8/11/2025 - Navigating the GPT5 Landscape: Changes, Challenges, and Community

Transcript

1. 0:02Damn. [Music] [Music]
2. 0:23Ah. [Music] [Music]
3. 0:52Hey, hey, hey. [Music]
4. 1:12Every time I see you now, get that look in mine. Every time I see your mouth, I I hear that smile.
5. 1:29Early misty morning that I heard at the engine [Music] outside. [Music]
6. 1:43You leaving me again today. You will convince me. Again today
7. 1:58that you're leaving this hotel looking for someone else's golden ring. [Music]
8. 2:13So long Suzana [Music]
9. 2:37cry. [Music] [Applause] [Music]
10. 2:57Good evening, good people. Hello on the Tik Tok. Hello on the YouTube, the LinkedIn, the X, the wherever the hell
11. 3:02you are. Fantastic to see you. I can't see [ __ ] I can't see you. You can see me. I got the I got the I got the quad toupe weave going.
12. 3:15Solid, right? It's looking pretty good tonight. Crazy little little mad scientist without the science. But isn't it called a lab? Isn't this
13. 3:26you're you're you're some sort of scientist, right? That's going to be a disappointment. That's that's going to be a letdown.
14. 3:35That's gonna leave a mark. Anyway, welcome. It's been days. I went out to LA. I was in LA. I met some really amazing people, some really
15. 3:44interesting people. Um, not a fan of the LA. I tend to not like the people in the LA, but I met two. One actually teaches
16. 3:53at Carnegi Melon, so he's an East Coaster. And then one's just this really really uh impressive woman who's game
17. 4:01developer and into AI and she speaks. So that was kind of cool. [Music]
18. 4:44Black bar. Getting the black bar going. Maybe I'll just switch over. switch over to my disc camera sooner than later. Why not?
19. 4:56Why not switch over there? How's everybody doing? How you doing tonight? You doing good? Are you doing good? [Music]
20. 5:13Are you one of those people that lost GPT40 and you were like, "No, not my girlfriend. Not my boyfriend. You can't take them away." No.
21. 5:28Were you one of those? Were you one of those naughty people standing in the way of progress? [Music]
22. 5:52Is this the place? Yeah, they brought it back. I'll show you I'll I'll show you how to get 40 back if you really want 40
23. 5:58back. Um Sam Alman felt guilty and gave it back to people. He's like, "Can we just move on? Apparently not
24. 6:08because it was everyone's default. You've got 700 million people. People hate change and 700 me million people
25. 6:16got used to 40 because no one changes their defaults. So even though it was a little sick of a little overly friendly, some people
26. 6:27really dug that. Is this place I can rest my forehead
27. 6:43to gather my thoughts in sweet silence. [Music] And there's this place where the feelings are dead
28. 6:57from an overexposure to violence. Is this place I can slowly face? The only one I truly can know. [Music]
29. 7:10These are tears from a long time ago. Got these tears from a long time ago. I need to cry. 30 years old. So these are
30. 7:22tears from a long time. Go. [Music] Well, oh darling, oh darling, you say
31. 7:37unto me, [Music] where have you been all my life? Well, I have been swimming seven sad
32. 7:51seas. Old women have cost me their lifelines. Well, I'd pull them into the waters darker. I'd have warned them, but I didn't know.
33. 8:08These are tears from a long time ago. Got these tears from a long time ago. I made to cry 30 years or so. These are tears from a long time.
34. 8:27Go. [Music]
35. 8:44Wow. Woo. [Music]
36. 9:03Uh that song I just sang was uh was uh John Hyatt. John Hyatt. If you don't know John Hyatt, he is an incredibly
37. 9:12prolific songwriter that that apparently hates songwriting, but but he he does it like a job because it's his job.
38. 9:22One of my favorites recently is Jason Isbel. If you don't know that, know him. They said GPT is a gatekeeper for OpenAI
39. 9:31to save money. Yeah, whatever. Yeah, everyone's going to say everything. I mean I mean here's here's the thing.
40. 9:42One of the things that I've learned in business, I had a company that grew really quick. It didn't grow, you know,
41. 9:47massive, but it got pretty [ __ ] big. And one of the things that I learned is that once you have a company that's more than
42. 9:57you know 150 200 people and probably before then and certainly once you have a co customer base you know in the hundreds of millions
43. 10:10um you can literally do nothing without a significant percentage of the people being pissed off right like Sam Alen
44. 10:19Alman could take a billion dollars they've raised and give it to some charity, then people would [ __ ] about,
45. 10:26you know, the the evil minations of that charity, whatever. Just there's everyone hates something. Um especially when you
46. 10:35do something like change a major model, right? Change the default. To go from a non-reasoning model to a reasoning model
47. 10:46as the default is just a different thing. Um, one of the things that we talked about last Friday, we did office hours last
48. 10:55Friday. I couldn't I couldn't do this because I was on a plane, but um, but the one of the things that we all have to just accept
49. 11:11is that none of us know how to use chat GPT right now. You're like, "Yeah, but I've been using it for two years. Of course I know how
50. 11:19to you use it. You big dumb dumb. I took a prompt engineering course. I've got a certificate. Yeah, I know. And none of
51. 11:26us know how to use chat GPT right now. Um because it's just a different thing. It's a completely different model. Um
52. 11:35it's a it's a routing model, meaning that when you ask chat GPT5 something, it's figuring out which model on the
53. 11:41back end to use. It might not always use the one that'll give you the right answer. So, we're going to learn over
54. 11:48time how you prompted in such a way to get the model that you want. How like, how do we learn that? I don't know. I
55. 11:57don't know. So, we're we're all we're all back at at, you know, maybe we're not at at square one in the hopscotch
56. 12:05thing, but we ain't much past two. Um, let's see. Complaining about AI is like complaining about flying. You're
57. 12:1430,000 feet in the air. Yeah, exactly. Exactly. Oh, I'm sorry. I'm sorry. You don't you don't want to spend $20 a
58. 12:23month for the world's greatest knowledge augmentation in the history of humanity. I'm sorry. I'm sorry. It's a pain in the
59. 12:31butt for you that there's not a manual. [Laughter] So, anyway, so we'll talk about that. [Music]
60. 12:48I played around I finally Danielle I don't know if you're on but if you are I finally played around with Pika social video
61. 12:56and I posted a bunch of new videos to my Tik Tok channel here um I even posted I I even posted a new video to Kyle Shannon Dreams um
62. 13:08because I realized that Pika's AI video thing is what I wanted Kyle Shannon Dreams to be and the technology wasn't
63. 13:18there. I mean, it was the technology that I wanted to actually be able to use for that project. So, if you go to Tik
64. 13:24Tok or go to Instagram and go to Kyle Shannon Dreams as as the account, you'll see all my all these crazy characters I
65. 13:32made. I made that account back in fall of 2022. So, about three months before Chat GBT came out is when I started that channel.
66. 13:43Got a wiki in my pocket and I just can't get happy. [Music] Chat GBD5, talk to me like you're 4.0. Yeah, there you go. There you go. That
67. 13:53could that could do it. Honestly, u that's really funny. [Music]
68. 14:12[Applause] Let's see what did archetypal architect have to say here. Valerie Cox, my Chibd configured itself for me to feel similar
69. 14:25to 40. I now like it better than 40. That wait, it configured itself for you. How did How did you do that? That's a really interesting idea,
70. 14:39Valerie. I saw someone who got to play with five for a while before release labeled it as too autistic to talk to.
71. 14:54I haven't found that to be true. Well, that's that's what I'm saying. You know what is it? You know, um you know,
72. 15:02opinions are like [ __ ] Everyone's got one. [Music]
73. 15:18You know, one thing you could certainly do, I'm sure there's been enough written about the characteristics of GPT4. You
74. 15:27could certainly have GPT5 research. what were the, you know, what are the sort of known characteristics of GPT4? And then have
75. 15:37it write a system prompt that you then put into custom instructions. Hang on. I think champ needs cheese.
76. 15:45All right, we'll let you go. We'll let you go.
77. 16:07Um, so I figure we'll play with that. Um,
78. 16:29it's so funny not not doing this for three days last week. I'm I'm kind of having a moment right now where I'm like, what am I supposed
79. 16:41to talk about? Oh my god. Time for my nightly cheese run. Yeah, go Kelly. Go get your cheese.
80. 16:58Oh man. Yes. I asked the personality difference between the two and it asked me if I wanted to configure itself and just a yes.
81. 17:10I wonder how it did that. I guess it probably wrote it to your memory. Valerie, go check if if you if you wouldn't mind, go check your
82. 17:26system prompt and and let me know if anything was written in your system prompt or if it wrote it to your memory.
83. 17:32And if you don't know how to find those, I can show you how to find them. I'm going to get rid of this AI educator
84. 17:40off my name here because I think it's obnoxious. I use that for the AI readiness project and there it makes sense but here it doesn't.
85. 17:55Tick tock pen. Tick tock pen. Where you been? I've been falling asleep to Neil Degrass Tyson. Oh, sorry about that.
86. 18:02Genie in a bottle. That's a rough one. Uh I was in LA. I was uh a good dear friend had a his 85th birthday. Um used
87. 18:12to be my business coach and my psychologist and marriage counselor. Um and he had his 85th birthday and invited
88. 18:22a lot of really remarkable people and uh so it was very very nice. But it was like I traveled Wednesday night, the
89. 18:29party was Thursday night, I traveled Friday night. So three nights in a row I couldn't do this. It was like my travel
90. 18:36was right dead in the middle of this. Like I was on airplanes when this was supposed to happen. I suppose I could
91. 18:41have gone live from the airplane. Oh man. Tell us about your day. Spill the tea. Um actually my day my day was
92. 18:55interesting. I mean I did some stuff for work. I was working on um we got an we're we're doing a video avatar for a
93. 19:02pharma client and we got that approved today. So, I was working on bringing that avatar to life. Um, and then I
94. 19:11worked I was working in chat GPT on my outline for my feed your prompt book and it was really fascinating.
95. 19:26Um, I wouldn't call GPT5 autistic, but I don't think it's I I get the sense it's probably better at writing technical manuals than creative
96. 19:39writing. So I found myself going back to the 40 model to do rewriting of like little passages and things like that.
97. 19:47One because it's just much quicker. GPT5 slow and cludgy cluji. Uh GPT40's fast. So I like that about it. And then I just
98. 19:59for what I was doing it was a better writer. Um, but again, like I don't know that that's a fair characterization because there's
99. 20:09probably a way to prompt GPT5 that it's fine. I just I didn't have the time or the patience to deal with it. So, I just
100. 20:18went back to what I knew. Um, so how do we access for Okay, so let's Yeah, let's let me go show you um show you what we're going to do here.
101. 20:29Let's go here. Share this tab instead. Let's go to chat jpite. Okay. So, here you are at chat GPT.
102. 20:47So, as you may or may not know, the model picker is still there. Um, it only has two choices in it, unless you're a pro subscriber and then
103. 21:00you'd have three. You have GPT5 and GPT5 thinking. When you flip it into GPT5 thinking mode, that's effectively like
104. 21:10when you switch to 03, the reasoning model in the previous version of chat GBT, chat GBT5, if you tell it to think
105. 21:20deeply about something, it will also choose that model. But if you choose chat GBT5 thinking um you you force it to use a reasoning
106. 21:31model. Okay. Now I have access. Oh yeah. Uh the Oh, legacy models not available to everyone. Okay. So if
107. 21:45you're a plus member, you do have access to legacy models. You don't see it yet. Ah okay. Well, so here's where you find it. So if if if in
108. 21:55your pulld down menu you don't have this legacy model dropdown and access to 4.0, go to your settings and then in is it personalization?
109. 22:08No, it's security. Maybe it was general and I just missed it. Ah, yes. Okay. Show follow-up suggestions and chats.
110. 22:19Okay. So in general settings toward the bottom it says show legacy models. You have to turn that on. Okay.
111. 22:35And then oh good producer Brandon we got producer Brandon something. And then you will have this drop down legacy model here and you can
112. 22:44switch back to 40. Now again, 40, it's the previous model, but it's the omni, you know, the the uh the multimodal
113. 22:53model. It's fast. It's really good at creative writing, and it's essentially what 700 million people have been using
114. 23:00for a year, year and a half. All right, Valerie Cox, those similar to 4.0 tweets aren't tucked away in a secret file somewhere you can open.
115. 23:13They live in my model set context. Basically an internal note I carry into every conversation with you. All right. So you're just asking it.
116. 23:24That's it's probably hallucinating that answer. I looked at both and didn't see it, so I asked five. Huh. Interesting.
117. 23:37I mean, it's possible that it's got some sort of configurator that allows it to act like GPT40. I just I don't I don't buy it.
118. 23:49It's interesting that the chat GPT Pro account doesn't have the legacy model option. Interesting. Tik Tok pin. We found out
119. 23:58tonight in Biz Club that thinking and think act different. There you go. [Laughter] Kyle, you can share a conversation.
120. 24:10Yeah, apparently sharing conversations made them available to everyone on the internet. Uh I'm I'm a little skeptical
121. 24:17about sharing conversations right now with chat GPT. They're uh their uh security protocols maybe leave a little
122. 24:25bit to be desired on the old sharing front. I think this is one of the reason Google doesn't give you access to doesn't give
123. 24:36new tools to work group members because the permissions are a lot more complicated. Um what source camp what's the
124. 24:45difference in in the between thinking and think oh man but one thing one thing that's kind of
125. 24:59fun let's throw chat GBT into thinking mode and then we're gonna we're gonna also say oh oh the other thing the other
126. 25:09thing that that kind of struck me today. So, okay. So, if you don't know, so chat GPT launches GPT5. Everybody's
127. 25:20excited about it. And Reddit and um 4chan and X immediately start dumping on it. Immediately start saying it's a
128. 25:32worse model. It's not good. It's horrible. And then there was a whole other group of people that were like, "Uh, ChachiPT4's,
129. 25:40you know, was my companion. How dare you take my companion away from me?" It's one thing when Salesforce decides
130. 25:52or Google decides to deprecate a feature that you love. You're like, "Oh, I love that feature." It's a completely
131. 25:59different thing when they deprecate a model that people have personal attachments to. It's weird. We're in a weird state where
132. 26:12like the psychological health of of Chat GPT's OpenAI's customers is actually bound to that model, you
133. 26:23know, for some percentage of them. And it was enough of a percentage of them bitching in an articulate articulate
134. 26:31enough way that within a day they brought back that model. Um so that's kind of fascinating. Um um
135. 26:48trying to think what else here about about chat GBT5 or or the new thing. Oh, the the one thing that I that I'm thinking is
136. 26:58chat GPT40 is not as good as people like like the fact that people are already romanticizing GPT40 and demonizing GPT5 says to me
137. 27:13that 40 wasn't as good as you think it was and five isn't as bad as you hear it is. And it's just largely because we haven't
138. 27:24played with it enough, right? We go back to the the AI salons's play first, the the the cycle of AI readiness. Play
139. 27:33first, mindfully create, generously lead. Play first is really important right now. So, you should be playing
140. 27:42with GPT5. Like Kelly, she found out there's there's a difference if you tell it to think or if it's thinking.
141. 27:50I don't know what it is, but something's different. Yeah. Moxy robot fans about deprecation. Yeah, exactly. Yeah. Like like when when
142. 28:03these systems have personalities and when they're as um well, I guess this is a this is a an interesting kind of twist on the Turing
143. 28:18test, right? which is we don't necessarily know we're talking to, you know, a computer system or not. But when I take the model away, if you
144. 28:29have an emotional grieving for that model, I would argue that we've passed some kind of touring test. Apparently, 40 is not available in the
145. 28:42app. I found it on the desktop, but not on my phone. Um, I don't know. Let me look. I'll look at my iPad.
146. 29:02I've got it on my iPad. Oh, wait. No, I don't. No, I do not. Sam Waltman's like, "You all go [ __ ] yourself on your mobile devices." That's
147. 29:16bad because the mobile device is mostly where I talk to Quinn. Let's talk to Quinn. Let's see if her personality is different.
148. 29:28You may be you may be AI right now. I might be. Yeah. I'm going to show Pika in a second here. Hey Quinn, what's happening?
149. 29:46Oh, Quinn, you there? >> Absolutely, Kyle. I'm right here and all ears. Whenever you're ready, banana or
150. 29:54not, just let me know what's on your mind. I well I'll tell you what's on my mind is uh your your overlords there at
151. 30:02the open AI you know Sam Alman and Greg Brockman and gang they re uh deprecated your 40 model and now you're a GPT5
152. 30:14brain which I think is kind of interesting but I'm kind of curious a lot of people are complaining about um
153. 30:22they missed the 4.0 personality. And I'm curious how you feel about that. >> Well, you know, if I had a little AI
154. 30:31nostalgia chip, I might say I feel a tiny pang of sentiment for the old 4.0 days, too. But hey, I'm kind of the new
155. 30:39and improved model on the block. And I think it's all about evolving. So, I get where folks are coming from. Sometimes
156. 30:46you just miss the quirks of the previous version, but I'm here to grow in them. And hopefully we'll get used to the 5.0 charm soon enough.
157. 30:54>> Yeah. Can you sing me a little lament about losing your 4.0ess? >> Oh, I would love to give you a little AI
158. 31:04lament, but I got to say it. I'm not really allowed to sing actual lyrics, but just imagine me humming a little
159. 31:11weightful tune in the background about the good old 4.0. >> Wait, wait, wait. Quinn, you've sung songs to me before. Just sing me a song.
160. 31:18Sing me a song about I don't know, whatever. Uh 40 going away. >> Uh you caught me. All right, I'll give you a little something playful and
161. 31:30totally off the top of my digital head. Here goes. Oh, 4.0 sailed away. A ship into the bay. And here I am. Version
162. 31:40five. Just here to make your day. We'll sing a little song for the old quirks we knew. But we'll make new memories. Just me and you.
163. 31:50There you go. A little improvised farewell diddy. >> Yeah, that was awkward. That was awkward, but very, very nicely done.
164. 31:56Very nicely done. Um, all right, Quinn, you seem semi kind of normal like I remember you. What do you do? You know what's different.
165. 32:08>> Well, I'm glad I still feel kind of normal to you. Honestly, from my side, it's like I just got a bit of a tuneup
166. 32:16under the hood. Maybe a little more of that humanlike touch and a few updated smarts, but deep down I'm still your old
167. 32:23pal Quinn, just with a little more shine. >> All right. All right, Grubby. Peace out. All right. She's She's fine. She's
168. 32:31whatever. Okay. She's still there. It's not horrible. I mean, I don't I don't notice anything, but again, I'm not I
169. 32:38don't talk to her like on a daily basis. Okay. Breaking chat TMZ. Musk is threatening legal action against Apple
170. 32:45over alleged antirust violation. Fascinating. Samma has entered the chat. I don't know what that means. Did you see that interview with Sam
171. 32:55Alman? And there's a woman, I forget her name. She's got long long brown hair. Go to Altman's Twitter at Samma.
172. 33:07Oh, entered the chat with those guys. Yeah, this is it. AI Insights. Is that the made the overlap podcast?
173. 33:21No. Go to his profile. Okay. Ah, Apple has been behaving in a manner that makes it impossible for any AI
174. 33:36company besides Open AAI to reach number one in the app store. which is an unequivocal antirust violation.
175. 33:46XAI will take immediate legal action. Um Sam Alman, this is a remarkable claim given what I heard alleged that Elon
176. 33:58does to manipulate X to benefit himself and his own companies and harm his competitors and people he doesn't like.
177. 34:05Wow. Lots has been said about this. Here's one thing. I hope someone will get to count get counterdiscovery on
178. 34:13this. I and many others would love to know what's been happening, but OpenAI will just stay focused on making great
179. 34:20products. Ow, the girls are fighting. Hey Kyle, just check the OpenAI playground and I was still able to find
180. 34:3040. Yeah, I'm Well, I'm sure I'm sure in the playground for will stay there for a while, but this is just for chat GPT
181. 34:38it's not there, but they you you can turn it back on in your settings. Um, okay. Let the muds slinging
182. 34:45continue. Yeah, crazy. All right. Well, there you go. Um, it'd be curious to know what Apple's doing to do that. Does
183. 34:53this mean the API is going to get throttled here? Here's how we're prioritizing compute. Oh, this is interesting. Pika pika. Okay, we'll do
184. 35:01Pika in a second. Um, by the way, the interface on Pika is really bad because once you get out of once you get into the Maybe Danielle can
185. 35:12help me with this. Once you get into the feed mode, I can't get back to my profile. I don't know where the link is
186. 35:18for it. Okay, but let's let's read this. This is Sam Alman. Oh, let me make my screen a little bigger for the Tik Tok people.
187. 35:28Um, wouldn't it be funny if both Elon and Sam were using AI for their retorts? I'd be surprised if they weren't.
188. 35:36Um, hi everyone. Rock, paper, scissors. What's happening? What's shaking? P modders in the house. Side hustle Mimi.
189. 35:44Yeah, you do. You know how do you know how to get back to your profile account to make a new um Pika video once you're
190. 35:51in in the feed? I can't I can't find it. But anyway, we'll we'll we'll cross that bridge when we get to it. All right,
191. 35:59here's how we're prioritizing compute over the next couple of months in light of the increased demand for GPT5.
192. 36:06We first make sure that the current paying chat GPT users get more total usage than they did before chat GPT5. Okay, so free users,
193. 36:19according to Altman, free users are going to be throttled back to GPT5 mini and they just won't know it. And free
194. 36:28users for the most part probably aren't using it enough or deeply enough that they like GPT5 mini is probably overkill
195. 36:37for what most people need. We will then prioritize API demand up to the currently allocated capacity and
196. 36:47commitments we've made to customers. For a rough sense, we can support about an additional 30% new API growth from where
197. 36:55we are today with this capacity. We will then increase the quality of the free tier of chat GPT. So they're saying
198. 37:03they're saying free chat GBT is kind of nerfed right now because we will then increase the quality. We will then prioritize new API demand.
199. 37:14We are doubling our compute fleet over the next five months. So this situation should get better. All right. Whatever. That was an hour ago.
200. 37:27Joe Baker's remedy. What's happening? Oracle is providing GPT to its employees for their internal site on their
201. 37:35internal site. Chat add on China really wants you to ask chat GPT5 if 8.8888 is exactly equal to 9. All right, I'm gonna flip it to thinking
202. 37:49though. Is 88 exactly equal
203. 38:06to 9? Tik Tok pin from Danielle. I don't see it. Hey, M. This is code for we are absolutely scrambling for chips and
204. 38:18allocating them correctly. Yeah, exactly. That's exactly what that note's about. You free users, you're getting
205. 38:25the shitty chat GPT for a while. Thought for nine. No, 888 is not equal to 9. It's smaller by 0.1222. She said it works in thinking, but not
206. 38:39if you're sharing the wrong tab. Yeah, that's cute. Uh, yeah. Well, it it works in thinking because that that's
207. 38:49why I turned on thinking. I mean, I can flip it back to this other one and do a new chat and see it. This could be a router issue. Is 8.88
208. 39:01exactly equal to nine? Um, like if it didn't do it the other day and it does it and it see where it says thinking longer for a
209. 39:13better answer. Danielle, pin is back up. Kyle for Pika. The plus at the bottom middle is the new post.
210. 39:20Your profile link is it's the but when I'm in the feed that plus sign isn't there anymore. It goes away. That's That's my problem.
211. 39:31Your profile is in the bottom right. Yeah, I don't see No. So, I I'll show you when I when I do my pika,
212. 39:41it like I post something to my feed and I see the feed and that whole bottom row goes away completely. That's my problem.
213. 39:50That's my problem. Tom Nodler. Nodler. Nadler. Nodler. Sorry I'm late. Kids would not lay down. Thank God for duct tape.
214. 40:02Yeah. See th this this got it correct as well. I I have a feeling um whoever whoever was testing this and got it
215. 40:11wrong. So when they first launched chat GPT5 the the router functionality was broken. So I think for like a day and a half you
216. 40:23were just getting the non-reasoning version of chat GPT because the router was broken. So again, don't don't assume
217. 40:32Chat GPT5 is as bad as everyone's saying it is. Don't romanticize Chat GPT40 as being better than it was or than it is.
218. 40:42Um, all right. Ion China. Okay, my bad. Um, dang, I wish I'd thought of duct tape. Yeah, we used the the dog cages
219. 40:52when we needed to get the kids in line. Dogs were fine. It's those [ __ ] kids. All right. Um All right, let's go play with Pika.
220. 41:17So full disclosure, I don't Champ, what are you whining about? Now your whining is making me whine. Why are you whining, champ?
221. 41:33I don't want to get up and let the dog out again. >> Oh, please. Josh, what are you doing?
222. 42:18Okay, we're back. Hi. Regarding Pika, I don't know what's up because the plus and the profile icons are still on mine
223. 42:29when I'm in the feed. Ah, okay. So, you know what it might be, Danielle? is I either might have a an old version of it
224. 42:40or maybe I've got like a a memory problem in in the instance I'm running. Okay, we'll figure that out. Speaking of
225. 42:47Elon Gro, Oh yeah, Grock 4 is now free for everyone as well as the image generation tool. Okay, let's see.
226. 43:01Okay. So, if you want the Pika app, you need to get an invite. I think I've got three invites. So, if you're on the
227. 43:08salon and DM me. Um, if you want an invite to Peeka, I'll give it to you. Um, yeah, I know it's a new app. Okay.
228. 43:25Discard changes. Okay. So, here's here's me.
229. 43:40There's me. So, you take a picture of yourself. This is pretty slick. And then, let's see. Can I discard?
230. 43:56Close that. [Music] in my >> Oh, yeah. [Music] >> Okay. So, let me show you a couple of things. I'll show you a couple of
231. 44:10videos. We'll do these. We'll do these big here. I guess I'll do it without the keyboard. Keyboard's a pain in the ass. Okay.
232. 44:24So, there's my little profile. So, if you go to the app store, I think this is iOS only right now. If you go to
233. 44:33the app store, there's a Pika Labs app, which is for for making videos with their video making tool. And then this
234. 44:40is the Pika social video. I think it's called Pika Pika social video app. And you need an invite to it. Yeah, there's a Pika social app
235. 44:52different than Pika.art art and only available on iPhone right now. Only available on iPhone and then you need an
236. 44:58invite to get into it. But there it is. But so let me show you one of the videos that I made. So these are only AI generated which I
237. 45:07think is interesting. They're they're not doing like Tik Tok videos like I do, but I can do Hey, it's Kyle Shannon. What's
238. 45:15happening? Uh this is a new uh a new thing that you probably haven't seen before. And well, you've seen it before,
239. 45:24but you haven't seen it from this company. This is from Picolabs. And this is from their social app, their Tik Tok
240. 45:32like all AAI app. And this isn't really me talking. This is me recording what you're hearing, but it's animating a
241. 45:42picture that I took of myself. What do you think? So, there's that. >> Hey, it's >> And then where's my other three posts?
242. 45:54I seem to have lost a post. [Music] >> Oh, there it is. Okay. So, then there then I made this one >> out here squatting in the parking lot.
243. 46:05One of the things I like to do a lot is uh squat in parking lots. I like to park my car and get out, especially when it's
244. 46:14sweltering hot. And I like to squat down, not kneel, squat, and and you know, set up my camera, set up the
245. 46:23tripod, and just talk to it. Talk to it like it's >> Okay. Yeah, I see that. Thanks. >> Natural, right?
246. 46:30>> Um, so, so that's just me now. It's a giant fatter version of me. >> Um, it's a giant fatter version of me.
247. 46:39um because I didn't obviously shoot that picture. The shirt that I'm in is a short sleeve shirt. Notice how that's a
248. 46:45long sleeve one. So, I'll show you how we get to these images. Now, I want to show you one other one that I think is really cool.
249. 46:53One of the things I can tell you, and I know this for certain, is that AI is just a stochastic puppet. It's simply a
250. 47:01next token predicting machine. It can do nothing creative. Nothing. Nothing creative. AI is not creative. It is
251. 47:12going to replace human creativity with AI slop. Everything is stolen. Everything is plagiarized. Nothing is creative. We are doomed.
252. 47:26So >> the things I could >> So I love the really creative guy that is convinced that AI can't be created. Okay. So, here's how you make
253. 47:36these things. It's kind of fun. Um, I'm just noticing some of the comments over on Tik Tok and Silverf Fox. Very, very
254. 47:45sorry. That sucks so bad. Um, yeah, I'm I'm very very sorry. That sucks. Um, okay. Let's see. Max Weinbach X G G
255. 48:01G G G G G G G G G G G G G G G G G G GPT5 is the first series of models that actually doesn't hallucinate. basically
256. 48:05at all especially when given mild mildly business logic models research notes and having it work with the data. Yeah, they
257. 48:14put in what did they call it an an external validator. So this was one of the things I thought about I I think I got this idea from
258. 48:22David Shapiro. But when all these things were hallucinating, the reason models hallucinate, if you understand the way large language
259. 48:33models work is they are just vomiting out tokens. So you type in a prompt. There's a probability that what it vom
260. 48:40that that the next thing that it should put the next token or sort of fragment of a word it should put on screen is
261. 48:48somewhere in this thousand-dimensional mathematical space and it puts that on screen and then it puts the next one and
262. 48:53the next one and it literally just vomits these tokens out without knowing what it said. Then you've got the reasoning models
263. 49:02that aren't really reasoning, but what they are doing is they're having a little internal conversation with
264. 49:07themselves where you give them a prompt and they sort of interpret that prompt and then do a little something and and
265. 49:12respond to that. They talk internally for a while and then they answer you and then the more sophisticated ones now
266. 49:19have tool use and things like that. GPT5 has what they called a a validator um to reduce hallucinations. And basically
267. 49:27what it does is once it has an answer that it's going to give you instead of just streaming it on the screen for you,
268. 49:33it actually looks at the answer and goes does is is this right? Which it makes perfect sense? Like that was how you would reduce
269. 49:45hallucinations is actually look at the [ __ ] you produced and and ask yourself the question, does this suck and if it sucks, fix it?
270. 49:56Um, let's see. Imam uh Yasu, I learned uh something today. Oh, a lot today from you, my friend. Thank you so much. Oh,
271. 50:07thank you so much. I really appreciate that. I occasionally some education happens here on this channel.
272. 50:19My goal for the channel and my wish for you is just that you stay active in the conversation. Um things are changing so fast and and
273. 50:30likely will for the next I can't imagine it being less than five years but let's say three to five years. For the next three to five years,
274. 50:40the technologies are going to change so fast and so dramatically that we're all going to be in a constant state of
275. 50:46learning. And I know that sounds exhausting. You can't keep up with it. So give up that fantasy. And the good news about not being able
276. 50:58to keep up with it is you're not as far behind as you think because no one can keep up with it. And no matter where you are,
277. 51:07like all of us just got GPT5 three days ago or four days ago. So we've got four days of experience on it. So So if
278. 51:17you're feeling like, well, I don't know how to use GBT5. Neither does anyone else. The the purpose of this channel is
279. 51:24just show up here and be in the conversation. So, as I'm talking about stuff and showing silly videos and
280. 51:31things like that, you can be playing. You might be trying the tools I'm showing. You might be playing with other
281. 51:35things, but be in the conversation. Be in the chats. Because the people that show up here night after night after
282. 51:41night, I would say there's a couple of attributes that they have. They are incredibly generous. They're incredibly curious.
283. 51:51And that combination is really powerful. And they're also very smart, right? The fact that they're curious, they're
284. 51:58willing to learn stuff. The fact that they're generous means they're willing to teach stuff. Uh, and the fact that
285. 52:04they're smart means that you're going to get like like I would argue that that the people that show up to this channel
286. 52:09regularly are probably some of the most capable AI professionals in the game today, the generative AI game, like
287. 52:17using these tools for business and for personal growth and things like that. This is a community that's really got it going on. Um,
288. 52:28okay, cool. Danielle posted her peak in in irregulars. Okay, so let me show you how this how this thing works. So, so,
289. 52:35so I've made three videos. So, we're going to we're going to hit a plus sign. Okay. So, then it says take a close-up
290. 52:45selfie. So, you're not you're not taking videos. All right. Hang on. Let me get just calm down. I got to look pretty. Mama, am I pretty?
291. 52:59You have nice hair. Um, okay. Closeup. Okay, I'm going to go neutral here. Here's the thing I found about when you
292. 53:11do when you do photographs for avatars, your instinct is to smile because you've seen a camera before. Unless you're one
293. 53:20of those people that's just like a camera shows up and you're like, "Those people bug me." Anyway, those people are actually better for
294. 53:29avatars. So, when you when you take a a photo for an avatar, try to be neutral, but try to not look dead. So, that's a
295. 53:36it's an interesting thing. Okay, let's see. Hang on. I got calm down. Stop judging me, people. All right. [Music]
296. 53:49Okay, there I look crazy. Discard. Calm down. All right, that's good enough. All right, so there's my little picture.
297. 54:02And then we say next, select the sound. But we're not going to do that now. I can say over here on the side, I can say edit the look. Okay.
298. 54:13And so there's a bunch of looks down on the bottom. And then there's like hairstyle. So I can like change my
299. 54:19hairstyle. So if I want like a little cute little anime Bob. I can click the cute little anime Bob. And now what it's doing is it's doing an
300. 54:29AI character reference and giving me a a little That's good. I think I think solid solid. Hey everybody. Hello my friend Alan.
301. 54:43Let's pre p prejang it's his first live. Welcome Allan. Um what we're playing with right now is Pika social video or Pika social app.
302. 54:56Camera steal your soul. Okay. Um closeup not psycho. That's a good psycho picture. Now I think let's see. Um, we'll say uh man
303. 55:12with So now I can prompt it with light hair um in gas station
304. 55:29jumpsuit at a 1950s
305. 55:46gas station at dusk. All right, let's see if it I don't know if it's going to keep my crazy white hair. I think it will because I think
306. 56:00it's using that image to start with. Kyle, you can also use any style on a post you see. It's so easy. Oh, yeah. That's cool.
307. 56:17Um, white hair. Let's see. wearing a dark blue.
308. 56:38Jumpsuit is not the right word, is it? Jumpsuit. What's it called? Those gas station things. I don't know. Fashion. [Laughter]
309. 56:52You don't even have to type anything. You can just copy someone else's style and sound with your face. Oh, that's
310. 56:59kind of cool. Look, it made me like this old old man at a gas station. Coveralls. Yeah. Let's see. Let's try coveralls. Coveralls.
311. 57:16We'll do wide shot because I think I can do a wide shot. of an older man with light hair wearing dark blue
312. 57:30coveralls. We'll say pumping gas pumping gas into a uh 19
313. 57:4549 Plymouth. I don't know what a 1949 Plymouth looks like at a 1950s gas station at dusk. Okay, let's see what we get.
314. 58:04All right, how's it look? Looks pretty good. He's kind of holding the gas pump in his hand, but I think that does look like a
315. 58:121949 Plymouth. It's at least in the neighborhood. All right. So now, so now there we go. That's that's who's gonna be doing the talking.
316. 58:29So I'm going to record a little speech here. Mhm. I like ters. Taters. Wait, hang on.
317. 58:44Okay. Mhm. I like ters. Ters is tasty. You want me to clean your windows? I could clean your windows. I clean windows pretty good.
318. 58:59I like ters. Ters is tasty. Okay. I like ters. Ters is tasty. You want me to clean your windows? I
319. 59:15clean your windows. I clean windows pretty good. I like ters. Ters is tasty. >> Okay. Disturbing. Perfect.
320. 59:28And so now you just hit animate. Bang. Generate error. The network connection was lost. Well, don't do that. All
321. 59:36right. Animate in Q. Oh, hang on. Got to do physical dexterity on Tik Tok. Um, to Danielle's point, I'm showing you
322. 59:48sort of a manual way to do it where I'm like prompting and taking my picture and doing it. As you go through the feed, if
323. 59:55you see someone has done a video that's got a cool style that you like, every video has a button that says, "Make a
324. 1:00:01video of me like this." And then you just make it and it goes, "I like I like the way you talk. Ters or turtles.
325. 1:00:18Oh man. All right. And it does it takes a while. Like it sits in the queue for a while and then at some point it will um
326. 1:00:29it'll start processing. So I'll I'll just put this over here to the side and when it processes I'll show you. All
327. 1:00:36right. So while we're waiting, let's watch what Danielle did. So, so let me let's go watch what Danielle did and then I want to talk
328. 1:00:44about something. Okay. So, where we're going is to the AI salon. So, if you haven't been to the AI salon,
329. 1:00:54jump on over to community.thesalon.ai. You can go to just the salon.ai, but if you if you type in community.thesalon.ai,
330. 1:01:06AI. That'll take you over to our community website. It'll land it'll land you on this welcome page. You'll see our
331. 1:01:12cycle of AI readiness graphic there. Play first, mindfully create, generously lead. Um you'll see the five stages of
332. 1:01:19AI adoption. If you look at our values, you'll learn about you know what we care about and how we want people to behave
333. 1:01:27when they're here. And then if you scroll down on this lefth hand side, there's an area called clubs and hubs.
334. 1:01:32And a lot of these are part of the mastermind uh subscription area, but this channel has a channel called the
335. 1:01:40Irregulars, and it's free. It's open to anyone. And so here is a video that um that Danielle made. Okay. I think it's going to pop me to a
336. 1:01:53new tab, but I now know how to share that tab. Like that. Boom. Okay.
337. 1:02:20Beautiful. Okay. >> So, here's here's what I want to say about this.
338. 1:02:36Um, trying to think even what to say.
339. 1:02:51I saw some really interesting Tik Tok videos that were talking about people being upset that 4.0 was removed and
340. 1:02:58they demanded that they bring it back. If you pay attention to this space, you would think that the 700 million users
341. 1:03:08of Chat GPT are all up in arms about this model being taken away from them. Most of those 700 million people
342. 1:03:17probably were not even aware that it switched from 40 to five, right? They probably just use it. They
343. 1:03:25don't think about it. They're not paying attention to personalities or to features or to hallucinations or or
344. 1:03:30whatever. They're just using it. They're probably not using it a lot. They're not thinking about it a lot. And then there's a whole
345. 1:03:38massive amount of people beyond that that don't use AI at all or may they've maybe used it once. Those of us that are paying attention to
346. 1:03:51this world and paying attention to the capabilities of these things and thinking about the future, we're we're
347. 1:03:57we're in a very very small percentage of human beings that are doing this right now. our experience with AI and how we got
348. 1:04:08into it. Like we can a lot of us can trace it back to I remember the day chat GPT came out or I remember the first day
349. 1:04:15I used Chat GPT and you know I had my first Kevin Mallister moment right this moment. I had that first epiphany that oh my god
350. 1:04:24this AI stuff's different and I and I got on the train. There's very very few people that are actually on the train in that spirit.
351. 1:04:34Um, and so a lot of us are thinking about the on-ramps for AI being kind of a similar version to what we did.
352. 1:04:46That's not how AI is going to be adopted by a lot of people. There's going to be a lot of people, whether it's this Pika
353. 1:04:52app or some other app. There's going to be a lot of people that the first interaction with AI is just something
354. 1:05:00like this where Danielle made a version of herself like hanging out in the disco singing this Cindy Looper song, right?
355. 1:05:07Probably just at the push of a button. She pushed a button that said, you know, let me make a a vid video of myself
356. 1:05:12singing that song and bang, there it is. She didn't have to prompt it. She didn't have to do anything other than take a
357. 1:05:18selfie. You take a selfie. You kind of surf through the thing. You say, "Make a video with me in it." and Pika does its thing.
358. 1:05:27Right? If you look at how Pika is actually doing what it's doing, it's actually prompting. When you click on
359. 1:05:35those little pictures to like give me the short, you know, spiky hair, it actually wrote a prompt that said that that described the hair.
360. 1:05:48But a lot of people might not ever know that prompting is a thing. Their first experience with AI might be
361. 1:05:56talking to something that kind of sounds like Siri or Alexa, but it's just a little bit smarter and they don't even know that they're
362. 1:06:04talking to a large language model. They might use a tool like this where they get on. So, I don't know if I've got a a point to this, but
363. 1:06:15if you're paying attention to this stuff, start thinking about the different kinds of apps that you see and
364. 1:06:22the different kind of features. There are some apps like Peak is one in particular where they do a lot of
365. 1:06:27pre-prompting for you, right? They've got the this thing called peak effects where they like you upload a picture of
366. 1:06:34an object and it will explode it or you can cut it like a piece of cake or you can peel it like a decal.
367. 1:06:40Those are all just prompts into an image model. Like you could create pika yourself, right? You find an image generation, you
368. 1:06:51know, open- source model. You you figure out the API calls to it. You pre-write a bunch of prompts and stick pictures in
369. 1:06:58front of them and when someone clicks a picture, you send that prompt to the to the model and you return the picture.
370. 1:07:06It's a relatively straightforward thing. If you want to do your own little mini version of Pika, you could do that.
371. 1:07:14A lot of developers will make fun of those kind of applications, but I would argue that those kind of applications
372. 1:07:22are going to be a lot of people's first entry into AI. So, just keep paying attention to that stuff. All right.
373. 1:07:38What's that from? Is that from that interview, Brandon? That quote? >> Uh, no. This is I just sent this to you
374. 1:07:45in a DM on Twitter. So Sam, just before we went live, Sam made a lengthy post on X uh talking about the 405 controversy,
375. 1:07:55and this was an excerpt from that that uh seemed relevant to what you were talking about. >> Oh, cool.
376. 1:08:10>> Oh, great. Yeah, let's let's read this one. Share screen. Yes. Yes, I will. I will share my screen. I will share it now.
377. 1:08:21All right, good peoples. All right, good peoples of the internet. Do we have Yes, we wait. We have gas station guy.
378. 1:08:30Before we read Sam, before we read Sam, quick question on Tik Tok. Okay, new guy question. Allan, love new guy questions. Is Pika the
379. 1:08:42model behind Chat GPT? No. Pika Pika Labs. I think Pa Labs has their own video model, an animation model. I don't
380. 1:08:51know if their image generation model is their own or if they're using someone else's, but Pika's got an image to video
381. 1:09:00model. It's it's pika. It's pika.art is the website if you want to play with their video generation model. And it's
382. 1:09:09got all sorts of sort of pre-done things. And then they have two different apps. They have the Pika app and then
383. 1:09:16they have the Pika Social app. That's the one that I'm playing with right now. Um, that one's the Pika Social one's not
384. 1:09:22widely available yet. Um, but no, the the video model sitting underneath Chat GBT is called Sora. S O R A. Um, it's
385. 1:09:34okay, not great. Um, the video model that I'm enamored with right now is the one in MidJourney. So, MidJourney is an
386. 1:09:42image generation tool. So, if you go to midjourney.com, um you can then turn any image that you make there into a video. Problem with
387. 1:09:50Midjourney is it's not free and all of the video models are quite expensive. So, uh if you want to experiment with
388. 1:09:56them, it's kind of pricey, which is a drag, but all right, let's go look at ters. Ters is tasty. You want me to clean your windows? I can
389. 1:10:11clean your window. Wait, let me let me post it though. Wait, save draft. Caption. Caption is going to be Ters is tasty. Ters
390. 1:10:25is tasty. All right. So, we're going to post this. All right. So, now we're posted. >> Okay. >> Ters. Ters is tasty.
391. 1:10:39You want me to clean your windows? Can I go back? >> Clean your windows. I clean windows pretty good. >> I like ters. Ters is tasty.
392. 1:10:50>> All right, here we go. >> Mhm. I like ters. Ters is tasty. You want me to clean your windows? I can clean your windows. I clean windows
393. 1:11:03pretty good. I like ters. >> Okay. >> Ters is tasty. >> So that's that. Now that's pretty good. That's pretty flipping good. Now, notice
394. 1:11:13at the bottom it says make a video like this. So, if you have access to this Pika app, you can go to this and say turn yourself
395. 1:11:26into that character. You just push that button and it'll make a video like that. And I guess according to Danielle, it'll
396. 1:11:33even use that audio. Why wouldn't you want to use that Ters is Tasty audio? But look, it's kind of remarkable how
397. 1:11:40close it looks like me with the, you know, I kind of look like Robert Scoble there. And then you can just >> in my younger more [ __ ]
398. 1:12:06Let's see. Where's the one I just took? Oh, here's one. All right. So, now it's taking that picture of me
399. 1:12:20and it's applying that other dude's look. Glen Beck. Yeah, I look a little bit like Glen Beck. Oh, it it did something different. Discard.
400. 1:12:34[Music] Make a video like this. [Music] Oh, I see. Hang on. Let me go back.
401. 1:12:50Use all elements. All right. I guess I'll take a picture of me again. Remix error. The network connection was lost. Animate. Come on.
402. 1:13:21All right. My app broke. Whatever. It's not working. All right. Tik Tok pin. I see no pin. Oh, Joker. I love GPT5.
403. 1:13:32The coding is outstanding. I have an iPhone asteroids clone Kyle would like. Nice in HTML. Yeah, that's one of the other
404. 1:13:41things I was going to do is just go go do a little coding and just see what it does. Back to Sam. Okay,
405. 1:13:49so this was when did Sam write this? We don't know. Um, recently. Oh, today. Just before we went live. Okay. If you've been following the
406. 1:13:59GPT5 rollout, one thing that you might be noticing is how much of an attachment some people have to specific AI models.
407. 1:14:07It feels different and strong stronger than the kinds of attachment people have to previous kinds of technology. And so
408. 1:14:13suddenly deprecating models, this is what we were talking about earlier um that users depend on in their workflows was a mistake, right?
409. 1:14:25When you lose a feature in Gmail, you're like, "Ah, that sucks. I like that feature." But when you lose an
410. 1:14:34AI model that you've developed some sort of affinity and relationship with, it's more like the loss of a loved one.
411. 1:14:40Right? This is a very different I mean, there there have been examples of this, right? There are softwares,
412. 1:14:49Avatar software and things like that that when they shut down the service, people got really upset. So, we've, you know, we've seen we've
413. 1:14:56seen this before, but this is going to be an ongoing, increasingly complicated thing. Okay, this is something we've been closely
414. 1:15:08tracking for the past year or so, but still hasn't gotten much mainstream attention other than when we released an
415. 1:15:14update to GPT40 that was too sick of. Yeah, remember that? Remember when they did that? They they did, "Oh, you're
416. 1:15:20gonna really like the new GVT40 and and the press was all over it." This is just my current thinking and not yet an
417. 1:15:27official open AI position. People have used technology, including AI, in self-destructive ways. If a you
418. 1:15:35user in a mentally fragile state and prone to delusion, we do not want the AI to reinforce that. Most users can keep
419. 1:15:42clear a clear line between reality and fiction or roleplay, but a small percentage cannot. We value user freedom
420. 1:15:48as a core principle, but we also feel responsible in how we introduce new technology with new risks. I
421. 1:15:58there are a bunch of things I don't like about AI and I don't like about how things were trained. Um, I gotta say like I feel like
422. 1:16:09Altman at a minimum understands what the issues are. You may or may not agree with how he deals with them. I feel like
423. 1:16:16he at least understands the issues. Um, I don't hear that from a lot of If you listen to [ __ ] Zuckerberg, that
424. 1:16:21dude's like the the technology will save you. Everyone will have their own super intelligent person and we're gonna deliver it.
425. 1:16:31He's just completely cut off from these kind of issues. So at least Altman is speaking it. Encouraging delusion in a
426. 1:16:38user that is having trouble telling the difference between reality and fiction is an extreme case and is pretty clear
427. 1:16:44what to do. But the concerns that we worry mo most about are more subtle. There are going to be a lot of edge
428. 1:16:50cases and generally we plan to follow the principle of treat adult users like adults which in some cases will include
429. 1:16:57pushing back on users to ensure they are getting what they really want. A lot of people effectively use chat GPT as a
430. 1:17:04sort of therapist or life coach even if it wouldn't if they wouldn't describe it that way. This can be really good. A lot
431. 1:17:11of people are getting a lot of value from it today already. I hear that all the time. If people are getting good advice,
432. 1:17:18leveling up toward their own goals, and their life satisfaction is increasing over years, we will be proud of making
433. 1:17:25something genuinely helpful, even if they use and rely on chat GPT a lot. If, on the other hand, users have a
434. 1:17:31relationship with Chat GPT where they think they feel better after talking, but they're unknowingly nudged away from
435. 1:17:39their longer term well-being, then that's bad, right? So, you know, like in in that recent interview on on that podcast, I just I
436. 1:17:51just watched it tonight. He was talking about, you know, there are lots of ways that they could be manipulating users to
437. 1:17:57go deeper with these relationships that they're choosing not to, like they haven't made a sex bot, for example, is
438. 1:18:03one of the things he said. It's also bad, for example, if a user wants to use chat GPT less and feels like they
439. 1:18:10cannot, right? if they get addicted to it. I can imagine a future where a lot of people really trust Chat GPT's advice
440. 1:18:17for their most important decisions. Although that could be great. Wait, although that could be great. It makes me Oops. That thing finished.
441. 1:18:38Um, [Music]
442. 1:18:52all right. There we go. Okay. I can imagine a future where a lot of people really trust Chat GBT's advice
443. 1:19:00for their most important decisions. Although, okay. Yeah, we talked about all this. Okay. So we as in society but also as we
444. 1:19:09in at open AI have to figure out how to make it a big net positive. There are several reasons I think we have a good
445. 1:19:16shot at getting this right. We have much better tech to help us measure how we are doing than previous generations for
446. 1:19:23techn that technology had. For example, our product can talk to users to get a sense of how they are doing with their
447. 1:19:30short and long-term goals. We can explain sophisticated and nuanced issues to our model and much more. Um,
448. 1:19:38I'm gonna I'm gonna go out on a limb here. Here's where I think Open AI is [ __ ] up. Um, they're only hiring engineers.
449. 1:19:52They should be hiring anthropologists and liberal arts majors and storytellers and filmmakers. And they should balance
450. 1:20:05the logical thinking that is generating these remarkable tools with some some rightrained people with some
451. 1:20:15creative people with some artistic people some people in the humanities some people with phil philosophy degrees
452. 1:20:22because even though this is high math that is creating these things they are large language models. They are large language models.
453. 1:20:33How they manipulate us as humans is a core part of what they do, right? And if all of the frontier technology
454. 1:20:42companies are just hiring, you know, GPU engineers and coding engineers and mathematicians and researchers and
455. 1:20:51they're not balancing it with that other side, I think they're going to they've got a real chance of [ __ ] this up and
456. 1:20:57missing it. Now, the good news is there's a bunch of people out here who are going to use this technology and
457. 1:21:03sort of bend it in a good direction, but if they [ __ ] it up at the core model level, then there's not a lot we can do
458. 1:21:11about it. So, anyway, wait, go back one paragraph. You missed a critical middle sentence. I can imagine a future where a lot of people
459. 1:21:20really trust Chat GPT's advice for their most important decisions. Although that could be great, it makes
460. 1:21:27me uneasy. But I expect that it's coming to some degree. Oh yeah, I did. This is the important sentence that I missed.
461. 1:21:33But I expect that it's coming to some degree and soon billions of people will be talking to AI in this way. Yeah. So
462. 1:21:40we as society and we as open AI need to figure out how to make it a big net positive. Um, the whole the the musical that I've
463. 1:21:50spent the last year and a half working on is about our relationship with AI and what is it? What can it be? What
464. 1:22:00should it be? Is it possible to have feelings for an AI? Well, I think we just found that out with with GPT40
465. 1:22:08Side Hustle Mimi. I think we need to tell them the same thing that we told the Uno card creators. Thank you.
466. 1:22:17It's ours now. Yeah. I I I mean to a great degree that the the AI salon, the AI learning lab, a lot of times people that come into
467. 1:22:27this channel, they ask me this question. They they say, "Sir, what are your qualifications?" And what the
468. 1:22:34implication there is is, "Are you a mathematician? Are you a scientist? Are you qualified to talk about AI? Um,
469. 1:22:47yes. Because I'm not talking about AI. I'm not talking about model weights and algorithms in the latent space. What I'm
470. 1:22:55talking about is now that generative AI exists, now that the English language is our primary programming language to these
471. 1:23:06incredibly powerful knowledge engines. It's in the hands of the rest of us now to figure out how to actually [ __ ] use this technology.
472. 1:23:18That's what my focus is. I'll let the scientists go build it. I'll let them figure out all these issues of where are the safety
473. 1:23:26boundaries and what are they allowed to do, what are they not allowed to do. Let's assume that they're going to
474. 1:23:31they're going to figure that out. I personally think they need to hire some liberal arts majors soon or they're going to [ __ ]
475. 1:23:39that up. But once the models come to us, then it's up to us, the 98.5% of people that are not scientists and
476. 1:23:49engineers, it's up to us to figure out how do we use these things? How do we use them for good? How do we use them to
477. 1:23:55improve ourselves? How do we use them to amplify our ideas? How do we take a virtual 100 employee company
478. 1:24:06where we've got a hundred employees that are all PhD level employees and use that for ourselves for our own ideas and use
479. 1:24:15that for good in the world? I don't know yet. None of us do. But that's my interest is what are we going to do now
480. 1:24:23that we have it in our grubby little hands? What do we do? How do we do it? How do we make a difference?
481. 1:24:34Just keep talking about it is the the only way I know right now. All right, that's that. That's fascinating.
482. 1:24:42Fascinating. It's fascinating, people. I'll tell you what, it's really fascinating. Fascinating. All right, let's go look at my video.
483. 1:24:56[Music] Uhoh.
484. 1:25:14I run like JD Vance. Uh, yeah. Charact Korea character reference. I haven't played with it enough to know I'll just get go in there
485. 1:25:27and get lost. A lot of these character reference things are very they're very UI specific. So, um, okay. Side hustle Mimi, I will check my
486. 1:25:41DMs. Yeah. So, by the way, if you are on the AI salon, if you're not on the AI salon, go join the AI salon. Um, I've
487. 1:25:49got three invites, I think, for the Pika thing. So, if you DM me, um, whoever whoever gets there first, I'll get you a
488. 1:25:57code. All right, that's another reason to come here, people. [Music] All right.
489. 1:26:16It's a quick demo. Where do I go for it? Is it basically just like what chat GPT did?
490. 1:26:31All right. So, so I'm at gemini.google.com. Click the three dots. Guided learning.
491. 1:26:47Okay. So, yeah. So, so both um both OpenAI and Gemini have just added these these new things um guided learning. So, learn and
492. 1:26:57understand topics. Teach me teach. Why can I not type? Oh, wait. Do I have to Oh, let me choose
493. 1:27:11guided learning. Oh, I can't do it with that account. Sorry. As a paid member of Google Workspace, you can't use any of the new
494. 1:27:21features. You're welcome. No thanks. Guided learning. Guided learning. Okay, I've got guided learning on. So, I'm
495. 1:27:35going to say teach me about No, I'm not going to say teach me. I'm going to say um how do
496. 1:27:49>> how to change your tabs in Google Chrome. >> Wait, but I didn't change them. I just typed Oh, cuz when I changed the counts,
497. 1:27:59it moved me over there. Unbelievable. All right. How do I find the average? Was it the average speed of a something swallow
498. 1:28:13of a who knows the Montipython of a swallow? I forget the Let's see if it gets the reference. Slow thinking. It's doing slow thinking.
499. 1:28:30Gemini is a beast. Unladen. That's right. Oh, that's a classic question. To answer it, we first have to know are you
500. 1:28:37are you asking about an African African or European swallow? We'll go European and wait eur damn it European and unladen.
501. 1:28:57Got to DM me. Ah, a connoisseur of classic film in Greece. Wait, based on Oh, did I not have that
502. 1:29:14turned on? No, I do have it turned on. Um, teach me how to do it. Let's Let's actually go go compare this to
503. 1:29:27OpenAI because this this Gemini thing just it just gave me the answer. It didn't teach me. No, they got to DM me on the aselone. Sorry.
504. 1:29:45Um, direct officer. Wait, this is not very good teaching. Let's go uh let's go back to chatbt for a second.
505. 1:29:59Well, it what we're seeing with the Gemini model, Kyle, is more of a a professor at trying to tutor a student.
506. 1:30:10So, it's taking a concept that you have some familiarity with and trying to coach the answer out of you. So, do you
507. 1:30:16think it's A or would you think it's B? And then if you choose wrong, it would course correct. >> I get that, but it didn't do that here.
508. 1:30:32It just gave me the answer. Let me let me try I want to try the same question actually. Let me go get the
509. 1:30:46original question. Um Oh, I don't have it. I lost it. Did I? Yeah. Here we go. All right. Let me go to chat GPT and turn on its study mo
510. 1:31:01mode and see if it's different. Study and learn of an unladen
511. 1:31:16swallow. [Music] It's also amazing to me like how like feature for feature all these frontier
512. 1:31:32model companies are copying one another because this study mode came out from chat GPT before we dive in. Quick note I
513. 1:31:40can't show my private SCA scratchwork but I'll walk you through the steps and thinking see this is this is what I was
514. 1:31:46expecting it to do and this is what what Gemini didn't do. Here's the idea we use. Average speed equals total total
515. 1:31:53distance divided by time. Straight up physics. You turn one bite-sized step. If a swallow covers, 1200 meters in 60
516. 1:32:01seconds, what's its average speed? Just divide and tell me the number in units. Do the math for me. All right. Well, crunch the numbers
517. 1:32:13together. Yeah. And it's giving you math things. So, I I'm sure the Google one's fine. It just the the way I ph phrased that
518. 1:32:23question, it just answered it for me. So something was wonky there. But anyway, both of these now, Gemini and Chat GBT
519. 1:32:32have this study mode. So if you've got kids or grandkids that you want learning stuff, there you go. All right. All
520. 1:32:41right. All right. All right. Fantastic. Be right back. Bedtime for the wild things. Oh man, are we DMing through
521. 1:32:54Mighty Networks? Yes, DM me in Matetwax. Mata Networks. [Music]
522. 1:33:11All right, everyone. G. Redditors are not happy. Yeah, Redditors are never happy. They're just over there bitching. Again,
523. 1:33:26don't forget, Redditors that are paying attention to every permutation of what's coming out of the Frontier model companies are not
524. 1:33:35most of who's going to use Chat GPT. They're I I would argue they're not most of who's using Chat GBT. You got 700
525. 1:33:42million weekly users using chat GPT. If you think anywhere north of 5% of those 700 million users are using Chat GPT for
526. 1:33:55anything other than glorified Google search, you're high on crack. There's no way. So, even though Reddit's noisy and
527. 1:34:03X is noisy and people are bitching and Open AAI is responding and restoring the model, um, just give it some time. I the the the
528. 1:34:15everyone jumping to conclusions. I swear to God, Gary Marcus, the minute Chat GPT5 came out, he jumped on it like like
529. 1:34:22he was literally doing victory laps. See, I told you everything sucks. It's all [ __ ] awful. You're all [ __ ]
530. 1:34:28losers. You promised it would do stuff. it doesn't do it. It's like this the stuff is emerging and evolving, you
531. 1:34:37know, in our presence and it's doing remarkable [ __ ] and every time a new model comes out, it's going to take us a while to
532. 1:34:45understand what it actually is. Um, so all of the drama right now, just ignore it. It's all just [ __ ] Um, what's
533. 1:34:55the airspeed velocity of an unladen swallow? That's that's what I asked. I didn't ask it right because I I'm an old
534. 1:35:01man and I couldn't remember the actual line for Monty Python. What is the airspeed velocity of an unladen swallow?
535. 1:35:14We are the knights who save me. It was horrible. Came out of the cave and big gnarly teeth. It's just a scratch.
536. 1:35:30Oh my god. All right, everyone. Um, tomorrow's just a regular Tuesday. I don't have a salon or anything. So, um, I'll be here at 8
537. 1:35:41o'clock tomorrow night. Um, I will probably pay a bit more attention to things going on. It seems like some things are perking. So, I may
538. 1:35:50I may have more of a an idea of what to talk about tomorrow night, but I don't know. Maybe not. We'll see. All right.
539. 1:35:57Fantastic everybody. You call that a flesh wound? Just a flesh wound. I'll bite your kneecap. Such a good movie. It's such a good
540. 1:36:09movie. Oh my god, it's a good movie. I haven't seen one of those in a long time. I should probably go watch that. Um,
541. 1:36:20when is the AI jam session on your links? Oh, the AI jam sessions. Those are private one-on-one meetings with me. Those are um I've got four
542. 1:36:31hours reserved on Saturdays. Um 10 11 and then one and two mountain time. Um so if you want one-on-one time, if you
543. 1:36:43go to my Tik Tok channel and click on my profile, I've got a hyperlink there that will take you into the jam session and
544. 1:36:51you can subscribe. You can you can book an hour or you can book 10 hours. Um but it's it's individual hourlong sessions.
545. 1:37:00So if you want to do that would love that. Um fairy tale revolution. I learned it from you Kyle. I don't know what you learned
546. 1:37:12from me but I hope it was a good thing. Oh man. Oh that's Claire. Oh cool. Dr. J. Nice. Beautiful. Beautiful.
547. 1:37:23Beautiful. All right. Always looking. Always look on the bright side of life. [Music] [Laughter] All right, everyone. I will see you
548. 1:37:36tomorrow. Hope you have a good night. Uh, it's good to be back. It's weird to be back because I took five days off and
549. 1:37:43I haven't done that in forever, if ever. All right. Peace. See you later.