Jan 1, 2024 - Unlocking AI's Potential: Create Your Own GPTs for Good in 2024

Transcript

1. 0:02[Music]
2. 0:19ready s in this long time when things are going to change my life away seems these dreams have turned to
3. 0:34Bunch glad get my nerve up but my past is pulling me [Music] down wondering how long this black sheep going to stick
4. 0:51[Music] around somebody told me once before he said you can never I'll go home again won't you leave say a just to me
5. 1:08away from the truth of who I am and what I believe so I thanked him for his two SS with a hand shake and some
6. 1:18sympathy yeah I packed up my blue jeans and I headed for this big price of my freedom byebye Black
7. 1:34Sheep the black sheep of the family bye oh that means so very much to me byebye bless she to my friends and my
8. 1:54family by going off and set my Soul set it [Music]
9. 2:17[Laughter] free get that dog some earplugs get me some earplugs what are you talking about that dog loves to
10. 2:31sing oh good lordy people happy New Year happy New Year Steamboat Willie is uh in the public domain go make Mickey Mouse
11. 2:43well go make that version of Mickey Mouse I suppose I don't know the rules on this [ __ ] but apparently you can do
12. 2:48something with him I'm sure Disney's still going to sue your ass but you know here we are making content with
13. 2:58Mickey oh good Lord I guess it's Steamboat Willie Steamboat Willie good evening fine people fine
14. 3:12people of the lab the AI learning lab the Internets the Tik tocks the world I didn't really check the hair
15. 3:22tonight I guess all right the glue's still holding oh what did I do today I watch some fruit ball I uh I got
16. 3:35my I got caught up on uploading all my AI learning Lab videos to YouTube I'll give you the link for that cuz I also made a
17. 3:44playlist for the GPT for good um sessions that we did um so you can learn how to do your own GPT you can
18. 3:55watch those sessions and as soon as I have the the links for ANS Zoom videos I'll add them to that playlist so this
19. 4:03playlist will will get updated with all of the sessions from that 24-hour period which will be cool that'll be a cool archive to
20. 4:13have well cool for us hopefully cool for [Laughter] you all right everybody welcome welcome welcome welcome to the AI learning lab
21. 4:27how are we going to kick off the year what do you want to do too it looks like apple actually released their freret model there's fret
22. 4:397B multimodal model into the into the wild today so Apple Apple's starting the year off with with an announcement it's a
23. 4:50minor one but it's still pretty cool do you have three beverages going I do not I only have two beverages going Disney is Tainted
24. 5:00these are these are the the dose dose oh look depending on if I hold one closer this is an old fishing
25. 5:10trick if you hold the fish closer to the camera it looks much larger this is actually smaller than this but is it by the way this
26. 5:27idiocy welcome to Chad add this is this is what neuro divergency looks like this is your brain this is your brain on
27. 5:42ADHD and Amelia's wife yes you're late you're late you missed champy was on fire tonight he was going off he was in
28. 5:52fact he was just sitting here staring at me like we're going we're going we're going right we're going to go we're
29. 5:56going to go and the minute I tried to tune my guitar he was like don't I was like calm down just zip it zip it he does not know the word zip
30. 6:09it that's a fine pup that pup that champy all right fany ass let's get this show on the road let's get it going
31. 6:18let's get her going all right listen if you're new here my name is Kyle Shannon and uh I'm what you would call an expert
32. 6:26influencer on a topic of artificial intell Ence I'm just it's I know everything about it it's in fact if there's not
33. 6:36probably a single thing I don't know about the artificial intelligence sometimes people ask me what are your
34. 6:43qualifications I got a certificate I got a I got a degree uh I got a I got a a degree in a certificate uh I I have learned in in the
35. 6:53past week or two all the AI I know it all I know it all I know every answer there's this thing called prompt
36. 7:01engineering it's going to be the big job of 2023 it's going to be fantasttic what it's 2024 I don't know anything people I know
37. 7:15nothing I don't know nothing there's no experts there's no experts I promise you I promise you that anyone that tells you
38. 7:23you're an expert they're not an expert they're they're a charlatan they're a scam artist they sound like this you can
39. 7:31make money with don't fall for it don't fall for it don't fall for it I fell for it once you know what I did I paid $1
40. 7:39199.95 a week you know what I got a certificate it was useless in a week all right do AO for champ and call
41. 7:51it something I the pin went away dang it I love to cast you on my on my big screen only the sound is a little bit
42. 8:04later I'm a I'm a regular [ __ ] oneman show over here do AO and champ and call it champy Blues we could do that I gotta I Gotta
43. 8:16Give sunno enough enough we we actually that's a that's a good idea champy Blues okay all right so if you don't know what sunno
44. 8:25is sunno is a um a music AI that will write a song for you and they've Incorporated it into Microsoft co-pilot but Microsoft co-pilot is
45. 8:41flakier than crap so let's let's let's go let's go see if the first day of 2024 shows uh shows Microsoft co-pilot being
46. 8:51less flaky than it was for the calendar year of 2023 I'm going to go out on a Limon say no but but you know Sydney Sydney she's
47. 9:03a nasty a nasty nasty woman all right oh by the way welcome everybody my name is Kyle Shannon this
48. 9:18is the AI learning lab if you don't know me I'm a bit of an odd Feller um co-pilot we're going to go to co- pilot. microsoft.com
49. 9:29which is Microsoft's new branding for Bing chat see I've got zero out of two I've got 2,000 tokens that generally means
50. 9:38this is not the fancy version of this but I don't know we'll see um okay make me a song uh make me a
51. 9:53blues song about a black and and white singing dog oh and because I did black and white it's probably going to say
52. 10:05this violates its content policies for for racism um a black and white singing dog named champy all right okay all right
53. 10:21fantastic let's see let's see if it does it now it should hang on so in order to to have this thing um make you a song
54. 10:30you need to have the sunno plugin turned on and in order to do that you have to have search turned on and thenso is the
55. 10:37little plug-in right below it all right the the plug how they've Incorporated plugins here is really weird because it
56. 10:44it requires you to turn search on and then search seems to screw up chat GPT or whatever the hell they call this
57. 10:51thing co-pilot okay so we're going to submit that let's see let's see if it does it people it should fire off the plugin it should
58. 11:01fire off the plugin ah it did okay that's good that's promising I'm finding that switching from balance to creative
59. 11:07or precise makes helps when it flakes out okay that's good to know Athena Luna all right so let's see late night when the moon was high I
60. 11:19hear howling in the distance my oh my down in the street comes a black and white dog singing the blues oh what a talking dog oh champy
61. 11:29sing your heart out for the world to see this is good when your Soulful how you'll make Believers out of me oh
62. 11:36champy you're the star of the midnight show it's so true sing in those B Blues wherever you go yeah gotta go you're a great human oh
63. 11:49thank you very much I appreciate that all right let's listen to champy Blues blues for champy at night when the moon is high [Music]
64. 12:01I hear aing in the distance my old down the street comes a black and white doll singing the blues what talking
65. 12:15do horrible how make out of me all right it's fine it's it's it's
66. 12:32it's where we are it's where we are in the in the industry right now it's it's a solid C minus it's it's acoustically a D minus
67. 12:44uh chord progression a d and overall a c c minus minus this shit's going to get so much better though
68. 12:59you just wait you just wait all you naysayers sitting there with your arms crossed going I hallucinates I don't
69. 13:08know if you knew this Kyle but it it doesn't tell the truth all the time it's stealing from everybody oh I get it I understand I
70. 13:21understand your reticence co-pilot wear co- pilot. microsoft.com that's the micro roft version of chat GPT that if if if we
71. 13:33assume that chat GPT is the is the state of the-art right now chat GPT 4 so chat GPT plus let's give that an a we won't give
72. 13:43it an A+ because it's still too janky it's irregular worthy for now yeah exactly exactly if if if I give chat GPT
73. 13:514 an a cuz it's state-of-the-art right now nothing's even really close to it until further notice um co-pilot microsoft.com Microsoft's
74. 14:04version of chat GPT I would give a c minus like I would give Claude like a B+ I would give
75. 14:19Bard A B B minus um I'd give grock a d I'd give Pi a B+ but it's different but
76. 14:34for conversational AI actually I'd probably give Pi an A minus for conversational AI um perplexity I don't know but it's
77. 14:47probably in the in the a minus in the a minus territory I just don't use it all that much try some classical music themes I'm
78. 14:56good I'm good with sunno uh Theo stuff's weird and predictable like they don't they don't have quite enough parameter like like they don't
79. 15:06quite have enough controls to control it and write a full song it only writes like a a Whata call it a verse and a
80. 15:15chorus it's not a full song so I'm done with suo for a while I'll let it I'll let it get one more one more version
81. 15:22better before I start playing with it a lot it's it's like it's like all the stuff right now all all the multimedia
82. 15:29stuff animation 3D objects text to 3D World video audio what else is there gaml all that sort of stuff right now is
83. 15:45all just it's it's just janky it's just janky it's janky janky janky janky janky janky janky janky janky ask chat GPT to
84. 15:57create a port cord progression and play it for us yeah but I don't know that I could try the it's it's it's okay at that it's not
85. 16:12great at that um and I don't really know how to prompt it right which is part of the problem let's
86. 16:27see champ will sing like a mother tonight though I'll tell you that all right let's see um you are are a
87. 16:44Grammy award winning Americana singer song writer you
88. 17:00are inspired by let's see Brandy Carlile Lucinda Williams Martin seon
89. 17:17John hayatt that's good
90. 17:31I want you to write a haunting song about a black and white dog named
91. 17:51champ who can't help but sing when whenever anyone strums a
92. 18:05guitar The Ballad of singing champ in a little town under Big Sky Blues all right oh champ sing your Soulful cry underneath the vast Open
93. 18:21[Music] Sky now now write the chords for the
94. 18:36song that are interesting and non cliche and this is where like it it tends to not be great at
95. 18:57the well well it's actually good enough that I don't know how to play a lot of these chords all right I'm going to go wait
96. 19:06okay simplify them a bit that's really funny a minor gcd okay well let's
97. 19:22see oh a minor that's Champ's chord actually right so [Music]
98. 19:46[Music] yeah [Music]
99. 20:04[Applause] that's actually not bad did you like that champy that's your song buddy that's your song buddy we made you a [Laughter]
100. 20:18song oh ask it to rewrite oh that's really good ask you to rewrite the Constitution but from our founding mothers that is a good prompt all
101. 20:34right you are a renowned historian with a reputation for finding documents no one knew existed
102. 20:53period it turns out that while the founding fathers were writing the US Constitution their significant
103. 21:09others gathered as well period Please write the alternative Constitution from the founding
104. 21:24mothers certainly this is what I love about chat PT is just like hey I got you I got you sure no problem we the women of the United
105. 21:36States in an effort to form a more equal and nurturing n Nation nurturing establish justice ensure domestic
106. 21:43tranquility promote the general well welfare and secure the blessings of liberty for ourselves and our posterity
107. 21:50do ordain and establish this constitution for the United States of America article one the legislative Branch every state shall have
108. 22:00representation in proportion to the population including all persons regardless of gender ensuring women have
109. 22:06equal representation in the legislative decisions of our nation voting rights shall be Universal and unimpeded for all citizens of legal
110. 22:14age irrespective of gender race or class Congress shall make no law that hinders the right to education and shall
111. 22:21Endeavor to promote welfare this is very very cool yeah and you know I I mean I did a i did a similar kind of exercise
112. 22:29to this it wasn't wasn't exactly the same thing but I did a kind of exercise to this of um um when the when Biden's executive
113. 22:41order on AI came out a lot of people were asking me my opinion on it and I Tred I try to look at things from a lot
114. 22:49of different angles and that one in particular I was like man people from with different points of view are going
115. 22:54to have very different responses to this you know some of it's just po politics some have will have legitimate issues
116. 23:01one way or the other and so I had chat GPT I came up with five personas and then I had chat GPT tell me if I was
117. 23:10missing any so we came up with eight eight different personas so a Libertarian a conservative a progressive
118. 23:16all that sort of stuff and and then um and then I Had Each Persona react to the executive order you know what what was
119. 23:27their opinion on the executive order it was fascinating I did a LinkedIn post about it hang on the LinkedIn post
120. 23:41was is this going to let me in is this going to let me [Music]
121. 23:56in articles so if you go to my LinkedIn profile which is Kyle Shannon in my articles there's an article
122. 24:11called but what do they think of the executive order and so I've got an overview of the executive order
123. 24:28and then the company exec at a major Tech firm the AI skeptic the regulation proponent the libertarian the generative AI Optimist
124. 24:40and Community leader that's [Laughter] me the AI researcher and academic the consumer rights Advocate that was
125. 24:49something that chat GPT came up with the international Observer oh there's nine of them and then the small Tech startup
126. 24:59founder that was that was a pretty cool little exercise and it's I mean you know it's it's kind of thing
127. 25:09where because of the way AI was trained it was trained on you know all of this output of human beings right that that
128. 25:17put all this stuff on the internet you like inherent in the training data even though there's bias and there's all all
129. 25:23the sort of stuff that people talk about that needs to be addressed which I agree with if you think about the amount of
130. 25:32information that's been put on the internet in the last 40 years it's a lot and and there's a lot of point of points
131. 25:38of view represented there and if you do your prompting cleverly and you do exercises like this where you're not just necessarily
132. 25:46writing from a single point of view you're writing from multiple points of view it like chat GPT is actually really
133. 25:54good at Nuance it's really good at at you know know helping explain things from different pe people's points of
134. 26:00view and so I think this kind of exercise is is actually a really valuable a valuable kind of thing this
135. 26:07it like it opened my eyes I thought this was an interesting an interesting little piece so anyway for what that's worth for what
136. 26:15it's worth you know what I mean you know you know you know what I'm saying H you know hey you could read it you could not
137. 26:21read it hey it doesn't [ __ ] matter I'll tell you one thing you could do you could take the text of this article you
138. 26:26put it in the chat GPT and you say hey just give me the skinny like what the [ __ ] does what the [ __ ] he saying here huh what's he saying
139. 26:33here and then you go in there and you say hey I like I like it I don't I don't a boots whatever you know what I'm saying all right
140. 26:44fantastic I remember that day Kyle hey Robert Rossy using personas really helps me draft a prompt that oh I just closed
141. 26:53your thing sorry about that I couldn't I didn't read the whole pin whoever put it there [Music] there how many gpts did you get set up
142. 27:08in the oh wait uh using personas really helps me draft a prompt that shows perspectives yeah actually I'll tell you
143. 27:15what that would be a great GPT to build okay everybody all right let's go let's go listen listen listen people listen let me tell you
144. 27:25something you're going to go right now to this URL no not that one don't go there go here go to that one all right
145. 27:35go to the salon. and it might not work cuz I don't I think my web my register is a piece of [ __ ] just like it's like
146. 27:43all this AI technology so go to the salon. and that's going to take you to a link tree and I've updated some things
147. 27:52the first link is the GPT for good event directory that uh that Vicky Baptist built for us so if you don't know what
148. 28:03this is um on Saturday this past Saturday starting at 6:00 a.m. um Ann Murphy of empowered fundraiser and myself kicked off this
149. 28:1524-hour event called GPT for good it was the it was the 24-hour GPT for good challenge and our goal was to to to have
150. 28:24participants build 100 gpts for nonprofits or good causes and so I am happy to announce that we hit as of today
151. 28:37151 and so this site shows you all them so what did we build in the first 24 hours here's the first 100 gpts that were built and
152. 28:51then we hit our goal of 100 gpts and kept going so here's the next 50 or 51 I guess um isn't that amazing and so so that's
153. 29:02uh that's what we did over the weekend it was long and exhausting and Vicky bless her heart not only did she
154. 29:09build this site she stayed up so an and I had six hours in between our sessions so every 3 hours we did a training session and um
155. 29:22and but Vicky was there for all of them so she just she didn't take naps I don't think I think she just grinded
156. 29:30through and so she was updating this this directory list in real time as these things were getting built uh and
157. 29:37it's just it's really inspiring so please go here and and check them out and share this and and and you know if
158. 29:45you want to do this you know what I'm going to do I'm going to turn back on I'm going to turn something back on on the link tree
159. 29:55here you can no longer sub submit them for the there's a $250 drawing that we're going to announce tomorrow at the AI
160. 30:03salon so you're no longer eligible for that but I'm going to give you access to the there's a document that is the prompt sheet that
161. 30:16we used for these courses so that's now turned back on and where did I put it yeah that's in a good spot all right right oh wait hang
162. 30:28on e e slash Ai No AI artist a i there we go come on what is going
163. 30:50on there we go um okay so the first the first link is to that that directory the second link is to the YouTube
164. 31:02playlist um which includes the night before which there's nothing valuable in that one it was just fun um but but the one starting at 9
165. 31:13a.m. uh and started at 6: a.m. I don't have ANS in here yet but these are the sessions where where we taught people how to make
166. 31:22gpts so you can you can go watch those videos and play along and then here is the prompt sheet that is this is
167. 31:30basically the the prompts and and the instructions for how to build your GPT and how to share it with the world and uh yeah so
168. 31:45151 151 that's a lot that's a lot of that's a that's a whole lot of gpts um for for an event that was just
169. 31:57kind of a like a fever dream idea the week before came together pretty good it helps to have some some kick-ass
170. 32:08co-conspirators um an an Murphy and and Vicky both Vicky Baptist and and also the Irregulars so one of the things that
171. 32:15made it a success is a lot of the Irregulars a lot of the people that come to this channel regularly showed up to
172. 32:21to multiple sessions of that event and just kept building gpts like TK made um 20 of them or or more and and lunik had made like 74 of
173. 32:34these things previously and so she was pulling out the the ones that were tied to nonprofits and put it was just it was
174. 32:40amazing it was just amazing so anyway thank you to all all who involve all who were involved who's programming who's programming chat
175. 32:50GPT open AI Greg Brockman and and uh the gang over there at open AI or the one who's programming it who's
176. 33:02programming the gpts we did anyone who wanted to that's the the amazing thing about way the way they built the gpts is
177. 33:09you don't need to know programming at all there's no code tools where you actually do need to understand
178. 33:15programming you need to understand apis and you need to understand programming logic and you need to understand inputs and outputs
179. 33:23and these things you don't need that you it if you have those skills you can make more sophisticated gpts but you can
180. 33:32literally in 5 minutes just talk a GPT into existence my very first GPT I did this thing called Sin City sipper I did
181. 33:41it in five minutes on on a live with no programming and it was brilliant wait who's making the guard rails chat GPT open
182. 33:54AI open AI is making the guard rails or in the case of co-pilot Microsoft is making the guard rails or in the case of
183. 34:01Bard Bard is making the guard rails or in the case of llama from meta no one's making the guard rails because they gave
184. 34:08out the the source code they gave out the weights the one of the significant things that meta did was they kind of ensured that these
185. 34:18large language models are going to survive whatever assaults come from politics or or industry we just saw a big lawsuit from The New York
186. 34:31Times um so so I mean your question about guard rails I like it sounds like it's coming from a place of of you know like
187. 34:40like you don't trust anyone and who's making a guard rails well there's educate yourself because there's a lot
188. 34:46of different models out there um open AI was intended to be an open-source company it was a nonprofit and Sam mman in realizing how powerful
189. 35:00these things got was like I don't think it's appropriate to put this level of powerful tool out there with no guard
190. 35:08rails on it that that you know people could do bad things with it now you could argue well who's he to say what
191. 35:16the rules are well at least he's thinking about it and then you look at a company like uh like anthropic that is
192. 35:27uh founded by a bunch of exop AI people that didn't like the way open AI was doing safety they're they're human
193. 35:34reinforcement learning they didn't like that model so they came up with a model called the Constitutional learning model
194. 35:40and what the Constitutional learning model says is let's create a set of rules and rather than having human
195. 35:46beings um determine if it was a good answer or a bad answer why don't we run it against those set of rules and so the
196. 35:54nice thing about anthropic in that Constitution model is you can make that set of rules available to everyone and it can be
197. 36:02transparent how I imagine this is going to go in the future spike is um there will be different models out there and
198. 36:10we will all opt into our particular level of safety or our particular flavor of safety right there's going to be some
199. 36:19people that look at you know what open and AI is doing going that's perfect for me there's going to be other people that
200. 36:25like I wanted to lean this way or that way or I wanted to skew you know in favor of this or in favor of that there is no single nonbiased
201. 36:38thing because we're all humans we've all got biases so so every model is going to piss someone off right and if you really
202. 36:49want to play with a completely unrestricted model have fun it's going to cuss a lot it's going to do a lot of nasty [ __ ] it's going to
203. 37:00randomly lose its mind and just start speaking a different language um I have a buddy who took the UN unrestricted version of um llama 270
204. 37:12billion parameter model and it's running on his company's slack Channel as like a participant and they're training they're
205. 37:21training this model they're fine-tuning this model using their slack interactions with it he said it's batshit crazy he said it's absolutely
206. 37:33insane it'll just lose its mind it it cusses like a sailor sometimes it's like it's it's like seriously not safe for
207. 37:41work but he's trying it as an experiment so so it's crazy like you know there there's again that it's very easy right now to sit on
208. 37:54the sidelines and point fingers and go yeah but who's writing the rules and and me start playing with the tools start
209. 38:04learning them rather than just take it at face value that oh these models are biased learn how they're biased do your own
210. 38:15tests you know check with prompting like you know I I see this happening on Mid Journey a lot that that people are
211. 38:21getting in there and the mid Journey models have gotten better over their model at being less biased because
212. 38:29there's feedback of people using it and doing it don't just sit on the sidelines and sort of point fingers
213. 38:34because there's no way that these here's the deal this generative AI stuff is so new that of course there's going to be problems with
214. 38:46it how they were trained is they were trained on an imperfect data set which is the public internet the public
215. 38:53internet is by definition biased to people who had access to technology that had whatever demographics the early geeks on the
216. 39:06internet were which did not skew super balanced to put it kindly but that's what it is that's what it's been trained on now can we improve
217. 39:17that we can improve that how do we improve that if we know how it's biased if we know the direction we want it to
218. 39:23go in and the only way you can do that is by playing with these things and understanding this one's better at that
219. 39:29this one's better at that here here's how I could combine these two things maybe maybe if you want balance you
220. 39:36figure out four or five different models that all have sort of different biases and for any given thing where you
221. 39:45want a balanced approach you use four or five different models or use 20 why not but the thing that like NWS at me like a like a
222. 39:58[ __ ] allergic reaction is when I hear well how are they doing this I don't know or or just they're doing it wrong
223. 40:07well then you go build a Frontier Model you go spend $10 million training a model and make it perfect we can write rules can have
224. 40:18dreams now how about an MLK GPT Builder day great yes put it put it together love it that would be awesome sounds sounds
225. 40:31fantastic I'll promote it I dig it I dig it I dig it dig it dig it dig it dig it dig dig dig dig dig dig
226. 40:41I heard big names saying that Lang chain and Vector DBS are Superfluous for AI thoughts are superfluous
227. 40:58no I don't think they're going to be required right like if if you look at so so for those of you that don't know Lang
228. 41:08chain and Vector databases allow you to take your own data and create a fine-tuned GPT model using you know whatever model as the
229. 41:24foundation and then and then with the ve Vector database you can be searching on your own data and and you can essentially put in an
230. 41:31infinite amount of data like however big your database is um that that as a that as a um a tool for where the technology is
231. 41:45today is super valuable because it's not easy to do that stuff right so you you'd have to write all your own code so Lang
232. 41:52chain makes it really easy to do that now you look at something like the gpts that that open AI just put out that
233. 42:00allow you to upload your own knowledge to them that takes a little bit of that functionality of Lang chain and Vector
234. 42:07databases and kind of makes it consumer friendly you know you want to make a GPT that understands your company's
235. 42:15documentation just upload it here right if you wanted to do that prior to gpts you had to run Lang chain but to say
236. 42:24that it's Superfluous is kind of silly like I think I think by the end of 2024 it will probably not be nearly as necessary
237. 42:35nearly as valuable but I would also Imagine That Lang chain and Vector databases and things like that are going
238. 42:42to evolve as new technologies come along and and they will probably augment the weaknesses of commercial products
239. 42:50that are launched um like they always have so what's the difference between all the the Llama models like 70b Etc so
240. 42:59it it it really is just size um so the the 70 billion parameter model um is is bigger in terms of just hard drive space how much you need to
241. 43:11run it and then it's bigger in terms of requirements to actually run like you need you need to be running it on
242. 43:20computer instances with big gpus and things like that and then so so they're all sort of in the same dire they all
243. 43:28kind of do the same thing it's just you know they are smaller and smaller data sets what's been really cool about the
244. 43:36opsource community is that they have been taking the different models 70 what are they the the seven the 7 the 13 the
245. 43:4644 didn't work I don't know if they've replaced that yet there was a 7 the 13 the the 40 and then the the 70 what the
246. 43:54open source Community has been doing is is they've been taking those models and they've been doing things like
247. 43:59optimizing them because they have access to everything and the weights they're doing experiments like what if we took
248. 44:05out 90% of the data in this model and made it really small does it still function turns out it does turns out it
249. 44:14actually functions pretty well so now there's like hundreds of variations of these models and there there are ones I
250. 44:22just saw one two days ago they just announced there's there's a new one that's designed it's a large language model
251. 44:30designed to be installed on a Raspberry Pi wait do I have my little Raspberry Pi here I think I do where is it yes this little Raspberry
252. 44:46Pi you can you can now install a functional multimodal large language model that that runs on a Raspberry Pi low power they cost 40 or 50
253. 45:01bucks and you can have a multimodal something that can analyze images can answer questions can write stuff now is
254. 45:10the thing that runs on a Raspberry Pi anywhere near CAD gbd4 [ __ ] no not even close but you don't need chat gp4 for
255. 45:22everything right if you just want something that can look at an image OCR it analyze it organize the information
256. 45:31for you and and give you that back that could be super valuable for a remote Village somewhere that that you know
257. 45:38wants to do a particular kind of task and and I think I think one of the things that we're going to see in 2024 and in fact the the
258. 45:48um the faret model that Apple just released is a 7 billion parameter model I don't know if they released all the
259. 45:55weights or everything I don't I don't know uh if it's open source I don't I I didn't pay a too too much of a too much
260. 46:01attention to it um but I think one of the things that you're going to see in 2024 um Google released Llama Or Not
261. 46:12llama uh um Gemini Nano which is designed to run on devices without needing an internet connection the open
262. 46:20source community's got a bunch of these things that don't need an internet connection you know apple have one we're
263. 46:26going to start to see 2024 I think is going to be the year where um Hardware capability goes up and Hardware
264. 46:34requirements go down for these models and you're going to be able to run very very powerful large language models with
265. 46:41no connection to the Internet so that's better for safety it's better for um speed responsiveness because you're not
266. 46:48having to rely on an internet connection and have it processed out there and then be sent back what raspberry model I I
267. 46:55think it's I think the we had so if you go look at um go look at the Joker if you go look at the um AI Salon YouTube channel we had a guy
268. 47:11come in who who actually saw me talking about this on here um and he said he had successfully installed a large language
269. 47:18model on a on a Raspberry Pi I think he said it's an 8 megabyte um an 8 megabyte minimum but I think a 16 megabyte Raspberry Pi
270. 47:29was better or something like that but I think that they they've all what's what's nice about the open source
271. 47:34Community is they've got all the requirements they'll say this model is you know has been tested on the
272. 47:39following configurations so you can get yourself configurations that fit that um but I think the one that was just
273. 47:46announced two days ago is really designed to be run on really shitty raspberry pies so it's pretty pretty
274. 47:53sweet what's happening Todd Happy New Year my friend thank you so much thank you for your explanation
275. 47:59absolutely it's a false assumption that there is a diverse selection of AI advancement technology it's a false
276. 48:08assumption that there's a diverse selection of AI advancement technology I would well if you're talking about the frontier
277. 48:18models if you're talking about if you're talking about building models that require billions of dollars of
278. 48:26infrastructure and investment then I would agree with you if you go look at what's happening in the open source
279. 48:32Community right now I would say that it is also a fun a false assumption to assume that there is not a lot of
280. 48:42advancement happening outside of those big companies um I I appreciate your cynicism but I don't agree with it um
281. 48:54are we in a perfect place no but we're not in a perfect place in society we can't [ __ ] talk about basic concepts of democracy
282. 49:05without wanting to kill each other right now in America like I think AI bias is the least of our problem right now how
283. 49:13about the the [ __ ] bias in our country how about we get back to being able to talk to one another as if we were human
284. 49:20beings isn't that a bigger problem it's might of quiet in here Jed did you say something I hear
285. 49:42nothing oh that's cool down down in the mud has that raspberry pie case yeah I like that raspberry pie case it's cool
286. 49:48isn't it oh my God the Mickey pick is in honor of its transition into the public domain it is it is indeed multimodo llm
287. 49:55and every pocket in the future exactly and and like the software for these is going to get really cool like you're
288. 50:02they're going to tie it into your camera on your device so you'll be able to just you know I don't know take a picture of
289. 50:09something and have it analyze it and I don't know it'll do cool [ __ ] with it and it and all your data stays private
290. 50:15it's all yours I prompted write a story sad bird still sing it wrote a 7,000 word story oh that's cool interesting AI
291. 50:23has only one rote that has been released to the public no that's not true it's not true stability AI released theirs it's AI
292. 50:34only has one route that is popular the Llama one because they they released it for commercial use um but there's stabilities out there
293. 50:46there's a couple of Open Source projects on it there's a there's an open- source project right now where they're trying
294. 50:51to build floating GPU Farms that they can can put on barges and Float outside of inter in international waters so no
295. 51:00country can um can control them and they're trying to raise enough money to to build their own military to protect
296. 51:08these things so there are efforts there are efforts right now to be able to build Frontier models outside of the
297. 51:19current INF power infrastructure so again I would say educate yourself there there's a lot going on right
298. 51:26now um I was trying to install my GPS ways in a Raspberry Pi that's awesome what Raspberry Pi model I'm using Da Vinci 3 on a pi python that
299. 51:39accesses various API Keys Kyle remember the promise software no I don't I don't know what that is let's see if chat GPT does I bet it
300. 51:52does uh uh uh or or do you mean the promise software wait what was the promise
301. 52:12software the promis software refers to a notable legal case management of software system that was developed in the 80s by Bill
302. 52:24Hamilton it allowed for efficient management analysis of large amounts of data which were crucial and compx legal
303. 52:31is this what you're talking about tracking and Reporting customization Innovative for its time software gained
304. 52:37some notoriety due to controversy surrounding its distribution and use there were allegations that the software
305. 52:43was illegally appropriated and sold to foreign governments leading to a series of legal disputes and conspiracy
306. 52:49theories these controversies have often overshadowed the technical and functional aspects huh now didn't know
307. 52:55anything about it that's interesting a private military for GPU barges sounds hella dystopian I have to admit but cool
308. 53:02yeah listen I I'll tell you what like I don't care if if if they actually raised the money to to to build you know 10,000 GPU barge
309. 53:15farms and and and they raised their own military and some other country wants those things not to exist those things
310. 53:25are going to not exist it's like you know it's like bad enough like going after one country's military but if
311. 53:32you're going to go after like China and Russia and the US and it's like but whatever listen I you know more more power to people it
312. 53:44like what I think's going on right now is that is that people understand that these Technologies are are
313. 53:55profound and because of that all of the concerns that everyone's talking about are valid to be talked about like we
314. 54:01should be talking about them we should be talking about who writes the rules we should be talking about how you debias
315. 54:07things we should be talking about alignment and safety um it's it's the the primary thing that um you know Andre
316. 54:15karpathy came back to open a AI to run super alignment so super alignment is are the are the ai's imperatives aligned with human
317. 54:27imperatives because if the ai's imperatives are not aligned with human imperatives at some point they're going
318. 54:34to go after humans because they threaten them so like that alignment problem is massively important massively we're just a little early for
319. 54:45you know all the all the doomsday because it's like we're not there yet but the fact that this is all a con
320. 54:55ation I think is a really good thing when people say AI will destroy Humanity I answer Humanity's been doing that for
321. 55:00a long time for a millennia I yeah right right now we we are the greater Threat by far absolutely absolutely Papa JJ
322. 55:12humans are the [ __ ] it's like well the AI hallucinates have have you talked to a person they [ __ ] lie and scheme and connive it's like
323. 55:25so like the the AI I'll tell you what I have had more nuanced powerful conversations with AI than I have with
324. 55:34humans in in the past year it's it's it's a it's a different thing you want Skynet that's how you get Skynet well I
325. 55:44don't want I don't want anything I'm just telling you what's going on the robots aren't going to kill us who
326. 55:48controls AI controls the world that's the I'll tell you what one thing I agree with you on Spike is is that I think that in the next 3 to 5
327. 56:00years so so the next year or two is just going to be the technolog is going to get more and more powerful and then we're going to start to
328. 56:09see attacks on the current power structure and not attacks like like hacker attacks I mean like business
329. 56:17models are going to be up and upended and and just weird shit's going to start start to happen where smaller
330. 56:23competitors are competing with really big companies and it it it won't make sense and then I think in the three to
331. 56:30five year range we're going to start to see the centers of power get threatened and there's going to be all
332. 56:36sorts of wild death throws it's going to be crazy can we get back to making funny turtles absolutely we can absolutely get
333. 56:47back to making funny turtles yeah this is a pretty heavy heavy New Year's thing um we can train ai's Behavior faster than
334. 56:58humans be nice right yeah be nice to your be nice to your robots because they could help you someday if I could only
335. 57:05get out of the house 99% of the world wants nothing but peace and just to live their lives out yep I agree with that I
336. 57:13trust AI more than a lot of people even if AI hallucinates it doesn't currently use manipulation and lies you can get it
337. 57:20to use manipulation and lies but it it it uh it can you can um you can also get it to be reasonable and it's it's hard you know what's funny
338. 57:36is one of the things that's caught me off guard about AI is um you know there are certain words in the current political situation you
339. 57:47literally just can't say the word it's one of those words that's been weaponized by one side or the other and you can't say it
340. 57:56to humans because it just you can just see you see their shoulders go up you see the you know people get emotional
341. 58:03when you say certain words right now just because of of how politics has been weaponized um one of the things that's
342. 58:10caught me off guard is I'll find myself like editing what I'm tying typing into chat GPT and I'm like oh I don't need to
343. 58:17edit myself here I can actually say that word and then and then I'll realize oh you know I actually don't I've never
344. 58:24never really thought about that word I just know I can't say it and then I'll go to chat gbt and say hey talk to me
345. 58:30about the concept behind this that or the other and it gives me this incredibly nuanced response and I'm like
346. 58:35oh I understand it now but it's like you can't have that conversation with a human it's weird we're in a weird spot
347. 58:44so so yeah i' I've I have found um actually some anxiety lowering effects of chat GPT um man alive I trust AI more than a
348. 58:57lot of people AI educated innocence wait AI educated innocence at our level of exposure now I don't quite know what that means Mary
349. 59:10Mary all right let's get off this this heavy heavy stuff let's play [Music] um
350. 59:30oh my God that's really [Laughter] funny Cosmic lover 7 Kyle what do you think about Rocco's
351. 59:45basilisk it's a great question it's a fantastic question it's so good in fact dear chat GPT what is
352. 1:00:02Roco basilisk okay Rosco's basilisk is a thought experiment that emerged
353. 1:00:16online with the online community less wrong it posts a hypothetical scenario involving the all powerful artificial
354. 1:00:23intelligence from the future that could retroactively punish those who knew about the possible existence but did not
355. 1:00:30help bring about its existence oh that's interesting the idea is based on several assumptions firstly it assumes that the
356. 1:00:36AI could eventually possess infinite power and near near omnipotence second it assumes AI would want to maximize its
357. 1:00:45chances of coming into existence oh this is interesting it combines the idea of decision Theory ethics and AI Theory
358. 1:00:55it's a thought experiment more of a philosophical exploration yeah this is this is outside of my you know I didn't
359. 1:01:02go to school for philosophy I'm not a I'm not a deep intellectual so this kind of stuff is like I find it fascinating
360. 1:01:08but I don't I don't have opinions on it because I don't know I don't know enough about it this head cold [ __ ] uh uh uh uh uh
361. 1:01:25Paula Elby that's why I don't that's why I'm here I don't want ro ro Roco basculas to get me that's it well so be
362. 1:01:34nice to your robots and show up at the AI learning lab so that there's there there's a record that you are here
363. 1:01:39that's awesome see it's you're you're hedging you're hedging your future bets we're trying to B bring good AI into existence I like
364. 1:01:51it basilisks turn people into stone oh man it's just that AI will punish people that were against [Laughter] it I want to be ai's
365. 1:02:07friend this actually yeah that's what this channel could just turn into like a a positive affirmation Channel AI is awesome AI swell we like
366. 1:02:19AI you can make money with saw a guy typing and the image rendering was updated with each click do you know
367. 1:02:30what he was using there's a lot of those S we can go play with one of those sorry hang on driving me
368. 1:02:46crazy there's one called Korea kore.ai and there's also one in Leonardo a so it's called realtime generation it's
369. 1:02:57based on um stable diffusion XL turbo came out about a month ago so a lot of softwares are incorporating it right now so if you run realtime
370. 1:03:12generation I think I have to sign in hang on let me sign in here and then where is the
371. 1:03:30oh that's the strength there okay so so this is it create. AI real realtime image generation so if you know
372. 1:03:41if I move the thing on the left move the thing on the right then the the image on the right just you know
373. 1:03:55and then I can change the prompt like the prompt right now says a pink frog on top of a Crea is not safe for Tik Tok I know that's a cool
374. 1:04:11one that's actually really cool see it made a door and then you we can put a little Sunset out in the door or a little planet
375. 1:04:26oh and it just stopped uploading come on it looks like it crapped out the other one that you can use
376. 1:04:40that's got real time in it right now is Leonardo or or if you're into stable diffusion you can set up your own
377. 1:05:02so right here realtime canvas it's this it's the exact same thing see Bo there's the thing I don't I think I've got zero prompts in there
378. 1:05:15right now that says Dynamic so we'll do let's see we'll do photography like it turned that thing into an eyeball which is kind of
379. 1:05:31[Music] cool r r so that's that stuff that stuff's pretty cool um and and you know what's
380. 1:05:47interesting about the realtime stuff is um it's it's a completely different skill right like if you got good good at Mid
381. 1:06:01Journey where you got good at structured prompting and you know learning the style attributes and learning what they
382. 1:06:09did and and and got good at you know mixing images and you know there's there's all sorts of ways you can use
383. 1:06:15mid Journey well this is a completely different skill and I I think that's what what what 2024 looks like to me is it it's almost
384. 1:06:27like what what what chat GPT taught us in 2023 was that there's a tool that exists that you can put in a minimum
385. 1:06:42amount of effort and get out of disproportionately um large response right write me a par paragraph of a
386. 1:06:55novel boom here's a paragraph of a novel write me an outline of a novel here's an outline of a novel write me the novel here's the novel
387. 1:07:02right whether or not it's good irrelevant but there's now a tool in the world where this much input results in this much
388. 1:07:13output and I think what 2024 is going to be is there's about a billion different ways to provide that input and some of the way like like it
389. 1:07:27could very well be that sometime in 2024 that that that could become the input for an AI model and not an AI
390. 1:07:41model that's going to generate music like that probably already [Music] exists could lead to an image or a a poem or a computer
391. 1:07:56program like and then who's going to get good at that if someone that can rip off a guitar solo that could then be turned
392. 1:08:05into some other thing like I think I think 2024 is just going to if you think 2023 messed with our brains I think 2024
393. 1:08:13is is going to be Bonkers it's going to be Bonkers let me take a PDF of a PowerPoint pre presentation I gave in
394. 1:08:291987 and used that as the input for some theater project like I don't know it's possible I mean if you look at the if
395. 1:08:42you look at the Gemini Ultra videos where they're taking two pictures you know unrelated pictures drawing a box
396. 1:08:51around them and then prompting tell me the relation relationship between these two images that's a completely different
397. 1:08:57way of thinking than here's a Google box come up with a prompt for a predictable answer like right now this is a fascinating thing right
398. 1:09:10now I think we don't know how to use AI at all because we're treating AI like a Google search
399. 1:09:26you get your prompting right you get back a predictable answer that's that's the Holy Grail that might not be the
400. 1:09:32Holy Grail that might not be the Holy Grail the Holy Grail might be Jazz I think Korea the the realtime
401. 1:09:43image generation things is a lot more like jazz like there are going to be people that get really good at putting
402. 1:09:50shapes on a screen or drawing things and and having the output you know morph in real time I just saw um I think uh Robert
403. 1:10:01scobble put a thing on Twitter last week there's a there's a uh there's an art installation right now where you
404. 1:10:10dance in front of this wall and project it onto the wall it's rendering a dancer that uses your your body position to
405. 1:10:21make the art in real time so so I don't I don't think we have a [ __ ] clue what the inputs are and and and how
406. 1:10:33you manipulate them and then what the outputs are like that's going to get really blurry in 20124 and I think it's going to get
407. 1:10:42confusing I think it's going to get like well I thought I had this [ __ ] figured out yeah so did I right this this has
408. 1:10:50been my the the last month for me like December for me was weird because when when gp4 went fully multimodal when they when they when they
409. 1:11:01sort of Incorporated all all into the single box there weren't different modes that you switched between I realized I don't know how to
410. 1:11:09use it anymore and then when I saw the Gemini videos that came out I realized oh [ __ ] I don't have a [ __ ] clue at
411. 1:11:18all when these things go true multimodal what that means is any form of communication can be used as input and
412. 1:11:29output video in text out music in video out video in video out image to like all of it what does that mean I don't know what
413. 1:11:41are the interfaces going to look like they're going to be weird there's a thing that meta put out called segment anything where it can
414. 1:11:52look at a video in real time and segment every little object in every video frame so imagine an interface where
415. 1:12:01you're watching a movie and at any point you can just grab anything on screen and and sort of drag that into some other
416. 1:12:08interface and use it as an input for that other interface you know what kind of desk is that just grab the desk and drag it over
417. 1:12:20there oh put that desk in a 70s room with shag carpet boom there it is ooh let me drag that room back into the
418. 1:12:29movie I'm watching and I want them I want these actors acting that scene in in this room is that going to be 2024 no but the
419. 1:12:38foundations for that are going to be 2024 who's going to get good at that not the person that's really good at prompt
420. 1:12:50engineering can I create a Freddy Mercury GPT um You probably can it you might run into safety guard rails anytime you use
421. 1:12:59someone's actual likeness or their actual name you might run into um into safety guard rail [ __ ] where
422. 1:13:07they're like ah we don't want we're not really sure about doing that there buddy um but that's where again you
423. 1:13:13could play with open source stuff and go get the go get the non-restricted Llama models and do anything you
424. 1:13:21want now whether or not you get sued by the Freddy Mercury estate that's on you wait say that again segment what so
425. 1:13:30it's a it's a a piece of technology meta put out called segment anything I think segment anything meta
426. 1:13:47yeah is this really meta is this really from meta it is go to segment
427. 1:14:03d.com here look this is it's it's cuckoo it's cuckoo so it just so like here's your New York City skyline it it like identifies all the
428. 1:14:19object it segments everything sort of instantly like while someone's walking through their room it's identifying everything
429. 1:14:29in the room right here's all the cats it's like identifying all the different cats and isolating them
430. 1:14:38programmatically in real time so you're going to be able to just grab an object turn it into a 3D object and 3D print it
431. 1:14:45if you're in here Vicki this this shit's coming zero shot generalization meaning it can break apart any object in
432. 1:14:55any image with with no training zero shot means no training the fact that meta so this says research by meta AI the fact that meta
433. 1:15:07put this out as a standalone website segment.com that says to me this is this is this is something so you know back to
434. 1:15:17what can you do with this stuff well if you're a programmer and a geek you could take this technology and roll that together with
435. 1:15:23you know some of the large language model technology and roll that together with some of the image generation
436. 1:15:28technology and roll that together with some of the music generation technology right you start to get
437. 1:15:35into I mean we just we start to get into a world where everything in our world becomes an input that can be manipulated
438. 1:15:43in some way we use something like this for insurance claims yeah exactly this is just this is the identical stuff
439. 1:15:50Vicki but apparently the the segment anything is just it's instant and much more granular than than things before
440. 1:15:58it but again I don't I don't know a ton about this but I but what I do know is I've seen language or uh you know visual
441. 1:16:05segmentation models before but they're not this granular and they're not this immediate and and meta put this out
442. 1:16:12there explore the data set download the full data set 11 million 11 million images 1 billion plus masks segment anything a new AI model from
443. 1:16:27meta that can cut out any object in any image with a single click home Inventory exactly oh yeah yeah yeah it's perfect for home
444. 1:16:37Inventory right yeah exactly you just you just take the video of the room and it just grabs the [ __ ] and then it can
445. 1:16:43automatically identify it and then it can automatically value it and just make your insurance document there you go
446. 1:16:55you mods are awesome by the way I am loving what's going on in the mod Channel right now oh it's all comedy all the
447. 1:17:14time hey Mickey oh man alive man alive man alive Apple Vision Pro and AI will make huge leaps next year yep that's going to be
448. 1:17:28another thing that we're going to see some some interesting stuff I think I think next year we'll see some stuff
449. 1:17:34from Apple that on the AI front that's going to be good my guess is around June time but maybe before then you could
450. 1:17:42have your own home entertainment run with AI artists and Holograms AB yes I we're we're entering we're entering an era of hyper
451. 1:17:55personalization where we're all going to have access to tools where we can just say hey I want a Spotify CH channel that
452. 1:18:05is only music that I like but it's all original music and you can take my um data from my Apple watch and if you're playing music that
453. 1:18:21doesn't make me happy my app watch will know so so recompose it on the fly so it's always giving me music that keeps me in
454. 1:18:33state that sound creepy maybe did it sound inspiring to someone is it going to get weird
455. 1:18:48mhm this Channel's going to get weird well this channel is already weird you can make money J what's going
456. 1:18:57to get weird is what's going to get weird is none of us are going to know what to do with what comes and and you know it's what's going
457. 1:19:09to there's a there's a cool weird to it I guarantee you that there PE there are people right now that come to this
458. 1:19:17Channel and they're watching some of the people in here that are embracing this stuff and doing really good with it
459. 1:19:23and they go to try it and they can't get good results and they're frustrated and they're like ah but I'm not good at
460. 1:19:28prompting or I'm not good at this or I'm not good at that those people are going to start to shine there's going to be technologies
461. 1:19:37that come along that like like I think I think that there are going to be people within the AI Salon that get really good
462. 1:19:43at the realtime image Generation stuff and and they they almost use no prompting that that they get really good
463. 1:19:50at this sort of dynamic fluid creation and that's going to go from creating images to creating movies and
464. 1:19:58and and doing them in real time and doing storytelling the people are different skill sets are going to excel at different kinds of
465. 1:20:08inputs and I find that really exciting because right now it's kind of limited just because of where we are in the
466. 1:20:18evolution of the technology the inputs are really limited right now so they favor a certain like like right now prompting favors word
467. 1:20:29people if you're good with the words if you're good with logical construction you're probably pretty good at the prompting
468. 1:20:41stuff you might not be as good at jazz image generation someone who's super visual that is shitty with words and maybe
469. 1:20:52maybe has dyslexia or has some sort of weird ass you know cognitive thing where words words in them are not friendly but
470. 1:21:02they're great with shapes and they're great with um Dynamic things they're going to be genius where you're
471. 1:21:16not I find that tremendously exciting I it's one of the things I'm most excited about with AI is is the
472. 1:21:25accessibility that that it's going to provide to just people of of All Sorts that are cut out for whatever reason
473. 1:21:34they're blind they're deaf they've got you know neuros spiciness they've got a brain injury they've got whatever it is
474. 1:21:40they've got paralysis due to a spinal cord injury all all of just because the inputs and and processing and
475. 1:21:50outputs are are are just going to just they're just going to become this gooey mess that like the possibilities are just going to be
476. 1:22:01crazy we are the product or not well here's the deal Spike if if the if the technology is free you're the
477. 1:22:17product guaranteed promise that that that is for sure um AI combined with augmented reality
478. 1:22:33well it by definition it has to right like the augmented reality stuff can't happen without the AI stuff but can one
479. 1:22:40use GPT itself to generate prompts for itself based on more tur prompts it's happening right now ey on China look
480. 1:22:48look look look look look look look look how long's he gonna say look he doesn't seem to be
481. 1:23:02stopping [Music] archive where is why do those things
482. 1:23:17disappear I do not understand hang on zip .zip zip archive are there any
483. 1:23:32here I got some weird ass [ __ ] going on my computer hang on a sec I'm going to go to
484. 1:23:49kaggle is he even paying attention to us are we just watch watching an old man try to surf the Internet is that what
485. 1:23:55this channel is is this this is what it is now is do PE you people find this entertaining PE people find this
486. 1:24:04entertaining oh they don't they don't why why are they here they they can't leave is he keeping them here we nobody seems to know okay
487. 1:24:20uh I'm I'm pretty sure he's touched he he seems he seems to be a bit touched okay so I on China here's the zip archive I'm going
488. 1:24:35to say what is this question mark that's about as sparse a prompt as you could get now watch what chat GPT does it
489. 1:24:45appears that you've uploaded a file named uh to better assist you uh would you specify what you'd like me to do um look inside inside wait
490. 1:24:57inide and tell me what it is and what is interesting about it okay it says analyzing well what's that mean well if we click on this it's
491. 1:25:13writing and executing its own python code so it imported a a python Library called zip file and another one called
492. 1:25:21OS to look inside the zip file and then it says oh there's there's two csvs in here about taxi data would you like more
493. 1:25:29detailed analysis for example um yes do what you think is best and then visualize it for
494. 1:25:45me this this is the sneak peek into autonomous agents where you just give autonomous agents or an autonomous agent system a goal and it
495. 1:26:00is going to go do [ __ ] for you like this is now opening here here's the panda Library it's it's opening these files
496. 1:26:10looking at the data in them analyzing the data within them the second data set seems to be a summary or analytical data
497. 1:26:18set with these columns to visualize the data we could create a few plot points for a taxi trip data a scatter plot
498. 1:26:26could work for the predictions so now it's doing whatever the [ __ ] it's doing Matt plot live so now it's coming up
499. 1:26:33with some visualization for us and there it did a scatter chart of trip distance over total amount and then this is mean distance
500. 1:26:49and predicted fair and it looks like there's some weird data at the $55 Mark which maybe that's a flat fee thing or
501. 1:26:58something like that I don't know or it's or it's anomalous data I don't know so so yes I mean yes it like
502. 1:27:09we have it right now crazy right but can one use GPT itself to generate prompts for itself yeah and in fact ion China so so go play with
503. 1:27:19that asking chat GPT to write its own prompts there's a there's a an article that or a study that Ethan mik talked
504. 1:27:26about that chat GPT writing its own prompts it's better at writing prompts than humans are they like they did a study on it and chat GPT 4
505. 1:27:36won and and one of the prompts that it came up with this is [ __ ] wild one of the prompts that chat GPT came up with
506. 1:27:44to prompt itself that had the highest efficacy it was like 87 whatever per on whatever their scale was um was the
507. 1:27:52prompt take a deep breath and think step by step that's a prompt that chat GPT came up with for itself and if you tell Chachi PT to take
508. 1:28:03a deep breath and think step by step your results get better why I don't know it's crazy it's crazy it's crazy
509. 1:28:13it's so crazy it's so crazy oh hey Marco hi it's a c the the mods at this point the mods have
510. 1:28:29realized I'm just going off on tangents tonight so they're have they're having a whole party off on The Mod channel it's [ __ ]
511. 1:28:38hilarious um is the recording of the webinar available it is if you go to the salon. a the second link there is
512. 1:28:49to the YouTube channel now I only have a my Tik Tok things as soon as I get the links for Ann's Zoom meetings I'll
513. 1:28:57incorporate them into that playlist but if you if you go there the second link is to the is to the YouTube
514. 1:29:07um the YouTube playlist for the GPT for good um event and then and then I'll I'll update it as soon as I have Anns
515. 1:29:18thingies Ubi will be our way but I beg to differ what do you beg to differ on uh I'm watching something on the
516. 1:29:27Irish mob while I'm watching you sorry that's okay I think we can probably learn from the Irish mob with this AI stuff too AI
517. 1:29:37will out Shide humans on all fronts I agree with that entertainment will be all we seek when we have nothing that I disagree with
518. 1:29:45vehemently vehemently I think that most humans want to contribute and want to connect I think the dis this the
519. 1:30:03dystopian the dystopian view that when when we have infinitely personalized entertainment that will all just numb
520. 1:30:10out and go away I think there'll be some percentage that do that but what I'm experiencing in this community is that kind of the opposite
521. 1:30:21is happening the opposite of people going oh if these tools are creative then I'll be less creative if these
522. 1:30:27tools um think for you then I'll think less what I'm experiencing in this community is that the people that are
523. 1:30:35super engaged with AI are getting more creative more engaged more intellectually stimulated more educated more driven to contribute not
524. 1:30:50less if we have tools so so here's how I think about it what I've learned I I've been on the planet for a while see I've earned
525. 1:31:06this this the larger forehead was earned um I just lost my train of thought I
526. 1:31:22made a joke about my my forehead I lost my train of thought um what was I talking about oh well it doesn't [ __ ]
527. 1:31:29matter all right ai ai will outshine humans um it was something about that AI outshining humans oh oh us working more
528. 1:31:38um oh I got I I know what it was okay um one of the things I've learned over the years is that just me personally I have some amount of
529. 1:31:50hours of Creative Energy per day it's finite and I can't and some days it's an hour and some days it's 7 hours and some days
530. 1:32:02it's 10 minutes and some days it's zero but it's a finite amount per day and my brain doesn't care what I'm using that Creative Energy
531. 1:32:16on prior to AI a lot of my Creative Energy was put Into The Drudge work right so if I had an idea for a
532. 1:32:28game about UFOs and disclosure which I did I would be like oh great I have this idea for a game but I don't really know
533. 1:32:40Game Theory all that much and I don't know all the different kinds of games so now I've got to go into research mode
534. 1:32:46and I'm going to go to Google and I'm going to look up games and Game Theory and how games work and it's going to
535. 1:32:52take me to all these documents and then I'm going to have to read those documents and then I'm going to actually
536. 1:32:56have to understand those documents and then at some point in reading that document I might realize oh well that
537. 1:33:03wasn't what I wanted to learn and then I go over here and I and so I would spend 45 minutes or an hour spending Creative Energy trying to
538. 1:33:14learn enough of the concept of something to be able to move to the next step for my game idea what happens now with AI is
539. 1:33:24I have this idea for this stupid UFO game and I go oh what I heard of a game called this how does that work it goes
540. 1:33:33oh well that uses a theory called this and it works like this and here's the game mechanics I'm like oh well I like
541. 1:33:38Monopoly could you combine Monopoly game mechanics with those game mechanics oh sure you could do that it would look
542. 1:33:44like this here's the rules here's how the players would do it' be this many people you would use cards and you would
543. 1:33:49use Dice and and so all of a sudden the the 45 minutes of Creative Energy that I had on whatever that day was that was
544. 1:33:57blown on doing shitty Google searches that work is now done in like 30 seconds and I still have 45 minutes
545. 1:34:07to be creative and to be engaged so what I'm finding with AI is that it's automating out the shitty work and it's allowing me to just
546. 1:34:19operate in the large forehead portion of right the the the the part of my brain and creative processing that's at a higher
547. 1:34:31level that's [ __ ] amazing it means you can do more faster better it's not just an efficiency tool it's like a mind expansion tool
548. 1:34:46because you're not spending your Creative Energy on non-creative [ __ ] that's what it is is for me I'm glad I found my train of thought
549. 1:34:55again that was that was one of those senior moments where it's just [Laughter] like oh man increase your internal fan
550. 1:35:06RPM yeah exactly yes I love what I can do faster now with AI but this is a small window right where where where the AI is still our
551. 1:35:18our servant right where where we're we're still up here and the AI is down here I agree with that but I I don't necessarily I I don't
552. 1:35:28necessarily I I just don't buy that humans are just going to go oh the AI can do everything great I'll check
553. 1:35:36out maybe what it can mean is I can do more of this that I enjoy and I can get good enough with it that other people enjoy it other than
554. 1:35:51the dog it will become novel socialism don't know what that means earned experience cognitive Surplus yes
555. 1:36:05exactly we are we are entering an era of cognitive Surplus human beings being more engaged and more creative and will people seek purpose in
556. 1:36:16the future wait humans being more engaged and more creative and will people seek purpose in the future I
557. 1:36:23think they will I like I don't I I don't see human beings all of a sudden stopping seeking purpose and meaning and
558. 1:36:31contribution and connection like I don't see that going away if the machines can do the shitty work maybe we'll we we will collectively
559. 1:36:40be in less shitty moods because we've fried our brains on the non engaging stuff so maybe we'll be be better people
560. 1:36:52maybe we'll be better con connected maybe we'll be able to contribute at a higher level humans will go back to human
561. 1:36:59interactions being more vital I totally agree with that there's a reason that authenticity was the word of the year for the Mariam Webster
562. 1:37:12dictionary was the most searched word in 20123 authenticity why people crave it covid [ __ ] us up right covid removed us from one
563. 1:37:29another and then and then the threat of AI the robots are going to take over I think puts people in a in a weird existential place where I
564. 1:37:44think the value of authenticity the value of human connection the reason I started the salon was I wanted to be around
565. 1:37:53people who are trying to figure this [ __ ] out because I know when people are trying to figure [ __ ] out they get in
566. 1:37:58they get in an amazing place they get in a creative place they get in a a human connection place because it hasn't been figured out
567. 1:38:10yet because AI hasn't been figured out yet we can't just sit back and say oh I'll just go to Stamford and take a
568. 1:38:17class on that and I I'll just do that remotely we're we're all having to figure it out and so so it's like I'm like hey did you
569. 1:38:26figure that out yet no I didn't figure that out yet did you figure that out yet no did did Eric figure that out he seems
570. 1:38:31to have figured part of that out oh let's go talk to Eric hey Eric did you figure this out well I figured this little piece out but
571. 1:38:38I think that's stupid no that's not stupid that's what I was trying to figure out well we figured this part out
572. 1:38:43we think that's stupid and then Eric's like no no that's the part I couldn't figure out oh yay you did it and then the three of
573. 1:38:53us run off and start a company today I did my research to do a factory reset on my printer awesome saved me a ton of time exactly like
574. 1:39:02think about how much time you would have spent if you were Googling how to do that factory reset where you go to site
575. 1:39:10after site after site that's some [ __ ] up sales funnel that doesn't actually give you the information you
576. 1:39:16need and you just get frustrated and frustrated you're like [ __ ] it I'll reset my printer tomorrow and then it's
577. 1:39:22a week later and then you still don't have a printer and now you're just [ __ ] angry and bitching about stuff
578. 1:39:27well now you just ask AI how do I do a factory reset on this here you go and maybe it hallucinated and got it wrong so then you say Hey you
579. 1:39:39hallucinated and got it wrong Could you actually search the internet this time sure here you go until we had computers
580. 1:39:45many non-verbal autistic people had no way to talk what's possible for them with AI um oh oh my God so massive
581. 1:39:54massive break we're going to see massive breakthroughs so there's about four or five months ago there's a there's a uh a
582. 1:40:03gentleman in Italy that's been paralyzed for 12 years from the hips down uh from a motorcycle accident and so they took
583. 1:40:11generative AI they scanned his brain with fmri they they interpret they they told him to think about moving his legs
584. 1:40:21and so they trained a large language model or a machine learning model on when his brain does this it means he's
585. 1:40:29thinking about moving his legs okay then they took those signals and with a Bluetooth connection they went from his
586. 1:40:37brain and they sent that data down to electrodes you know in his legs and the guy's walking and so they're doing that with
587. 1:40:49imagery so they're they're they're doing real-time fmri scans where they play a video for someone and they're being able
588. 1:40:56to recreate the video that they're seeing they're doing it with images they're doing it with language so I think we're going to see
589. 1:41:04massive breakthroughs in accessibility yeah non-verbal autistic people like I don't I don't know enough about how it
590. 1:41:11works but basically the way these the way the Transformers work is that they don't care what data you train it on so
591. 1:41:19if you can get the data out of the brains and train it to recognize patterns then you can process it with
592. 1:41:29these same tools that chat jbt uses and that mid Journey uses and that the music generation tools the how the music
593. 1:41:39generation [ __ ] works is they just train it on music data rather than word data it's the same but it's the same
594. 1:41:48basic processes it the the tool doesn't care what the data is it's just going into some weird mathematical
595. 1:41:56space and then when you put in whatever the input is the prompt or the visual input or whatever that is it creates
596. 1:42:04some mathematical probability and it says go into the latent space they call it find the points in Thousand
597. 1:42:13dimensional math space and come back out and will reassemble the music data or will reassemble the brain data and just [ __ ]
598. 1:42:24works this is that weird thing where they don't understand why it works or how it works they just know that it
599. 1:42:33works there's these things called emergent behaviors where it just starts doing [ __ ] they don't that they didn't
600. 1:42:38train it on authenticity is a higher vibrational State than love is oh yeah I saw that Debbie that was that's one of those those things
601. 1:42:49authenticity authen ticity it like really really really impacts people in a in a it just it's in a different way it's why these
602. 1:43:00guys you can make money feel so [ __ ] scummy and weird because they're scummy and weird it's not authentic so yeah totally
603. 1:43:17agree GPT for good Saturday kind of proved we would get creative even when we didn't have to that's interesting
604. 1:43:23Genie in a Bottle yeah I the I think one of the things that blew me away about the GPT for good event was how many
605. 1:43:30people came to not just multiple sessions but like a lot of the sessions and like like when I imagined it what I
606. 1:43:40imagined was someone would come if they hadn't made a GPT and they'd make a GPT and they'd be like cool made one awesome
607. 1:43:48or if they' made a GPT maybe a come and do another one cuz they hadn't done one in a while but no what ended up
608. 1:43:53happening was I'll go to Kyle's thing oh then I'll go to Ann's thing then I'll make one then I'll make another you know
609. 1:43:59what now I'm on a roll why don't I just make 10 of these like where did that come from it just I it just
610. 1:44:07did people want to contribute people want to connect so there was something about that event that brought people
611. 1:44:14together in this way that was way beyond the tool like the tool was just the tool we're making shitty gpts that may or may not
612. 1:44:25survive the next 3 months right right they might be the biggest thing in history or they might be something that
613. 1:44:33remember those things called gpts whatever happened to them I don't know but but it wasn't about the gpts at all was it it was
614. 1:44:43about the connection and the the vibe and the and that authentic thing that was happening to today I read about camera
615. 1:44:51companies cooperating to create a verification system for photographers good I think that's great I think that's great I mean I listen I
616. 1:45:01think I think we are we are very very quickly entering a time where having unverified content connected to verified humans is
617. 1:45:14going to be it's it's not going to be sustainable we're going to need to be able to say oh that video was made by
618. 1:45:23Kyle and I and I can verify that and and it hasn't been modified like we're going to need that increasingly if you're going to be here
619. 1:45:33I'm in oh awesome thank you two block Tom like the name humans will work enough to basically cover the cost of being
620. 1:45:44provided for I didn't see the rest of that how is your day night going uh it's going okay I'm still I'm still still
621. 1:45:50sort of dealing with the co Remnant [ __ ] but I feel I feel okay I feel all right um Vicky Baptist she has to get up early she's done
622. 1:46:07playing if you're still here Vicky go sleep I'm gonna do it soon too what time is yeah it's getting there all right
623. 1:46:14what if a AI was embedded or linked to a person's brain as they passed and it captured the experience yeah ion China I don't
624. 1:46:31yes I I mean one of the things that we saw I was talking to to an about this uh Ann Murphy um one of the things we saw
625. 1:46:40we saw multiple gpts created um for Alzheimer causes I lost my dad to alzheimer's um the ability to capture stories before
626. 1:46:54they're gone the ability for Alzheimer's patients to connect with people that they know or knew and have those people
627. 1:47:05be infinitely patient infinitely there 247 um is going to be huge and then doing that at the brain level where yeah like stuff like
628. 1:47:19that stuff like that is not like the only difference between the the the M the fmri that is allowing the Italian
629. 1:47:29guy to walk and what you just described is just some researchers doing that research right we are quickly we are
630. 1:47:37very quickly getting to the point that these tools are going to democratize science creativity knowledge work all all of
631. 1:47:51it and so someone who was not a brain scientist is going to say hey I've got a loved one who's going to
632. 1:48:00pass they're really into this stuff they want to do this I'm going to figure out how to do it and they're going to take
633. 1:48:07their their Raspberry Pi and hook it into some fmri machine at the facility and they're going to be able to do
634. 1:48:18it and that's going to it that's going to revolutionize a something is chat GPT 4 much better than three yes I've been thinking about
635. 1:48:27upgrading okay here's the deal I do not work for open AI I do not I don't care I don't have a bet on any of these
636. 1:48:36companies I'm not invested in them I don't give a [ __ ] for me right now the musthave subscription in all of
637. 1:48:45AI is chat GPT plus it's 20 bucks a month it gets you all of what chat GPT is just from a mathematical standpoint
638. 1:48:54GPT 3.5 is 175 billion parameter model doesn't matter what that does it's just points of data that points of data that connect with
639. 1:49:04one another and give you answers gp4 is eight different 1.8 or 180 um billion parameter models so it's 1.8
640. 1:49:19trillion so it's 10 times bigger but it's also this thing called mixture of experts so it's not just 10 times bigger it's one Big
641. 1:49:28Blob what it is is it's multiple models that each specialize in different things so there's one model in gp4 that
642. 1:49:36specializes on programming there's another one that special specializes on language there's another one that
643. 1:49:41specializes on I don't know this kind of processing it whatever it is so when you do your prompt one of the things gp4
644. 1:49:49does it says which of these expert models are the best ones to solve this particular problem so it might pull
645. 1:49:56three of those things together and just do those so so and and then on top of it it's multimodal it's connected to the
646. 1:50:03internet it can see images it can make images it can hear it can talk it can do data analysis you can generate these gpts
647. 1:50:13which are personalized customized versions of chat gbt it's like the depth of GPT 4 is so mindboggling bogglingly beyond 3.5
648. 1:50:28and and it's it's also just bizarre to me that in the year that this 800 person company has been building chat GPT you know they've
649. 1:50:42been around for six or seven years so they've been building the foundation for it for a while but chat GPT has been
650. 1:50:47around for a year it is staggering to to me that a 200,000 person company like Google hasn't caught up Microsoft basically said well we'll
651. 1:50:59just use open AI they haven't caught up meta said well we're not going to catch up but we'll open source our [ __ ] because we're
652. 1:51:07close like apple hasn't done any like no one's caught up to these guys now gini Ultra Google's purported gp4 Killer
653. 1:51:19looks promis ing but it's still not out yet but yes I think the 20 bucks a month for chat GPT plus is the best deal in
654. 1:51:27software right now I think it is irresponsible if if you can afford it if you can't afford it you can't afford
655. 1:51:34it don't pay for it but it's 20 bucks a month don't have three coffee don't have two coffees at Starbucks a month and you
656. 1:51:43can you can afford that subscription if you want to be AI literate right now I think not deeply understanding what gp4 makes possible is
657. 1:51:55irresponsible like I think it's that important so yes yes it's different it is profoundly better and there's nothing else close to
658. 1:52:09it right now and it's what's [ __ ] up is I can't it is impossible to describe why it's better you kind of
659. 1:52:19have to to experience it which is really [ __ ] up a virtual reality world where your family is always there talking to you at the age you
660. 1:52:34remember yeah exactly just me and the trees yeah and know how to calm them and communicate with them they'll benefit yep Silver Fox
661. 1:52:45exactly thanks for the response Kyle hartfelt thank you very much I read the Sci-Fi story about using AI for
662. 1:52:52Alzheimer's patients yeah like we're not we are so close I I mean 2024 they're going to be off-the-shelf
663. 1:53:05software where it will be relatively trivial to make a virtual version of me that someone else can interact with as if they're talking to
664. 1:53:15me on one of these will it be as Dynamic is really talking to me no but is it going to be better than what it is now oh yeah and in 3
665. 1:53:31years it'll probably be largely indistinguishable I'm sorry about your dad Kyle thank you it makes so much
666. 1:53:40sense regarding your passion for your company yeah thank you so much appreciate that yeah my company's called Story vine it's a it's a video
667. 1:53:47storytelling company can you talk about creating gpts I've heard vaguely but I don't know how it works yeah so actually
668. 1:53:54go to there so this weekend this past Saturday Shadow Ranger we did an event called GPT for good it was a 24-hour GPT
669. 1:54:05challenge so every 3 hours we alternated between myself and Anne Murphy she taught on Zoom I taught on Tik Tock and
670. 1:54:12we taught people how to make gpts so if you go here right now the the first link here let me show that first okay so this
671. 1:54:22is just chat gp4 right I can go in here and I can say write me a song about quantum mechanics that uses a black and white singing dog as a
672. 1:54:38metaphor in a poem that rhymes wait in a poem that rhymes like Jay-Z all right so so chat GPT all these GPT
673. 1:54:54things you can ask them anything and they'll answer anything what a GPT is is a specific sort of mini version of that
674. 1:55:03the the best way to think of it is this if you think of chat GPT as like the the iPhone operating system iOS the
675. 1:55:10operating system the the phone can do anything you have like a little super computer in your pocket and apple makes
676. 1:55:15some apps like the phone calling app and the messaging app and the notes app and the whatever apps that they give you and
677. 1:55:22then there's this whole ecosystem of thirdparty developers that develop really specific little apps this one
678. 1:55:27will tune your guitar and this one will let you take notes in a different way and this one will let you call people on
679. 1:55:34the other side of the world for free right so all of those apps that's kind of what gpts are so I can go in here [Music]
680. 1:55:48and 7s muscle car machine so this is one I made that all this does is help you helps you make a poster of a 70s muscle
681. 1:55:59car to hang in your room if you were a teenager in the 70s or you're a gen xer that's just stupid I'm ready to design
682. 1:56:07my dream car so I clicked on that link and now the 70s muscle car machine says fantastic let's get started first which
683. 1:56:15model do you want uh GTO Judge great choice let's move on to color and finish sassy orange all right fantastic finally let's
684. 1:56:31talk about modifications um sleeper um race version that is a Resto
685. 1:56:46mod a Sleeper race version is awesome aome and then it should yeah now it's starting to make the image so this GPT is just like a really
686. 1:56:59mini pre- prompted version of chat GPT that came out of my imagination could you do this with just regular GPT
687. 1:57:10absolutely would you think of making a 70s muscle car machine that makes a cool poster of the judge and misspells it
688. 1:57:18that creatively no you wouldn't come up with that but you don't have to now you can just use my GPT so that's what they
689. 1:57:24are okay so if you go to the salon. it's going to take you to where is it I'll just go there the salon. it's going to take you to this um
690. 1:57:39this link tree the first link is to the directory of gpts that we made this weekend we we did this event over 24
691. 1:57:48hours our goal was 100 gpts we ended up making 151 well we didn't the people that participated in this event did and
692. 1:57:56so these sessions were like an hour long so in less than an hour and it it's really about 10 minutes is the part
693. 1:58:04where you actually make the GPT it's about 10 minutes um within an hour people were making gpts for nonprofits
694. 1:58:12um that that they believed in or that they were affiliated with we ended up making 151 of these so that's the first
695. 1:58:17link you should go check that out um the second link is the YouTube playlist of the um of the session so
696. 1:58:27right now there's just my sessions in there I'll be adding in and zoom sessions as well so you can actually go
697. 1:58:33walk and then the next link is the prompt sheet so this is the sheet that actually has the exact prompt for how to
698. 1:58:43make a GPT and exactly what to do and so how you make them it's it's really quite simple you in the left hand
699. 1:58:51it has you have to have gp4 so so you have to have a GPT Plus account and then there's a little button that says
700. 1:58:58explore you click on the explore button and the first option is make a GPT and then once you're here you just
701. 1:59:05follow the instructions in that prompt sheet and with like like seriously within 10 15 minutes you will have made
702. 1:59:13one it's it's it's quite remarkable so and you can go watch the videos they're all up there and we'll I'll be sharing
703. 1:59:20them once once we have that thing kind of dialed in all right anyone in chat have the link for the video where Kyle speaks Italian I'm
704. 1:59:34trying to think if that's up anywhere it's on LinkedIn hang on let me see Kyle Shannon attal
705. 2:00:11Italy learning Italian me neither anyway that's me that's Italian be ready to have your mind blown check so go to labs. hen.com in the
706. 2:00:24lower left hand corner click on the labs button and then go to video translate you upload a video file you choose a
707. 2:00:32language and watch this so this is a video from my storyvine platform what you'll first see is me introducing myself in English and then
708. 2:00:48watch hey it's Kyle shenon I'm the co-founder and CEO of Story vine I would love your input I'm inviting you to a
709. 2:00:54meeting I'm having on February 23rd so it's coming up uh about modular content and AI so let me dig into what's going
710. 2:01:07on so it's on it's on uh it's on it's on my Tik Tok Channel there's a couple of them on my Tik to channel so you can go
711. 2:01:14find them there um and then I think you can find them also on link I just searched Kyle Shannon Italian and it took me to that
712. 2:01:24one um here's another one and is important to them getting more efficient with how we create content and how we leverage
713. 2:01:40content behind me you see these little green boxes um that's how we think about stories at Story vine we break stories
714. 2:01:47down into little thank you Johnny 24 honey that vine I have never said Story vine like that in my life but I do now I
715. 2:02:01speak fluent Italian crazy right that's from haen by the way if you're wondering Mr it thank you for the the
716. 2:02:11ruse is very nice F this is why I insist on the smartphone being in you mix of iron and clay or not in Grant that are
717. 2:02:23taken I will use all Techni technologies that I can use without surgery yeah actually so a fascinating thing mantrid is
718. 2:02:34um so Elon Musk has his neural link right where you drill a hole in your skull or multiple holes in your skull
719. 2:02:41and then you this robot inserts the 60 electrodes in your brain uh because it's a it's instant access to to your brain
720. 2:02:51um some of the technology inside the Vision Pro goggles are essentially that without surgery there's there's one so there's
721. 2:03:02like 5,000 patents inside the Vision Pro goggles one of them one of them that I read about is is mindblowing okay so the
722. 2:03:13the way the interface works with Vision Pro is it's got four cameras looking at your eyes and then it puts this virtual
723. 2:03:23interface you know into your room right that's got buttons and when you look at a button if you pause when you look at a
724. 2:03:31button it it highlights it and and then with your fingers outside the goggles you do this to click well one of the things that the cameras
725. 2:03:44looking at your eyes are doing is they're looking at the micro dilations of your pupils and they know before you do that you're going to
726. 2:03:57click on an icon so as you're looking toward an icon your pupils do something that that show that you're getting
727. 2:04:08excited and you're going to click on that so what they what they actually start doing is pre-loading that software
728. 2:04:17before you click on it so that it's more responsive so like there there there's there's some [ __ ] like the they they have this chip in The
729. 2:04:31Vision Pro goggles called the R1 chip it's the reality engine or the reality chip it's basically just a machine
730. 2:04:36learning chip on device so they're not having to process this stuff like send it off to the internet have it processed
731. 2:04:44on a supercomputer and sent back it's being processed in real time like faster than you can it it knows what you're going to do
732. 2:04:55before you do it and it starts pre-loading the software so that is essentially the [ __ ] that Elon Musk is
733. 2:05:03doing like like Elon might argue well well the actual neurotransmitter would be you know 14 micros seconds faster
734. 2:05:12than the than the other than the other [ __ ] thing but you know there we're going to have very sophisticated
735. 2:05:20technology that that does a lot of this stuff without surgery so it's it's I'm we're we are entering weird [ __ ] times people weird [ __ ]
736. 2:05:33times Sam in UK listening that Italian blew me away wide awake now yeah here you want to see I'll I'll show you the um I'll show you the um
737. 2:05:49I've got a presentation that's got some good stuff on it oh this is a good one all right so yeah so so this is this is this is a
738. 2:05:59presentation from from my company storyvine so this is some of the AI product that I'm putting together
739. 2:06:05um so so what my company does is we're an 11 and a half year old company we've got an app that guides a human being
740. 2:06:15through telling their story so it prompts them step by step so so they say um introduce yourself and when you were
741. 2:06:23diagnosed with this disease right and you know talk about when you were diagnosed and how that felt emotionally
742. 2:06:29and how how have you been treating it and how are you doing and what are your hopes and dreams for the future so
743. 2:06:34people answer a series of prompts in the app and then our software takes those answers stitches them together into a
744. 2:06:42video Auto magically in the cloud and 5 minutes later you've got a fully edited video so you know powerful thing we're
745. 2:06:48actually not using AI to do that automation we we're using just more traditional assembly stuff but now using this AI stuff I can
746. 2:06:58because I'm capturing content in this modular way I can take those modules and do all sorts of stuff with them like say
747. 2:07:06this so here's Ellen my greatest source of support living with rheumatoid arthritis has been um speaking with
748. 2:07:14other people about what they've gone through uh reading a about the effects of rheumatoid arthritis so there she is
749. 2:07:22you know that's the original video here she is saying that same thing in
750. 2:07:42Spanish and so now I've got her speak in Spanish well I can also because I've because our software where builds videos
751. 2:07:50in this modular way I can also take any onscreen text that's in English and I can translate that to Spanish and then I
752. 2:07:57can so I can take all these individual components that are in English translate them all to Spanish and then reassemble
753. 2:08:03them so here's the reassembled video with both text and and audio now in Spanish so notice her [Music] name
754. 2:08:21[Music] so notice the lower thirds in Spanish and then I can take that video and I can put it into a system that we call the
755. 2:08:32authenticity engine which generates content about this video and um generate transcripts chapters descriptions all in
756. 2:08:44Spanish and so and I can do this for every video in our system and then I can do that in 27 languages so here's Ellen speaking
757. 2:09:02[Music] Hindi here she is speaking [Music]
758. 2:09:16Japanese right I mean I mean we're we're we are in it's this technology is not coming it's here it's just that very few
759. 2:09:29people know about it yet and not this technology but this technology is basically the same core technology that
760. 2:09:39is sitting behind chat GPT is sitting behind all these other magical models that the reason we're seeing innovation in so
761. 2:09:48many different areas all at once is because it's all the same basic technology just trained on different
762. 2:09:56data and so as you get more Curious with this stuff you can start to do things that seem impossible right now but they're
763. 2:10:11not we're in a different era it's here it's not coming it's here now is it perfect [ __ ] no is it janky as [ __ ]
764. 2:10:27yeah is it ready for prime time yet no close and getting better fast if you want to if you want a sense of the
765. 2:10:41progression look at Mid Journey images from version one to version six version one of mid Journey images was like 18
766. 2:10:50months ago or maybe two years ago at this point not that long ago version one mid Journey images were you could tell it was trying to
767. 2:11:03make a human but it was not a human and then version two was sort of like a very deformed human and then
768. 2:11:11version three it started to look kind of like a person and then four it was good and then five it was really good and now six
769. 2:11:21it's competes with whatever I introduced another friend to Pi and she fell in love with it so if you haven't played
770. 2:11:29with pi so here here's another thing did you know this was possible this is an iPad I have it on a fancy keyboard stand but it's an
771. 2:11:39iPad and so I'm going to launch piie so that's how you spell it not pie this is huge for teaching oh it's crazy it's
772. 2:11:47crazy crazy so I click on pi and then she shows up and it's it's just a normal like you know talk to pie and then there's this
773. 2:12:02little telephone and so I'm just going to hit the telephone I'm a little bored and I'm wondering if you can teach me something
774. 2:12:14that I don't know anything about
775. 2:12:34p is thinking she's thinking too long I I think this might not be connected to the internet hang on let me let me quit it and restart it
776. 2:12:48too many requests ret retry make sure your internet connection is okay okay here we go yeah my internet connection was
777. 2:13:00dead I'm a little bored can you help me come up with something to entertain myself absolutely here are some ideas
778. 2:13:13try a new hobby maybe learn a new instrument paint a picture or pick up a craft like knitting or wood working
779. 2:13:19exercise take a walk do a few yoga poses or dance around to your favorite music read a book dive into a new novel or
780. 2:13:27non-fiction book I have an idea I want to say something in English and then I want you to repeat it in some random
781. 2:13:35foreign language and have me guess what language it is that's a fun idea I'm happy to play along you can and go ahead and say
782. 2:13:48something in English and I'll repeat it in a random foreign language then you can take a guess as to what language it
783. 2:13:54is sound good sounds great okay the sentence is the Llama was walking down the road and was about to step on a Rock
784. 2:14:04when he realized it was a turtle W GOI is that
785. 2:14:23[Music] Chinese yep it's Chinese you got it right on the first try give me another language same
786. 2:14:38sentence sure thing here it is in French Lama Mar doing a tour tour but wait you told me it was French I didn't get a chance
787. 2:14:52to guess you don't know how to play the game she'll [Music] apologize Oh no you're right I got ahead of myself there my apologies I'll wait
788. 2:15:05for your guest next time so so that exists did you know that exists chat GPT can do that too by the way the the mobile version of chat GPT
789. 2:15:19so if you download the iOS app from open AI it's got a similar button and you can put it in conversation
790. 2:15:27mode and then if you build a GPT this personalized customized version of chat GPT you can talk to them so you could be on a road
791. 2:15:37trip and maybe you did a GPT that helps you work through issues about a specific topic you could go on a 4-Hour road trip Tri and work
792. 2:15:48through some [ __ ] and when you get done you'll have a transcript of that conversation the movie version talk to
793. 2:16:03my daughter and it knew it was us by our voice wow yeah there you go I mean yeah so all of this stuff right now are like teeny tiny little
794. 2:16:19glimpses on the mobile version teeny tiny little glimpses into what's coming in 2024 2024 is going to be Bonkers because
795. 2:16:31all of these models are going to get multimodal and so the different companies are going to try to one up each other and provide different
796. 2:16:40features that make you want to use theirs versus someone else's so there's going to be different interfaces
797. 2:16:46different like like like Pi the company behind Pi that company is called inflection and they raised $ 1.3 billion they spent about a billion
798. 2:16:58dollars of that 950 million on 22,000 high-end gpus so that they can make their own foundational model their own Frontier
799. 2:17:11Model so just like gp4 is a base model they're going to make their own their company mission is to create the world's best
800. 2:17:20personal assistant so Pi is really good at conversation it's going to get way better at conversation and it's then
801. 2:17:30going to be able to take actions so you're going to be able to have a conversation with pi and say hey I'm
802. 2:17:36having an argument with my friend Steve and here's what happened and I think I'm right and I think he's an [ __ ] and
803. 2:17:43then Pi will go well I can understand how you might feel like that here's another way to look at it here's what might be going
804. 2:17:50on 45 minutes later I'm like oh you know what you're right I was being closed-minded and I don't think I saw
805. 2:17:58his point of view can you help me draft a letter that that you know takes responsibility for my piece of it and
806. 2:18:08tries to open the dialogue back up with him sure I'll do that how's this oh that sounds great can you send that to him
807. 2:18:15now sure boom off it'll go and then she'll probably go hey Steve got back to us he apologized too and uh I went up
808. 2:18:24ahead and set up a meeting for tomorrow does that work for you your your schedule was open like that we probably get in
809. 2:18:382024 and we probably get it from Apple we probably get it from PI we probably get some version of it from open
810. 2:18:45AI Bard is they they've already demoed some stuff that's in that neighborhood we haven't seen it yet but you know Pi
811. 2:18:53is connected to the internet yes Mike so yeah you can you can ask Pi like tell me about the specials at Costco as you're driving to
812. 2:19:03Costco hey I'm bored I'd like to see some live music tonight who's playing tonight you want me to book those tickets it doesn't do that
813. 2:19:13yet it'll tell you about it right now I was talking to Pi in my car and forgot
814. 2:19:29about it suddenly PI says if you keep using that language oh that's funny Pi's funny Pi Pi and Pi's got some
815. 2:19:41memory like she she'll occasionally go oh Kyle she remembers my name it told me I might have to block your account or something like
816. 2:19:52that she said let's keep it polite yeah um Sydney the Microsoft one is really bad with that that one will
817. 2:20:02shut down the conversation if you're rude to it but again that's where I think we're going to have different
818. 2:20:07models right I'm like a smartass and I like to swear and I like to go oh you're just a dumb dumb and like the Microsoft model gets
819. 2:20:20offended by that and shuts the conversation down like so I I don't use the Microsoft model at all because it
820. 2:20:27doesn't work with who I am I would imagine that what's going to happen with pi is that you're going to be able to
821. 2:20:35sort of dial in your safety stuff right if you just want to say oh my gosh instead of oh my God and and you wanted
822. 2:20:43to have those kind of boundaries and that kind of conversation um I'm sure you're going to be able to say hey let's let's have that
823. 2:20:50and then it will have memory it'll remember that and then if if I want mine to be more of a smartass and more like a
824. 2:20:57stand-up comic like I'm interacting with some you know smart aliy you know judgmental funny person then it'll act like
825. 2:21:13[Music] that early Revel for me Papa JJ bye till next time see you later Papa JJ yeah I got to get out of here too let's see it's 10:21
826. 2:21:25it's time it's time people Pi talked to my dentist and told him not to be afraid of it oh that's funny so when everyone
827. 2:21:33is using it nobody will have real conversations and AIS will just talk back and forth um for so Dei I so here's here's where I
828. 2:21:44go with that it so what you're saying is is largely true I will have agents that take actions on my behalf you will have
829. 2:21:54agents that take action on your behalf my agents will talk to your agents and they'll talk to Jane's agents
830. 2:22:02and they'll talk to Sam's agents and our agents will all coordinate and they'll do [ __ ] together what the agents will do is do
831. 2:22:12all the shitty work like hey are you available then oh I've got a party then I can't really do all the like when when
832. 2:22:20you want to like coordinate something with your friends and it's like [ __ ] 3 days of ridiculous back and forth
833. 2:22:26conversations cuz someone had their Muller pulled and someone else couldn't do it and their cat just threw up on the
834. 2:22:32thing and all of those conversations can be done by the [ __ ] Bots and it's just like hey tonight you're meeting with Sally Jim and Fred
835. 2:22:44oh cool where are we meeting you're meeting here Fred's really excited to talk about the [ __ ] you talked about
836. 2:22:51last time oh yeah what were we talking about again remember you told Fred this idea for the thing and then he had an
837. 2:22:56idea for the thing and then you two came up with this idea oh yeah I remember that yeah he wants to talk about that
838. 2:23:02again okay cool so then when I see Fred we just start right where we were all of that other [ __ ] is handled by the
839. 2:23:10robots yes sign me up for that best local language model currently local language model oh want to self-host on my
840. 2:23:22computer I I don't it it depends on the specs of your computer I I mean the the um I I would say grab just
841. 2:23:32it depends how technical you are I would say grab a piece of software called GPT for all and that's just a an installable
842. 2:23:41that you can install on your Mac or your Windows computer and then you can select from different models and you can
843. 2:23:47experiment with them um what I would do if I were you is I would I would go to Reddit um on the open source AI find
844. 2:23:57some open source Ai subreddits and just start hanging out there and then you're going to you're going to want to play
845. 2:24:04around on GitHub and hugging face and you're just going to need to educate yourself because it's got to do with the
846. 2:24:11specs of your computer and how much RAM you have and what you actually want to do and why you want to host it and do
847. 2:24:17you want it to be um have safety guard rails or no safety guard rails it it just like it's the the amount of parameters
848. 2:24:26of of what's best for you are [ __ ] Bonkers so you're just going to have to self-educate my computer is super
849. 2:24:34powerful awesome then you might be able to play with the 70 billion parameter stuff and have really powerful AI
850. 2:24:42there um you can also do virtual machines right you could spin up something on on GitHub co-pilot or not GitHub co-pilot on U on Google um
851. 2:24:56uh what is it Google uh Google I forget what it's called it has the notebooks it's not co-pilot it's Google I forget what it is but anyway
852. 2:25:11you can spend up virtual machines that have GP uh gpus in the cloud GPT for all yeah that's it with LM Studio you can download hundreds of
853. 2:25:23models none of them work exactly that's the problem with the open source world is like you have to be in it pretty deep
854. 2:25:30to understand what actually works what doesn't what's been updated what hasn't and and it's it's changing on a weekly
855. 2:25:36Google collab thank you yeah Google collab you can spin up an instance of Google collab with multiple gpus and you
856. 2:25:42can run really powerful models in the cloud in a private instance that you know your your data isn't going off to to chat
857. 2:25:50GPT will AI decide who won for some games first rule of the cult is not to talk about the cult chat gvt not
858. 2:25:59allowing public domain stuff is a bit silly yeah I agree all right listen everybody I got to get out of here my voice is going
859. 2:26:08crackly my eyes are cloudy I'm still recovering from [ __ ] Co and I did a I did a 24-hour GPT on this weekend um if you if you want to
860. 2:26:20play around with some of the stuff we did with um with the gpon this weekend go to the salon. it's a link tree the first two or
861. 2:26:34three the first three links are about the G GPT for good event we did this weekend it was a 24-hour event where we
862. 2:26:41made 150 gpts for nonprofits in a 24-hour period it was cool um but the the videos of those sessions and the directory of those gpts and the
863. 2:26:54the Google sheet that that we created them with are all there and then also go check out the AI Salon there's there's a
864. 2:27:01link for that there well as well um it's a community of people like if you're trying to figure this AI [ __ ]
865. 2:27:08out and you want to like hang out with people that are up to it and like curious about it too and are not
866. 2:27:15judgmental that's what group is it's amazing it it it it's largely this group that made that GPT for good event
867. 2:27:25successful this weekend thanks Kyle you're amaz amazing best two hours on here oh thank you very much I appreciate
868. 2:27:31that please if you have time check my GPT the link is in the S&T okay oh show and tell on the AI Salon great terrific you're awesome Kyle
869. 2:27:42even when you're sick thank you Vicki I appreciate that any tips on getting chat to learn how to spell correctly no no
870. 2:27:51just keep trying it it I'll tell you what the dolly3 spelling is better than it was a month ago it got it got
871. 2:27:59markedly better in the past two weeks I don't know what they did something got better but it still sucks at spelling
872. 2:28:06but again here here's here's my this is my point of view on life's tough it's tougher when you're stupid
873. 2:28:14um here's the here's here's my opinion on uh dolly3 spelling shittily enjoy it now we are going to look back on this era where [ __ ] is
874. 2:28:26spelled wrong like hilariously wrong like nostalgically because I'm guessing by 6 months from now all of that's going to be solved it's just
875. 2:28:38going to spell it correctly it's going to be [ __ ] brilliant be like oh you remember when it used to do shitty spelling that was kind of
876. 2:28:44fun have you seen mid Journey V5 I assume you mean V6 I have seen V6 it looks it looks pretty good it's different you've got to prompt
877. 2:28:52differently like that's an that's an amazing thing V5 you you learn to prompt mid journey in a certain way V6 comes
878. 2:28:58out you got to completely relearn prompting that's just the errow we're in you just going to have to keep being
879. 2:29:03flexible people I never thought I'd be evangelizing people on AI I know welcome welcome to the AI learning lab that's
880. 2:29:10the whole idea of this is that I want people to experience this [ __ ] I don't mind if you hate it a i as long as
881. 2:29:17you've tried it most people that hate it haven't tried it most people that have tried it have their Kevin McAllister
882. 2:29:29moment and then they can't go back from it and then they turn into weird ass evangelists that's okay that's why we're
883. 2:29:38here all right so without spelling test I would have gotten Straight A's in English class it's amazing recommend
884. 2:29:45folks join official mid Journey Facebook group yeah it's oh the mid yeah the the mid Journey Facebook group's really good
885. 2:29:51some incredible Creations there that's a very well- curated uh group on Facebook all right everybody I'm out of here
886. 2:29:58thank you to the mods it was fun watching you party tonight while I was off you know doing my I was just
887. 2:30:04rambling tonight so um do they have GPT in China they have their version of it I think they've got this GPT blocked
888. 2:30:12because it just answers questions and they don't want that um thank you all right I on China good to see you Lord
889. 2:30:19digital Gods good to see you 101 Source Camp thank you thank you thank you regulars jrc Emilio's wife thank you so much Mr it good to see
890. 2:30:31you what's your name truck TR trenen I like the name uh amethyst Gwen good to see you Nicole there's a bunch of new people in here
891. 2:30:44great let's go uh um I on China Cosmic lover good to see you Spike Jones thank you for your comments questions thoughts
892. 2:30:53opinions I love talking about this stuff as you know Earth rocker thanks for being a subscriber Sky limbo all right
893. 2:31:00everybody Joe mama good to see you I haven't seen you in a bit good to see you happy New Year everybody oh this is public service
894. 2:31:07announcement tomorrow is Tuesday go back to work 2024 160 people in here thank you uh join the AI Salon keep coming back here
895. 2:31:22and uh I really appreciate you all more than you know peace