't know if anybody really
488. 1:19:07understands they're all like oh this is over hyped this is over hyped I'm a question for you what if it's under hyped hmm
489. 1:19:22what was the original picture oh I don't know what I was doing but if you remind me what I was doing I'll grab the original picture Wild
490. 1:19:33ah Magnum CI yes Thelonious Thelonius is the one that coined the term Magnum CI what we talked about last week was that the
491. 1:19:42product name code interpreter is [ __ ] horrible because it's way more magical than code interpreter um so Thelonious came up with the Gen X
492. 1:19:54friendly Magnum CI which I like it's solid but I don't think they're going to go for that so we gotta we gotta rename it
493. 1:20:04at some point we're going to do that dang it it would have been funny to name the colors with puns of his name cold please
494. 1:20:28please all right [Laughter] is there a way to share gpt4 with my partner like sharing Netflix um well
495. 1:20:40so so you have a couple of options if if it's like a cost thing like you don't want to pay the 20 bucks I I assume you could
496. 1:20:48I don't know if you can both be logged in at the same time but claude.ai is free Bing is free Bing is gpt4 it's chat ebt4 but it's it's kind
497. 1:20:59of Microsoft's version of it so it's a little bit hobbled but but it's it's you can do image generation there um Bart is free
498. 1:21:11perplexities free Pi is free pose got a bunch of free stuff in it so there's a bunch of free options here
499. 1:21:20um I know that you can share prompts from chat gbt with someone else but they have to have their own uh kind of thing but
500. 1:21:31yeah I don't know I I don't quite know what you're trying to do river but but there's enough free options here that that I think you have
501. 1:21:39some options without having to have two subscriptions and with the ability to share your prompt um I think
502. 1:21:47I think if you share a prompt your partner could bring a GPT 3.5 which is free and just use your prompt within GPT
503. 1:21:553.5 so that might help what filter do you use to have your face in a square while on a computer it's called dual camera
504. 1:22:07so when you go live there's just a button there for dual camera if you don't see that button that's because
505. 1:22:13Tick Tock took it away from you because you're a bad human being like I had dual camera for a while and then they took it
506. 1:22:19away for like a month because I was a bad boy it was bad it was very bad they took it away and then for no reason then give it back
507. 1:22:30why because they're [ __ ] Tick Tock I don't know they make their own rules they do what they do they banned me twice for seven days and
508. 1:22:37I have no idea what I said one time I think I was talking about bias and I think they just went oh if it's biased it's bad get out
509. 1:22:48I was trying to explain what was going on with bias and they didn't want to know they didn't want to know they didn't want to know
510. 1:22:54you know what they wanted to know nothing they were like get off stop you don't do lives for seven days get out get out now poof gone
511. 1:23:04this is fantastic thank you thank you the website to make images of cars yeah that was that was uh um I always forget the name of a
512. 1:23:19clipdrop.co clipdrop.co let me go to the beginning of it so if you go here there's like all sorts of Cool Tools
513. 1:23:33here so this is stability ai's little playground so like you can they got stable diffusion XL that's where I made the cars they've got
514. 1:23:43uncropped they've got clean up they've got stable doodle is really cool have you seen stable doodle yet look at this um
515. 1:24:00[Music] foreign
516. 1:24:23what the hell is he doing I don't know he does this every night why does he do this we don't know why do you watch him we don't know
517. 1:24:34all right so there's that and then I can put the prompt uh let's see um blonde blonde haired funny old man
518. 1:24:51generate so this site is clipdrop.com that's actually kind of cool those are good I like that one
519. 1:25:06I mean is it made or is it is is no one else like just [ __ ] like just [ __ ] gobsmacked daily that we we get to play with these tools
520. 1:25:22I mean this [ __ ] insanity I drew a ridiculous stupid wait how do I get out of here how do I get it how do I get back oh
521. 1:25:33back to grid I just I draw this stupid thing and it turns into that without trying
522. 1:25:48I mean I don't know when I hear people like well it's not perfect it's got seven fingers I'm like have you ever tried to draw
523. 1:26:05you [ __ ] idiot you ever tried to write a screenplay you know how [ __ ] hard it is have you ever tried to outline a
524. 1:26:14screenplay you know how [ __ ] hard that is right
525. 1:26:29and don't forget all right it's an important lesson we like to learn here over and over and over again uh I'm also in my journey with
526. 1:26:43generative AI I love the conviction in your Tick Tock lives thank you very much what's your name wait a beer uh AB here up banerhi
527. 1:26:57abhira banerhi cool name very hard for my American Mouth to get my tongue around but it's very cool uh um
528. 1:27:13paste all the comments into chat gbt and ask it to highlight the best questions can I do that how do I get to all the comments
529. 1:27:25anybody now go left go right oh wait how do I copy them all no hmm
530. 1:27:40oh this is wild I didn't know this was here how do I copy all the comments anybody swipe left to close
531. 1:27:47these are a hell of a lot easier to read if I if I swipe left look look you know it's like oh man learn to use tick tock
532. 1:28:03he was so cute when he discovered copy and paste oh my god oh what am I doing with my life what am I doing people what am I
533. 1:28:18doing with my life I don't know we're having fun we having fun if we're having fun then glad jingus let's use terms like by jingus
534. 1:28:28I don't know how to copy and paste the comments but but that's a really good idea actually so if I can figure out how
535. 1:28:33to copy well I'll tell you what dear text does tick tock not realize that some people might actually have fun with the hat
536. 1:28:46they give it to you for a second and a half come on how do I get GPT bus you go to gptbus.com McKenzie's a good dude I would go I I
537. 1:29:02support you supporting him GPT boss is a cool project and by the way um uh open ai's custom instructions kind of [ __ ] him because custom
538. 1:29:12instructions are kind of gbt because uh but now he created custom instructions oh I lost my place sorry I lost my place
539. 1:29:20in the comments when I tried to figure out how to copy questions I completely lost my place in the question so we'll
540. 1:29:26get back to it we'll get there eventually it's gonna be so good it's gonna be so good it's gonna be so worth the wait
541. 1:29:33what'd you do Wednesday night I don't know I think I watched some old guy try to learn how to use tick tock that sounds awful yeah
542. 1:29:43what are you doing tonight I think I'm gonna watch the same guy what's going on with you I don't know all right
543. 1:29:49I've still lost in my questions I'm just making fun of myself at this point all right Magnum CI we're back to that
544. 1:29:57the website make the images and cars we're back to that oh you're so funny oh my God yeah but it looks aren't everything [Laughter]
545. 1:30:10is the sound randomly static for anyone else Dr J I think that's probably me every time I switch cameras I think it [ __ ] up the Stream
546. 1:30:19and I switched the cameras a lot because you know I'm touched I'm a survival crafting gamer okay awesome 10 out of 10. text to
547. 1:30:30personalized game would be a dream come true I'm telling you man I used to I used to joke with my kids because I I
548. 1:30:36just could not for the life of me figure out why they played these strategy games and so I said hey I got an idea I came
549. 1:30:42up with an idea it's called an RT Triple G it's a real-time grass growing game and you plant seeds and then you watch it grow in real time
550. 1:30:52it's like they play like that that's the level of excitement they have for me RT Triple G baby gonna be a new genre
551. 1:31:07I I think watching this channel is probably like playing an RT Triple G what's VBA uh Visual Basic something it's basically
552. 1:31:18programmed into Excel to make it do fancy [ __ ] it might not be Visual Basic it's probably not Visual Basic virtual basic
553. 1:31:26I don't know it's something to do with Excel what you have ADD I never would guess that okay solid troll solid troll
554. 1:31:34it's kind of funny kind of mean kind of a compliment not sure if it's a real put down or like that's my favorite
555. 1:31:44kind of troll the one where you can't quite figure out if you just got slapped in the face or not but that's solid it's solid it's good
556. 1:31:52stressful week um it's actually been a good week it's actually been a good week um a lot of this uh one of the things I talk about in here
557. 1:32:05a lot is go learn to use this [ __ ] it's what it's why I'm a [ __ ] walking billboard for open AI um and I'm not getting paid for them
558. 1:32:16from them by the way but one of the reasons I talk about it is I actually think it's really important I think good things will come
559. 1:32:22of it I've had a very good week this week because I've been talking about this [ __ ] um not just on this channel but just in
560. 1:32:28general I've been talking about it on LinkedIn I've been putting myself out there I got the salon and people are
561. 1:32:34starting to reach out to me and it's leading to real opportunities for my business so it's been a good week
562. 1:32:40because it's the business environment right now is just [ __ ] weird anyone else experiencing this like like I just
563. 1:32:48call it bad corporate Behavior no one's returning [ __ ] emails they don't pick up their phones they don't show up to
564. 1:32:53meetings they don't show up to meetings they don't really apologize for it they're like oh yeah let's do something
565. 1:32:58and then three months later you're still trying to chase them down like it's I've never seen it like this we've in in 11
566. 1:33:04years of my business we've never had collections issues until this last year and it's not just small companies right
567. 1:33:10oh they're struggling no no this big Pharma companies just not paying their [ __ ] bills like what is going on that's some Gen X sass
568. 1:33:27[Laughter] he started out kind of calm and dim and then he just got raging for a while I don't know what was going on
569. 1:33:38I'm by the way I'm trying to find my place in the questions again this is this is this is what I do now oh
570. 1:33:48I might be thinking Kyle why don't you just skip some questions like I don't know have you ever been in my head no
571. 1:33:58I'm codependent okay I want to answer everyone's questions I want you all to be happy mmm
572. 1:34:13welcome to Moviefone is that iced coffee why yes it is remember Juan Valdez ah don't we all those are fantastic days in the 70s
573. 1:34:28when your coffee was tasty tasty and laced with arsenic oh yeah bring back the super fund hello Emilio's wife welcome welcome
574. 1:34:37welcome I know you've probably been here for half an hour I'm so far behind on questions but you know I'm trying to
575. 1:34:43figure out where I was in my questions I just saw you joined river is there a way to share ah here we go this was River's question about
576. 1:34:50sharing Chach EBT with her partner this is fantastic the website that makes images for cars now I'm too far back the beta was released
577. 1:34:59with my Creative Cloud screen subscription in the US best AI for iOS okay so for me so there there is an official chat GPT app for iOS
578. 1:35:15there is an official Claude is there an official Cloud app I can't remember I think Bing has an app or doesn't
579. 1:35:30um I use I just use Google Chrome or Safari with the chat GPT website I don't use the chat GPT app because
580. 1:35:44it's hobbled it's not as good and the site is so simple that it works as a website so I I would say the best one is that but
581. 1:35:52the one that I think is most interesting right now I wouldn't say it's the best but the one that's most interesting is
582. 1:35:57pi the PI app Pi dot AI or just the pi app on iOS it's just like chat gbt but it's way more conversational and it's got a mode
583. 1:36:09where you just sort of flip it into phone mode and then you just talk to it and it talks back there's six voices you can choose from
584. 1:36:15and like what I've been doing is I I pop it in my car I put it on Bluetooth and I just start having a conversation with
585. 1:36:22all of the history of the internet you know and it's connected to the internet so you could ask it like you could be
586. 1:36:29driving to a place and say you'd be driving to the mall and say hey what movies are playing at the mall and it'll
587. 1:36:34tell you and then you say what are the reviews like for that one oh here's what's going on with the reviews for
588. 1:36:38that one it's really interesting and it like for me pi kind of gives me a glimpse into what's coming so if you imagine pie without the delay
589. 1:36:49with a less annoying robotic voice and with an animated you know either version of you or stylized you or your dog or whatever
590. 1:37:01where you just have this persistent assistant oh someone trademarked that the persistent assistant huh
591. 1:37:10we're doing nothing but printing money in this channel like you take persistent assistant you turn that into an app no
592. 1:37:18one can pronounce it and you'll probably lose some money but anyway but seriously someone will remember it and then as much about
593. 1:37:27per month this is mrr recurring Revenue up to ten dollars per month with persistent assistance you're welcome
594. 1:37:37you heard it here first thank you thank you thank you all Bob good to see you all right but my colleagues in India
595. 1:37:49didn't get the beta on theirs yet oh that's interesting oh oh for Photoshop yeah well so Adobe right now
596. 1:37:58I could potentially get you an answer on that I know a guy at Adobe um they're really still in beta like
597. 1:38:06they're really still trying to figure out like how this stuff they're in a weird position as is Google
598. 1:38:13right how do you how do you how do you take something that in a lot of ways is a direct competitor to your existing product in Google's
599. 1:38:23case this stuff is a direct competitor in adobe's case it's tangential um but I I would assume that they're slow rolling it internationally
600. 1:38:32um is just my guess and that's probably something you could you could Google uh so yeah I don't have a clear answer
601. 1:38:40for you that on on that one Heather sorry about that um hey it's Mr Shannon Jeff will AI replace software developers
602. 1:38:47yes because I'm a CS student I'm worried about my future okay so Jeff here's the deal yes they will and
603. 1:38:59they're still going to be the need for engineers and programmers and computer scientists for years and years to come
604. 1:39:05but what you do and how you do it is likely going to shift dramatically so if what you're spending your time
605. 1:39:11doing right now is like learning python or learning C sharp or learning Swift or whatever it is whatever framework react
606. 1:39:18node.js whatever the [ __ ] you're learning and you're thinking like okay I'm gonna learn to code and I'm going to they're not ready
607. 1:39:29that shit's going away now what stays well critical thinking product design um understanding code logic how things
608. 1:39:42go together understanding things like um like like the the operation side of things like like all of these tools are cool but if I
609. 1:39:52want to do this Within a large organization how do we do stuff that's this cool but is also safe and private and has correct permissions
610. 1:40:03right so so the operations of AI is going to be critical um learning automation um you know whether you do it with code or
611. 1:40:14not learning things like Lang chain where you can um uh train these AI on on your own data or on company data learning things like
612. 1:40:23Auto GPT and baby AGI autonomous agents um playing around with code interpreter if you're not right now pair programming
613. 1:40:32with github's copilot you should be right because if you're a competent programmer getting out of school
614. 1:40:40you're going to have some level of value if you're a competent programmer that's AI literate getting out of school like
615. 1:40:46your value goes like this so keep learning what you're learning but later on this stuff and and if I were you like if I had the
616. 1:40:55chops and the and the the brain to be able to do more engineering stuff like I would be joining open source projects I'd be I'd be
617. 1:41:04participating in Lang chain I'd be participating in the autonomous agent things um I'd be go check out Mark Shapiro is
618. 1:41:13it Mark Shapiro yeah I think it's Mark Shapiro and YouTube who talks about AGI he's written five books on it he's got
619. 1:41:19three or four open source projects for that so I would just [ __ ] go all in learn everything you can be the most
620. 1:41:31and then learn whatever the [ __ ] they're teaching you at college and quite frankly the deeper you dig into this you may end up disrupting
621. 1:41:40yourself so rather than saying you know hey I just learned all this [ __ ] and you know they don't need it now you might figure out oh
622. 1:41:49I might start a company doing something because I'm really excited about this new [ __ ] I would lean in heavy with this
623. 1:41:54stuff if I were you because I think there's going to be massive demand but not a few ignore that stuff what's going on champion
624. 1:42:05you gotta pee what are you doing oh he does
625. 1:42:31[Music] come on computer come back come back come back come back is that iced coffee wow yes it is movie phone
626. 1:42:46yes cool stuff which does everyone prefer stable diffusion or mid-journey okay so here's my thoughts on this let me flip over to the slide
627. 1:43:00to that slide so I've got mid-journey Leonardo and Adobe fireflies as the three uh image generation things I put here
628. 1:43:11the only reason I put Leonardo there instead of stable diffusion is it's just a web front end in front of stable diffusion um for Pure image
629. 1:43:24amazingness mid-journey is the winner in my opinion but I think it's really good together and they just released Excel 1.0 today so
630. 1:43:38that's probably my guess is it's in the neighborhood of a mid-journey competitor um the thing about mid-journey is
631. 1:43:46because it's open source there is so much activity from the development community that if you're geeky and have
632. 1:43:53the the tolerance to keep up with all that stuff I I would go mid-journey because there's just so much going on in
633. 1:43:59the development Community right there's all sorts of Cool Tools there's control net there's automatic 1111 which is a
634. 1:44:05cool front end for it there's all these extensions there's Laura's which are all these different styles that you you know
635. 1:44:12you can sort of fine tune um stable diffusion to do pretty much anything you want it to do but it it requires some some
636. 1:44:23rather robust tolerance for the geekatube a lot of geeky Tunes with stable diffusion so and mid-journey you know mid Journey
637. 1:44:34so if you haven't played with mid-journey you you can only generate images right now in Discord and if
638. 1:44:41you're wondering well what the [ __ ] is Discord it's basically slack for gamers you know if you know slack for your
639. 1:44:48business it's kind of like slack but more annoying and um and so you have to kind of do command line things to generate your images but
640. 1:45:00the images are [ __ ] miraculous like they're just mid journey is amazing um so I think that's I think right now it's
641. 1:45:09just like Different Strokes for different different folks and and you know if you're if you're in a corporation if if you're in a commercial
642. 1:45:16environment the Adobe products are the way to go because they've been trained on public domain and Adobe stock images
643. 1:45:23they have the license for they're going to be more artist friendly where they're going to allow artists to opt out of
644. 1:45:29being trained opt into being trained attribution models um compensation to do it right so if you're in a
645. 1:45:38corporate environment I would say the Adobe products and then if you're if you're just out in the world and want to make part of pictures
646. 1:45:46then you know for me it's mid-journey because mid journey is just prettier and I like the pretty you know I like the Purdy
647. 1:45:54I like Purdy I don't mind the geeky but I like to Purdy and and you know they're close enough now that I would definitely
648. 1:46:05I'm I am definitely paying attention to stable diffusion because I think it's a really interesting company they're open
649. 1:46:11sourcing all of their Tech their CEO is an interesting dude I forget his name but he's super optimistic like he makes me look like a
650. 1:46:22pessimist he's just like she's like in a year you'll be you know describing the movie and then watching it you know it's like
651. 1:46:30yeah it's probably not coming that quick dude but but what they're doing is really interesting he entire ass thing
652. 1:46:36if I have thousands of hours of live stream video is there a way to feed an air model that data maybe I don't um
653. 1:46:46what do you want to do with it do you want to take the transcriptions and find interesting bits of it like I would
654. 1:46:53think that would be expensive but if you've got it on a platform is it on Twitch or is it on like where where did you live stream it
655. 1:47:01and is it on a platform that's that's already done the transcription for you because if you've got transcriptions
656. 1:47:07from all that then you can just run that through um through some of these tools you could maybe even make an automation on zapier
657. 1:47:14or something like that to just you know point to you pop a URL in there have it take that transcript and generate a whole bunch of things
658. 1:47:22from it um I built one of those today for my business um foreign
659. 1:47:41like Auto analysis of video but it ain't cheap and it's like if you've got you know if your live stream is two hours or
660. 1:47:49an hour like these I've been doing and they're um they're I've been doing these like on average they're probably two and a half hours long
661. 1:47:58and so there's there's some [ __ ] in here that's worth replaying but it's like I don't I don't necessarily want to spend a ton of
662. 1:48:06money going to find it all so I don't know Tick Tock does a thing where they try to identify some clips but they're they're
663. 1:48:13all over the place and they don't let you it's there's not an easy way to go in there and mess with them if someone
664. 1:48:19knows how to do that let me know because I don't I'm just like old man with Tick Tock I'm like here's three Clips can I
665. 1:48:25do anything with these doesn't look like it but maybe there's some magical [ __ ] pixel I gotta click that goes
666. 1:48:31into the actual editor I don't know uh can you have it perfect oh predict the lottery numbers well you could certainly have a pick them
667. 1:48:42uh that's the Kyle Shannon Comedy Hour my favorite time of the day oh Lauren I'm so sorry foreign here at the yeah learning lab
668. 1:48:55welcome to yuck yucks try the iced coffee tip your waitress uh AI is in Snapchat now yep I know it's
669. 1:49:09it's in that and it's in some other thing it's gonna be everywhere I'm telling you this stuff is I just I I still can't get my head
670. 1:49:18around the fact that there's so many major players right now and there's and there's some that are not on that list
671. 1:49:25there are AI skills wait there are AI skills you can enable in Alexa yeah oh in Alexa oh that's interesting
672. 1:49:36yeah I'm also curious if apple is going to um now Siri sucks less or are they gonna be like all right that was fun while it
673. 1:49:47lasted here's a new thing I think they have to bury Siri I think it's just I think it's got I think it's just not
674. 1:49:55I think it's got bad branding but I don't know Apple always surprises me uh elect the ability to become useful to
675. 1:50:04come to useful conclusions or I had to do all the thinking and feed it I don't know which one you're talking about
676. 1:50:10yeah the the I I mean the the art of prompting is you know in a lot of ways is more like a psychological exercise than an
677. 1:50:21engineering exercise um because you really do have to think in language and if you're not getting the answer you want
678. 1:50:28it's it's kind of like I actually I've never thought about this for as a metaphor prompting in a lot of ways is like an escape room
679. 1:50:37right or if you ever played the game Mist any if if I got got any uh gen xers any if I got any gen xers on the line
680. 1:50:45welcome welcome who played Mist it was on a shiny thing called A CD-ROM um prompting prompting these large language models is is
681. 1:50:57it's it's a similar cognitive exercise to escape rooms where it's just like like it gives you a shitty answer
682. 1:51:05and you're like I know the answer is in there somewhere and then and then you kind of go around this corner and you go around that corner
683. 1:51:15you're like I've looked at these drawers seven times is there anything different in any of these drawers oh
684. 1:51:21wait why is this one upside down click oh oh you know and then you figure out that layer of the escape room that that's
685. 1:51:29kind of what it feels like to do these so I wouldn't give up on it just yet I like I dig deeper and sort of peer
686. 1:51:36around the corners of whatever you're doing linguistically add more context sort of narrow the conversation before um
687. 1:51:48you know ask it things like do you understand you know what I'm asking you or if it gives you a shitty answer ask it
688. 1:52:00um to review the answer and tell you tell you how it would improve its own answer and that will sometimes give you
689. 1:52:07dramatically better answers that's the nerdiest thing I may have ever typed I don't know what you typed
690. 1:52:16oh wait if you have this chat to wait if you have this chat text to speech baby oh oh I see that's funny uh
691. 1:52:26Fender T-shirt oh yeah and the baby and uh dogs barking champ all right are there any safeguards to keep AI
692. 1:52:39from being used for for weird stick as in comedy uh there there are safety guard rails on
693. 1:52:55all of them except the that one labs.perplexity.ai sitting underneath that well so perplexity.ai that's the whole thing but
694. 1:53:08labs.perplexity.ai they're letting you play with meta's llama 2 models and those are open source and they don't
695. 1:53:17have any guard rails on them so people can do weird stick with them all they want um all the other ones have safety guard
696. 1:53:25rails and in fact they just made a big announcement Google open AI anthropic inflection and one or two others just agreed that when they all
697. 1:53:41make the next generation of these things here's the thing [ __ ] mind-blowing as gpt4 is like what's coming
698. 1:53:53is just gonna be like a whole other level so they've all agreed that they're gonna they're gonna collaborate on the safety
699. 1:54:02guard rails that they um that they put in place to keep these things from weird shtick is an interesting way to
700. 1:54:11put it you know this ain't exactly a Vaudeville act gone bad the AI learning lab is a Vaudeville act con pad hahaha
701. 1:54:20but but these things gone bad are going to be way more significant than that dogs back all right Kyle would you give
702. 1:54:28us daily homework assignments um would you create five project teams maybe ten possible um I don't know I mean I did a I did a thing on my
703. 1:54:44on The Tick Tock Channel called um the 2023 smart challenge and the 2023 smart challenge was this shit's coming
704. 1:54:55whether you want it to or not and so the challenge was this there were three three components to the challenge the first component was
705. 1:55:06um do two projects using a large language model so chat GPT or or one of these and one of those projects should be a
706. 1:55:15personal project and the other one be should be for someone else so personal project could be create a meal plan create an exercise routine
707. 1:55:27start the Great American novel you've been procrastinating for 27 years um I don't know teach yourself Raspberry Pi it doesn't [ __ ] matter
708. 1:55:39use these things to do a personal project come up with something for yourself and use these things to do it in the second half of that first
709. 1:55:49is do that for someone else so for your for your business for your family for a community group maybe your community center I don't know
710. 1:56:04needs to needs to come up with ideas for um programming for the next year and everyone's uninspired and burned out and
711. 1:56:12it's been 100 degrees for two weeks straight and no one no one's into it go into the community group say hey I'm
712. 1:56:19teaching myself AI would you mind if I came up with programs using AI for the group and they'll be like no okay whatever
713. 1:56:27I heard it's evil yeah I know but I'm just it's an experiment do you mind if I experiment and do something for them
714. 1:56:34in doing that thing for yourself and for someone else you will learn more about what these things can do than than any
715. 1:56:41amount of books or prompt lessons or any of that stuff if you come at this stuff from the outside like start with the
716. 1:56:46technology I've got to learn prompt engineering it's it's really foreboding and like complicated and it's hard to
717. 1:56:54see what's possible but the minute you go oh they need to come up with interesting activities for six different
718. 1:57:01age groups let me go try that could you do that oh [ __ ] I don't know I could do that could you organize that in a table sure
719. 1:57:15oh damn hour later you walk back in with a plan for 2024 activities for your community group so that does a couple of things
720. 1:57:27you'll learn how to use these better you'll learn what's possible you'll help someone else and then if you do it even remotely well
721. 1:57:39they're going to be like how did you do that so fast well let me show you and then you you have your own little
722. 1:57:47mini version of what I'm doing here so that's that's um task one of the 2023 smart challenge two projects one personal one for
723. 1:57:59someone else yeah I like it welcome Barbie um the second one was make 10 images using an image generation tool like Bing you
724. 1:58:12can make images on Bing or you could do mid Journey if you want to learn how to use Discord or you could do Adobe Firefly if you've got a
725. 1:58:21Photoshop or generative fill or you could do Leonardo make 10 images why because until you've had the experience where
726. 1:58:31you type in words and it [ __ ] miraculously generates something that you didn't think it could possibly know and it does it well
727. 1:58:40you just you you haven't quite experienced what these things make possible right because with the words you're kind of like well
728. 1:58:50I've been using Google for 30 years and 20 years and and I know like these things are just like Google but fancier but when you make images
729. 1:59:01where the thing that comes back is just in a whole other class of anything you could have imagined was possible just
730. 1:59:09replace image generation with every other media type music film um programming architecture interior design product
731. 1:59:19design like anything anything is coming so so the second half of it is to just get you in touch with the
732. 1:59:28multimodal thing right there's words and then there's the rest of the [ __ ] we deal with in life these things are gonna
733. 1:59:34be remarkable and transformative in all of that so that's the second thing of the 2023 smart challenge the third
734. 1:59:42component of the 2023 smart challenge is [ __ ] tell people about it tell people what you're learning be generous didn't she
735. 1:59:56like if you're like well I'm not Technical and I'm not really good at this doesn't [ __ ] matter you have friends over for dinner
736. 2:00:05so I was uh anyone checked out that Chat gbt thing yet oh I heard that stupid oh no I don't think it's stupid here's
737. 2:00:13what I did with it oh that's kind of cool have that conversation talk about it at work talk about it with your friends why
738. 2:00:23because it's a cult no because this shit's coming and like once you have your Kevin McAllister moment once you have this oh
739. 2:00:31you'll get it you'll you'll get that the possibilities here the opportunity here is [ __ ] massive but not if you're on the sidelines and
740. 2:00:39so talk to people about it because what's going to happen is one of those people at some point down the line
741. 2:00:46is gonna go I need to figure this out for my business oh Sally was talking about that at dinner the other night Sally yeah that AI stuff
742. 2:00:55you were talking about can you come teach me about that boom there's a job or a project or some interesting new thing in your life
743. 2:01:08so that would be the assignment that I think I would give is I would just go back to that thing I think I launched that in January
744. 2:01:18two projects one personal one for other people ten images and talk about it I'm surprised 30 bucks extra from
745. 2:01:29Microsoft I'm not because this stuff's expensive and it's the it's a business audience and it's a captive business
746. 2:01:36audience so they're gonna say well um the the the co-pilot the chat GPT that we build into Office 365 and that we
747. 2:01:47build into Azure and that we build into all this other [ __ ] it's going to be private it's going to be secure it's going to be out and
748. 2:01:57It's Gonna Make You three to forty percent more productive as an organization as you incorporate these tools minimum
749. 2:02:05within your organization so 30 bucks a month on a whatever it is 250 or 300 a month per seat license for all the other
750. 2:02:14[ __ ] you're paying for is is within reason for business so um and that's that's the thing that Google doesn't have right now right
751. 2:02:25Google's relying entirely on Advertising Revenue and you know they've given away their office suite they haven't charged for it and
752. 2:02:34and listen it was brilliant it was brilliant for a while because they took all the market share away from office but now you know now
753. 2:02:46Microsoft has this cool [ __ ] they're gonna charge 30 bucks for it and you know Google can't just start charging
754. 2:02:53for [ __ ] they've never charged for so I don't know it's it's fascinating this is fascinating like business competitive landscape times
755. 2:03:06it's wild it's just wild Claude wouldn't edit the style of Jane Austen but Chachi PT did a great job yeah I know gpt4 is just [ __ ] better
756. 2:03:16it's just better but Claude like here's here's how I would probably use Claude and gpt4 together if I I would take long documents or
757. 2:03:26disparate documents I'd put them into Claude I'd use Claude to organize them and to and to block out maybe the
758. 2:03:33outline or the structure of what I want to write you know sort of pulling in the raw materials that are in an unorganized
759. 2:03:41fashion I'd organize them in Claude I'd then take that synthesized [ __ ] I didn't Claude put that into gpt4
760. 2:03:49and then actually start working with it that that would that would be how I would handle that because because there's the the the
761. 2:03:59context limit that GPT is really frustrating and I haven't hit it yet with Claude some of them are making a hundred dollars a minute
762. 2:04:09oh some of the NPC people Emilio's wife well start putting cowboy hats on me I'll go cowboy hat cowboy hat cowboy hat
763. 2:04:20tutu pink tutu totally lights twirly lights I can do this could be an NPC come on um [Applause]
764. 2:04:37I guess I gotta do something about the hair like an NPC wouldn't have hair like that uh good Lord
765. 2:04:52complimenting the model and boosting ah it's a ticket attack yeah well dang it it took my hat away it took my head away they gave me a hat
766. 2:05:01they took it away oh any hats around here I just I'm just I'm just remembering back to that troll comment before you
767. 2:05:10add what a surprise what [ __ ] looking for a hat all right uh [Music]
768. 2:05:25uh neither Bard chat GPT will generate a woman in a bikini Bard or chat gbt neither of those do images do you mean Bing and
769. 2:05:39well she's got a stable diffusion you can see all the women in bikinis you want act like you have talent I know that was
770. 2:05:46that was a pretty cold burn uh so what's happening 30 minutes ago that's cute see okay app user three three eight eight six
771. 2:05:57three six eight by the way 2.5 out of 10 on the name come on you can oh Apple user all right I'll give you 3.5 out of
772. 2:06:0510 because it's because it's Apple user not just app user but uh that's a good troll what's happening 30 minutes ago
773. 2:06:12because I'm so far behind on the questions you see how this works here we troll that it's kind of funny it's kind of a
774. 2:06:19compliment and it's kind of a burn that's solid that's solid it's we got yeah we got clever we got clever participants here at the AI
775. 2:06:29learning lab by the way if you're new here looks like we got a fair amount of folks my name is Kyle Shannon this is
776. 2:06:33the AI learning lab if you ask me what what my qualifications are I'll [ __ ] tell you uh um all these things here are all
777. 2:06:43different flavors of chat GPT if you have not played with them go now go do it now just just go do it don't [ __ ]
778. 2:06:51sit there like I don't wanna I'm busy I'm swiping no go there and just tell it to write you a country song about your Tubby Dog
779. 2:07:03and then go oh that was fast that's not bad oh my God but it can't do what I do oh it can oh good lord should I be scared
780. 2:07:16go have that moment and if you don't know what to type there go to prompts.chat that thing right there and that'll teach you about prompting
781. 2:07:24and all of the [ __ ] that you can make chat GPT do for you like like this will start to scratch the surface of it and then um
782. 2:07:34yeah and then there's other [ __ ] there so yeah go do that all right and if you have questions about AI popping below I I do answer all
783. 2:07:43the questions if I can um although I'm 99 plus behind now so yeah I'm at least half an hour behind hey Mr cgo hey everyone the bear
784. 2:07:54yeah I'm a chief generative officer I don't know if you knew that ask for my qualifications I'm a chief generative officer what's
785. 2:08:01that you might ask if only I knew okay I'm generating [ __ ] could be Revenue could be bad copy um I like to ask Bing to tell me about
786. 2:08:16her dark side yeah Sydney Sydney Sydney got a little dark I don't know if you know that uh Taylor the uh
787. 2:08:24when they first launched Sydney and she told the New York Times Reporter he should leave his wife because she was in
788. 2:08:29love with him it got pretty got pretty dark and that made the cover of the New York Times and Microsoft was like oh we're gonna
789. 2:08:38have to lobotomize Old Sydney so she's a little more she's a bit more of a simpleton than she used to be she's not nearly as sassy
790. 2:08:52all right I wasn't receiving your nerd of no no notifications so I subbed win-win um are you doing the NPC thing if I can
791. 2:09:02make a hundred dollars a minute I'll [ __ ] do it cowboy hat cowboy hat I wonder if I could do the NPC thing and not be annoying Maybe
792. 2:09:15like all my stupid voices are kind of like NPC stuff I don't know I don't know I don't know all right I love your sense of humor and
793. 2:09:27absolutely admire your knowledge thank you very much thank you to buyer not bear oh wait the Hat a hat
794. 2:09:33oh wait and the bear you just hit some sort of level you hit you won an award I don't know what your award was but Tick Tock liked it
795. 2:09:42thanks tick tock the bear and I appreciate it all right I love your sense of humor absolutely admire your knowledge I'm learning a lot
796. 2:09:51from you thank you very much I appreciate that I appreciate that a lot um looking to analyze PDF bank
797. 2:09:57statements tried using GPT for coder but did not work go to just go to the uh Claude Claude you cannot look there
798. 2:10:09you tried using gbt Code interpreter and it didn't work it should work but sometimes it depends on the PDF you
799. 2:10:17know some bank statements their PDF is actually just an image that you have to OCR it so you might need to so check that out if if your PDF
800. 2:10:30actually has text in it then you should be able to do that at Claude if your PDF is just scanned images of bank statements
801. 2:10:41because we're in 2023 why would you have it be actual [ __ ] data what what is with banks and and medical
802. 2:10:51facilities just scanning pictures in but anyway Bard and Bing can both read images and I think both of them can OCR
803. 2:10:59the text into usable tables thank you for the heart the heart the heart I couldn't it's tough enough just like rambling for
804. 2:11:11this long but like that I don't know I'd want to just smack my face into a board much less the audience like does the
805. 2:11:22audience really like it I suppose it's like I suppose it's like oh you know I gave her a diamond and she said diamond
806. 2:11:31I you know funny back to that idea like we're entering this dawn of hyper personalization where you know right now it's real people
807. 2:11:41acting like that but not that long the NPCs and games are gonna do whatever the [ __ ] you want them to do so it's
808. 2:11:48there's a there's the there's the hint of a of what's in store for us there as annoying and weird as that is
809. 2:11:59um hey Kyle sure appreciate you helping getting to know this new world thanks so much so much you're welcome Daisy
810. 2:12:04sunshine I appreciate it what you have a d d I never would have guessed we already did that one I saw some of these questions could you
811. 2:12:14share some thoughts on bias and Equity concerns I would but the the the two times that I really tried to dig into
812. 2:12:21what was going on there I got my my life got booted um what I'll say on it is this is there's bias in the train data for sure
813. 2:12:32just because it was trained on you know the public internet that is a bias toward English it has a bias toward you
814. 2:12:39know whatever whatever it has um chat GPT in particular has what's called human reinforcement learning where it's humans
815. 2:12:48um saying that's good that's bad so anytime you have humans going thumbs up or thumbs down just humans have bias the
816. 2:12:54other thing about bias is there's no perfect unbiased set the thing that's unbiased for me is going to be radically
817. 2:13:01biased for you right so I think where we go in the future is that we will all be able to self-select into
818. 2:13:09different models that have different if effectively points of view I want one that's all verified news like open AI
819. 2:13:19just did a deal with the Associated Press so I might say I want to just talk about new stuff with open Ai and
820. 2:13:27Associated Press you know stuff because I trust that or whatever and then other things might be I want other kind of
821. 2:13:32sources in there and so I think in the future we'll be able to choose our models and and it might be the kind of
822. 2:13:39thing where you have you have you know a little dashboard and you can go I want to go you know more liberal more conservative
823. 2:13:46more funny more serious more medical more dating right like we might have different models for different purposes
824. 2:13:54and then and then sitting in front of those models are likely going to be agents where we're going to have personal
825. 2:14:00agents this is my dating agent this is my financial agent this is my entertainment agent this is my travel
826. 2:14:06agent and they're going to go off and do stuff for us right and and they probably um so so I think it is an issue
827. 2:14:19a lot of these companies know it's an issue and are working on it anthropic in particular claude.ai um they've got a thing called the
828. 2:14:28Constitutional learning model so where open AIS is human beings going thumbs up thumbs down and thropics says we're
829. 2:14:36going to write a Constitution and the Constitution is a set of four or five imperatives like you know don't harm humans
830. 2:14:45um increase Prosperity um I don't know be supportive and you know move the ball forward whatever it is they come up with their rules
831. 2:14:54and then their model is trained against that Constitution right so it's not a human being doing this that that
832. 2:15:01introduces bias but if you've got a transparent clear thank you very much for the hot if you got a clear and transparent set of rules
833. 2:15:10one you as a human being can look at those rules and go yeah I'm down with those rules I agree with those those are
834. 2:15:16great or those rules are horseshit I want a different set of rules okay let's go find you another Constitution right
835. 2:15:23so I like that model a lot um it probably has its problems like I don't know enough about it to know if
836. 2:15:33that's going to ultimately play out but I like the idea of having a transparent set of rules that I can go yep I'm good
837. 2:15:41with that and as long as all the questions are answered through that filter then that's good enough for me
838. 2:15:47and they'll get better and better and better at those imperatives and this is a place where also the other thing I like about the
839. 2:15:56Constitutional learning model is we're entering the 2024 election here in the states with all of this powerful AI [ __ ]
840. 2:16:04so we're going to see a level of quantity and quality of misinformation both words images and videos like just eye watering
841. 2:16:20and so I think we're gonna need post the election because it's like in the middle of it we won't be able to see
842. 2:16:28it but after all the dust settles we're gonna go well that was [ __ ] up we're likely going to need blockchain or
843. 2:16:34some sort of tokenization technology that we can start to verify things like verify humans verify authenticity and if
844. 2:16:42you have something like the Constitutional learning model you could say hey this is a learning model that's
845. 2:16:46been trained on this Constitution and it's verified on the blockchain that that's the case because there's just now that we have
846. 2:16:55llama 2 out in the wild right we have we have these open source large language models there's going to be all sorts of
847. 2:17:02crazy [ __ ] out there so so the bias is a thing people are aware of it there are some paths forward to deal
848. 2:17:13with it but there's no perfect solution so I think the perfect solution is multiple Solutions um but I don't know
849. 2:17:22I'm just a lowly out of work actor why do you ask looking to analyze PDF bank statements we talked about that what
850. 2:17:29plug-in was that for books it's called stories to anyone new here Kyle's answers everyone question everyone's
851. 2:17:36questions one by one thank you Lauren I do I do answer them one by one including yours where you told everyone I answered
852. 2:17:45them one by one see see I'm not a bot I'm not an NPC I'm not an MPC I'm not an MPC I'm not an MPS hat had had
853. 2:17:57I could do this I could totally we're going to change it's not going to be the AI learning lab anymore uh donut
854. 2:18:05I like donuts I like donuts I like donuts donuts is tasty Donuts is tasty hot dog hot dog [Laughter]
855. 2:18:21don't [ __ ] tempt me because you know I'll [ __ ] do it I like the spicy pepper them spicy peppers are tasty oh
856. 2:18:37[Laughter] when was it he lost his mind well it started innocently enough
857. 2:18:53David gave him a donut what ah it's so good it's so good hi Kyle hi all enjoy Chris Justine Bateman also outlined ai's
858. 2:19:08impact on entertainment that piece by Justin Justine Bateman if you haven't seen that piece by Justine Bateman I
859. 2:19:14think it was in Hollywood Reporter um or variety was it Hollywood Reporter I think it's Hollywood whatever you can just go look
860. 2:19:22up the Justine Bateman piece she's Jason Bateman's sister yeah former actress new computer scientist or yeah now computer scientist
861. 2:19:32she describes what's coming in the movie industry with like remarkable accuracy like like she sort of traces here's
862. 2:19:39what's going to happen with AI and basically from like an acting and writing standpoint it's [ __ ] terrifying and her point of view and the
863. 2:19:46piece is basically don't let AI anywhere near Hollywood which I don't think that's realistic at all I I appreciate
864. 2:19:53it but I don't think it's realistic but the point that she makes which I think is a really really good one is
865. 2:20:03the SAG strike and the wga strike these are the last two strikes that those organizations will have any Leverage
866. 2:20:12so what she basically says in the piece is get everything you can get everything you can in this country because
867. 2:20:22they will have leverage right because when you're competing against um dynamically generated young Bruce Willis
868. 2:20:32being put in any movie you want in the future you know you could be like well I'm a big movie star they're like yeah well we
869. 2:20:40could just use Bruce Willis if we want like so you are a big movie star but we're going to pay you 50 bucks for this movie so if
870. 2:20:47you want it you know take it but if not we'll use Virtual Bruce so it's it's rough I mean while but that's a wild piece it's a
871. 2:20:58very well written piece and it she really does point the future of what's coming like you can just sort of see
872. 2:21:04you can see what's coming she she kind of describes it as you know I want to see this kind of movie with this kind of
873. 2:21:09star with this kind of plot and it just does that South Park uh the South Park thing I mentioned was um it's called showrunner AI just go
874. 2:21:21onto Twitter and look up showrunner Ai and just scroll until you see South Park and basically what it is is you write in a prompt
875. 2:21:28and it generates an episode of South Park now it's it's a crappy version of South Park like the voices are all
876. 2:21:35roboty and the writings kind of stilted and shitty but it generates an episode of South Park so that tied to the
877. 2:21:43Justine Bateman thing you know the the showrunner AI if you don't know the term showrunner a showrunner is a producer in
878. 2:21:51Hollywood that can run a show right they can get a show produced and you know successfully executed that's a
879. 2:21:58showrunner so showrunner AI is we don't [ __ ] need no show runners we'll make your stupid [ __ ] shows with our AI [Laughter]
880. 2:22:09what was the one that wrote the e-book it's called Uh GPT author is the name of that project the other thing if you go
881. 2:22:16play with GPT authors and um the um what's cool about it is the Google collab notebook is you can see
882. 2:22:26all the prompts that they that they put in there it's like it's set up to write a fantasy novel but you could go in and just modify all the
883. 2:22:34prompts to write any kind of book you wanted and it's a it's a really clever it it's really clever what it does basically
884. 2:22:41what it's doing is it's sequencing a series of prompts so it starts out with let's get some inputs what do you want the book to be
885. 2:22:48about what do you want the genre how many chapters do you want give us your API keys and then it writes an outline writes chapter one writes
886. 2:22:56chapter two writes chapter three writes chapter four and then it gets to the end generates cover art for the book writes it into an
887. 2:23:04e-book so so it and so just go in and change it if you want it to be a thank you for the rose I appreciate that oh I could do I could
888. 2:23:15cowboy hat uh don't tempt me I'll [ __ ] do it you know I'll do it uh new closed captioning on tickety talk
889. 2:23:31really what's that David gold with the Wild what is that chili is that a gummy bear I can't I can't see
890. 2:23:40what that is oh that's a little dinosaur a little dinosaur ball tennis ball oh God okay I'm not gonna do it I'm not
891. 2:23:50gonna give in I'm not going to give in I'm not going to do it what program is this this program is NPC world
892. 2:23:58he got a little duck he got a little ducky [Laughter] can we make our own games so we can make our own games you can make your own
893. 2:24:11games right now with these things go to chatgpt and tell it to program you a um game of snake or a flappy birds clone
894. 2:24:23you can do it in code interpreter you can do it DBT 3.5 it's probably for that kind of stuff I think bards there's examples of out
895. 2:24:33there of people doing it on Bard so yeah you can make your own games now the games you're gonna make initially are going to be really
896. 2:24:42you know basic and and just like proof of concept like the the game equivalent of hello world programming right
897. 2:24:50but just like that GPT author thing so so you could do a single prompt to write a book and it'll write a really shitty
898. 2:24:57book so what they did with GPT author was they said well let's not write a shitty book let's start with some good inputs let's
899. 2:25:04start let's then generate a solid outline then based on that outline let's write chapter one right and I assume I
900. 2:25:12assume it's not just an outline it's probably an outline plus you know character like a character sheet and then they write chapter one then
901. 2:25:20they write chapter two so they're doing a sophisticated chain of prompts that generates that thing so we're going to
902. 2:25:25be able to do the same thing with games right so if you go to blockade Labs skybox.blockadelabs.com where you create
903. 2:25:32that 3D world that you play in right now it's just this static 3D World but when they start stacking together
904. 2:25:40multiple prompts and multiple tools you know your prompt will be some version of give me this world with these kind of characters and these
905. 2:25:52kind of weapons and this kind of gameplay and this kind of narrative go and it'll do it it'll [ __ ] initially gonna duct tape
906. 2:26:07it together and then you'll play a game and you'll be like it's not as perfect as I want it to be it took five minutes and just did
907. 2:26:17something miraculous but I'm bored so that's how people will be for a year and then someone will come out with the
908. 2:26:25real version of it just think again realize in front of your Vision Pro goggles that now look like these glasses that's five years out
909. 2:26:37um so yeah and again Sam like that's my the the whole idea here of why I keep pointing people just go try it
910. 2:26:47just go play with this play with the image stuff play play play is is to as much as you can get your head around not what these are but what
911. 2:26:57they represent like imagine these three years from now and it's it's hard it's it's actually really hard to comprehend
912. 2:27:11if you've got like Bing Bing is just Microsoft's crappy search engine right that's what it was for years
913. 2:27:19and now you can upload an image to it and it understands what's in that image and you can say make me an image of anything and it will do it
914. 2:27:27and you can say go find me a thing on the internet and rewrite the marketing copy for me and it'll do it now
915. 2:27:35because it's early and because they've got safety guard rails and all this it does all that kind of shittily but
916. 2:27:42imagine if it wasn't shitty and imagine if it was multi-layered and all you did was give it a go I'm thinking about starting this
917. 2:27:51kind of business great here's your website here's your marketing plan slide this slider for how much money you
918. 2:27:59want to spend on Advertising here's your products choose the ones that you like press go to start your business here's
919. 2:28:06all your legal documents right you start to you know don't think about these tools as individual tools think
920. 2:28:17about if they all came together as a series of capabilities that you no longer have to think about that your
921. 2:28:27agent who knows what you want and those preferences and knows your data and knows your background and knows that you have [ __ ] add
922. 2:28:37so it's only going to give you the shiny things it's it's going to take care of all the boring repetitive [ __ ] that it
923. 2:28:43knows you can't stand and then you just tell it what you want and it's going to go do it that's where these things are headed
924. 2:28:53so yeah you can make your own games right now but you know lower your expectations for now but imagine five
925. 2:28:59years from now what's that look like and if if you don't go holy [ __ ] [ __ ] then something's wrong you're not looking at it right
926. 2:29:06because I you know again I come back to everyone's like well this is over hyped I'm like no no the web in 1999 was over
927. 2:29:14hyped right I was there in 1994. it was pretty hyped 1998 when I had three employees walk into my office within one
928. 2:29:23month and quit because they had been funded for 10 or 20 million dollars for a shitty website idea that was overhyped this stuff
929. 2:29:33is not overhyped in my opinion it's ridiculously hyped but what it does and what's coming is it's just it's it's on another plane
930. 2:29:47like it it's just like like what it feels like to me like the the like why I keep dealing these Lives why
931. 2:29:56I keep doing this channel why I keep leaning into this is it just feels like I I cannot get my head around what's coming
932. 2:30:03and I'm normally like for me I can normally get stuff like that like when I saw the web I knew immediately what was coming I could see it
933. 2:30:12I could feel it I knew I knew kind of what the future would look like I remember getting kicked out of I was
934. 2:30:18invited to speak at a the newspaper Publishers Association somewhere up in [ __ ] Connecticut and I went up there and I was like you
935. 2:30:27know my cocky 30 something you know it started agency.com and I've got all my confidence okay before and I went up there and I said I
936. 2:30:38sat on the front of the stage and I said you know you should consider um you should consider experiment ING with blurring the line between
937. 2:30:48editorial and advertising and I got escort it off the stage and ask to leave the facility but look what happens right
938. 2:31:00that line is completely [ __ ] blurred you could see that from 1994 it was pretty clear that once you sort of collapse
939. 2:31:09communication into you know sort of instant jumping from here to there you can kind of see what the implications of
940. 2:31:15that are I have no I like it's almost impossible to comprehend what the implications of I can just make an image
941. 2:31:22I can just write a screenplay I can just write a marketing plan I can just write code I can just tell a thing to do something
942. 2:31:31code interpreter oh thank you thank you the bear I can just tell code interpreter I wanted to do something to
943. 2:31:38analyze some data and it will start writing and executing python code to accomplish what I asked it to do
944. 2:31:47what are the implications of that I don't [ __ ] know let's say that you've got a data Analysis company that's been around for
945. 2:31:5630 years and you're the most trusted data Analysis company on the planet what when you look at what code interpreter
946. 2:32:07does what what happens in your brain it can't be normal it's got to be like holy [ __ ] I've got [ __ ] 400 people that do that
947. 2:32:19what happens to them do they have to learn this stuff do they have to learn some new version of what they do yes
948. 2:32:27because that's who you're going to be competing against is it there right now no is it going to be there in a year yeah
949. 2:32:38I like Civilization and strategy yeah see I'm I'm more trigger twitcher Chris what is MMO uh massively multiplayer
950. 2:32:49online massively multi multi yeah massive multiplayer online lots of people running around doing [ __ ] I'm a Survivalcraft and gamer
951. 2:33:01um oh Tick Tock just took me back in the it just returned me to the Past sorry I'm gonna go oh now jumping to the
952. 2:33:11bottom all right I'm gonna skip toward the bottom I gotta it's getting late I gotta go soon um I'm just gonna skip a bunch of
953. 2:33:17questions I know I don't like to do that but I'm gonna do it so I'm gonna grab some questions at the
954. 2:33:22bottom here I really do appreciate y'all hanging out with me this has been cool you're an overqualified actor and we love it
955. 2:33:30that's awesome thank you I appreciate that um he does have the holly hollywood industry energy yeah thanks thanks
956. 2:33:44the hair makes you more MPC believable uh uh YouTube raised their prices by two bucks whoa wait what what's this hey pie
957. 2:33:57can can tell you movie times it has internet access it seems yeah yeah Pi Pi is connected to the internet
958. 2:34:07so it's heypi.com or Pi dot AI or the IOS app it's connected to the internet so it it will um it's it's I'm telling you pie is an
959. 2:34:20interesting one this the other thing about Pi is it looks like a really um it's it's a very thin application layer
960. 2:34:29right now it's not doing much um they just raised 1.3 billion dollars that company that little teeny tiny you are
961. 2:34:391.8 billion dollars seriously you know how old they are one year I'm 11 years into Story vines we haven't raised 1.3 billion dollars yet
962. 2:34:57so Pi dot AI is going to be something I think I think they're going to be something right now it's glimpse into the future but
963. 2:35:07honest question what is the point of pie dot AI oh so I was just talking about that I I honestly don't know their their
964. 2:35:14mission is to create personal assistance so I was talking earlier I don't know how long you've been here but I was talking earlier about
965. 2:35:22we're going to have all these different personal agents that are kind of different modes of our life right here's
966. 2:35:28here's my social agent and here's my professional you know get a job agent and here's my vacation agent and we're
967. 2:35:37going to have these different agents now maybe someone like apple figures out how to have a single agent that just does
968. 2:35:43all that different [ __ ] for you but I could see there being different kinds of uh personalities so I think what these
969. 2:35:49guys are up to the the 1.3 billion dollars that they're um that they've raised they're probably spending 900 million or a billion of
970. 2:35:59that on 22 000 high-end gpus so they're building a massive GPU Farm what what does that mean that means they're going
971. 2:36:06to be training their own large language models and I think they're going to be focusing on autonomous agents right so
972. 2:36:14someone that you talk to so the thing about PI right now is just go talk to it they've got it programmed so that it's
973. 2:36:21very very conversational and it's connected to the internet and it's sitting on top of a GPT 3.5 like large language model
974. 2:36:30just imagine that kind of blown out where not only can It have a conversation with you and go get internet information but it can go start
975. 2:36:40to do stuff on your backpack um and that's probably discoverable at their website so so if you go to chatgpt get the web
976. 2:36:52pilot plug-in point it to the inflection website and just tell it to summarize what the [ __ ] those guys are up to
977. 2:37:02because I think it's going to be big and very very interesting um I always thought hey pie was kind of
978. 2:37:09a therapist type AI yeah it's it is and it's it's very very conversational and it's like I I found it surprisingly
979. 2:37:16enjoyable I'll tell you where pie is not enjoyable as if you're trying if you're trying to do like research and work it's
980. 2:37:22really [ __ ] annoying because it's like let's have a little conversation about that shall we Bob no just give me my [ __ ]
981. 2:37:31go to Bart if you just want your [ __ ] go to pie if you want to have a conversation go to chatgpt4 if you
982. 2:37:37wanted to actually be good writing go to Claude if you have giant documents to deal with go to Bing if you have to um
983. 2:37:47so yeah and and go to go to llama2 if you want to build your own [ __ ] sponsored by Doc Brown hair tonic I need
984. 2:37:55some [ __ ] doc Browner Atomic oh I guess that's kind of what it's looking like man it's just it's Jimmy Neutron now um
985. 2:38:09someone said I have Hollywood energy there was someone on last night going I'm a talent agent you should have a show
986. 2:38:15never heard from them again but you know all right uh same question death by butterfly what a death by butterfly asked oh
987. 2:38:29he does have Hollywood energy not a playable character all right like the character Bots and video games that say the same thing
988. 2:38:38hat I got a hat this part all right a few more questions and I'm
989. 2:38:54gonna go I'm starting to play around with creating agents in steamship what's steamship why do I not know about that stem ship
990. 2:39:12um what is steamship AI hey it understood me that was pretty funny manage stack for language AI packages
991. 2:39:28steam stat steamship is a managed stack we would have wanted on our past large language AI products we've built
992. 2:39:36world-class Auto manage reference stack ingesting natural language data training and running models on it oh that's cool
993. 2:39:43so what are you doing oh you're building you're creating agents in steamship that's super cool yeah that's super cool if you if you get
994. 2:39:50some of those um if you get something built that you want people to play with drop drop it in here we'll go check it out maybe we'll even
995. 2:39:58check it out we'll check it out live on the internet live on The Tick Tock we'll go live on a tick tock we'll go to
996. 2:40:04Steamship we'll go into your little your little agent and we'll go play with your agent and then you can install malware in our computers
997. 2:40:12that'd be fun let's do that nowhere is so cute I love it when my screen just goes black [Laughter]
998. 2:40:26ah thank you for the rose I got a rose I like turtles I almost started the evening with that but we'll go we'll go hang on where is my
999. 2:40:44um back here live at the waterfront village with my friend the zombie Jonathan
1000. 2:40:59you're looking toy Jonathan just got an awesome face paint job what do you think I like turtles all right
1001. 2:41:08the Waterfront Village open for the next 11 days [Laughter] got an awesome face paint job what do
1002. 2:41:24you think I like turtles all right you're great zombie oh my God oh my God and there's there's another one from Kill Tony
1003. 2:41:37um that's almost identical to that you know I don't know if you know kill Tony um uh let's see
1004. 2:41:55something you think we should know about you as we get to know you what do you think about 3 000 people in front of you
1005. 2:42:00what's something interesting about you I'm not [ __ ] but I like rocks hahaha [Applause] [Laughter]
1006. 2:42:15and then he pulls rocks out of his pocket oh my god oh sorry sorry I think you know a little bit too much about me
1007. 2:42:30oh my God [Laughter] I like turtles oh who reminded me of that focal point.biz thank you thank you for the
1008. 2:42:45comic relief oh my God can't breathe how did he pass apparently a turtle took him out [Laughter]
1009. 2:43:04oh my god oh the closed captioning makes it it's now the official AI learning laugh oh I like that that's actually a good
1010. 2:43:16name good name Dr J I like Doodles oh my God oh and scene bring it down a little
1011. 2:43:32break the flop sweat AI is dangerous in the wrong hands the actors and writers should have seen the red flag when AI hmm models were used
1012. 2:43:44well okay here 's the deal I mean yes actors and writers should have seen the red flag and do what about it like that the problem is this
1013. 2:43:57and do what about it so so if you look at like one of the straws that broke the camel's back for the Screen Actors Guild was the The
1014. 2:44:06Producers After ignoring them for two weeks told them to go [ __ ] themselves on almost every request and then they had one thing where they
1015. 2:44:13but we've got one groundbreaking proposal and it was we'll pay background actors 200 to scan their likeness and then use
1016. 2:44:22their likeness in perpetuity for that 200 and never have to ask him again or even give him credit um and so
1017. 2:44:33sag rightfully said that sucks go [ __ ] yourself we're going on strike and but but the producers could could easily take the position oh
1018. 2:44:46fine well then we won't pay them the 200 we'll just create synthetic actors right and so so so they're gonna
1019. 2:44:56have to use whatever leverage they have now to get what they can where you know background can't be more than you know
1020. 2:45:02sixty percent NPCs right it's got to be some amount of of day players right um but but I don't like what can they do
1021. 2:45:12about it here's the other problem with this this so so the Arts have a particularly rough problem with um
1022. 2:45:24with labor issues because for every actor there are easily 10 000 actors behind them that would kill a family member for that role
1023. 2:45:39like not joking for everyone you see on screen there's a pile of actors behind them that would do anything for that role so it's a real
1024. 2:45:51tough position to take a strike right what they have to do is they have to leverage Their Stars and then you know in this case they may
1025. 2:45:59have someone like the teamsters come in like the UPS strike the teamsters have a lot more power and so if those two
1026. 2:46:06things kind of gang up they'll be able to get something but but you're not stopping the AI like you're not this this stuff's not getting
1027. 2:46:16uninvented if you if you've seen I showed this the other day we'll go back to the we'll go back to the totals I'm not [ __ ] but I like
1028. 2:46:28rocks uh Barb and Heimer so this is a clever trailer that someone
1029. 2:46:44put together using AI ladies and gentlemen we are in a race against the Nazis and I know what it means
1030. 2:46:55she's a scientist and I hear she's the best good she's also a doctor excellent she's also a flight attendant astronaut
1031. 2:47:03Marine Corps medic paleontologist veterinarian and three-time Olympic gold medalist I see the bright pink glow of a new era
1032. 2:47:18re-imagine a future in our imaginings look fabulous party together we're gonna beat those Nazis and we're gonna look so good doing it
1033. 2:47:34hey Barbie can I come to your secret laboratory later sure I don't have any plans just giving Humanity the means to destroy itself [Music]