7/13/2026 - Navigating the Chaos of ChatGPT's New Work and Codex Modes

Transcript

1. 0:01[music] [music]
2. 0:17[music] [music]
3. 0:32Good evening, good people. What's happening? Welcome. Do I need Do I have schmutz on my lens? I may I may be positive. Hang on.
4. 0:43I think that's better. Yep. That was I don't know if you noticed that, but I used a professional lens cleaning rag
5. 0:53called my shirt. Called my shirt. Good evening, good people. Okay. Now, what you're going to do here is in the upper leftand corner right up
6. 1:05here is the button that pops out the chat. So, if you have questions or comments um or hair product suggestions for me,
7. 1:17you can you can put those in the chat. Um both Brandon, producer Brandon and I can see those chats. Um so, you put them
8. 1:24over there. So, if if you have the chat open, say hi and I'll say hi back. >> [laughter] >> or if you just listen, it's okay if
9. 1:32you're lurking. If you're a lurker, that's fine. But, you know, isn't it more fun if you like show up? Um, okay.
10. 1:40So, that's that. Mary Chester, your hair is perfect. Well, thank you. Thank you. I don't like to brag, but you know, if you got it. Um,
11. 1:51okay. In this corner down here is a cool little feature where you can pop out the video. and pop it out. So, if you pop it
12. 1:59out, you can open a second tab and you can surf the AI salon and you can pf the AI surf the AI salon. Check those spaces out and go chat and
13. 2:10things like that. Mel, what's happening? Good to see you. Thank you for saying hi. Hello, everybody. Um, so that's it.
14. 2:18We've got there there's so much going on in the salon. If you have not been in it in a while or if you're new here, go
15. 2:27check out the events tab and go check out the spaces. So the spaces down the left hand side, think of those like
16. 2:34think of the AI salon is like a big hangout restaurant and the spaces down the side are like differentiz tables
17. 2:41with people hanging out talking about different things. It's like it's like a speed dating mixer, although there's no there's no
18. 2:49speed element to it. So, it's like a dating mix. It's a bad metaphor, but but anyway, you go hang out with different people in the different
19. 2:58places. And what you find in this community is that the people are very, very generous. They're very sweet and
20. 3:04they're very smart and they're very curious about AI. And if you have a question, they're willing to help you.
21. 3:10Even if that question is like, uh, where do I start? They don't make fun of you. It's weird. It's unlike school.
22. 3:24The AI salon is for people curious about AI but hated school politics. [laughter] It's not weird, it's irregular. Linda
23. 3:35Sherman, welcome from Kawaii. Thank you for for for hanging out. I love seeing you here. Linda and I met at Social
24. 3:43Media Marketing World in Anaheim, California. I spoke back in I think it was April and we met then and she's been
25. 3:49hanging out in the salon and coming to events and it's just great to see you here. It's great to have you here. Um,
26. 3:55okay. So, I'm going to talk for a little bit. And I know when I say I'm going to talk for a little bit, that could be 45
27. 4:02minutes. It could be three minutes. We just don't know. It was funny. I was on a call today. We were doing a practice
28. 4:08run. I've got a pitch tomorrow for Sydney, my musical. So, we're pitching a theme park. You're welcome, Linda. It's great
29. 4:18to see you. Um, so we got a pitch tomorrow. So, we were doing a runthrough of it today, and I actually took the
30. 4:24time to write out, you know, talking points and little points along the way. And the feedback I got from our little
31. 4:31run through today was just be yourself. Don't read stuff because they like there like when I talk on my own, I talk I can
32. 4:39talk good words good from the mouth. But when I write things down and try to say formal things, doesn't go so well for
33. 4:47me. So they they were like, just be you. [laughter] So that was nice. That was good reinforcement. Um, okay. Here's
34. 4:55what we're going to talk about. I want to talk about the dynamics of the frontier labs. Right now, I would consider us having
35. 5:07two frontier labs, two half frontier labs and and a a wannabe. If you don't know what frontier labs are, they're the people that have a
36. 5:20billions of enough enough billions of dollars to be able to like lead the charge with AI. The two frontier labs are Anthropic and
37. 5:30Open AI. And Anthropic, as you know, was founded by Daario Amade, who quit OpenAI. He was a founder of Open AI. He
38. 5:37quit there because he didn't like how they did safety. And so, Anthropic is the answer to what he felt were the sins of Open AI from a
39. 5:48safety perspective. Then you've got SpaceX AI, the rocket slash AI porn slash large language model agentic AI company
40. 6:03[laughter] and you've got Google who actually invented the transformer. They're the reason all of these
41. 6:14companies are doing what they're doing. But SpaceX AI is making more money right now um renting out their CPU or their GPUs
42. 6:26than using them. And Google right now, Sundar Pachai is just is like, "Hey guys, just let Open AI and Anthropic
43. 6:38screw themselves up with the government. We'll just sit back. We'll just sit back." So, I think Google is probably
44. 6:45sitting on some cool technology that they're just they're just taking a time out. They're just off in the corner.
45. 6:52We're good. Probably, you know, laying the the regulatory foundations to be able to do whatever the [ __ ] they want to do.
46. 7:00And then the fifth one that's just not a part of it anymore is Meta. Now, Meta did just come out with an image
47. 7:07generation model that's decent. Um, and I think they might have even even come out with some new LLM model. Perplexity,
48. 7:16well, well, Perplexity is certainly not a frontier model company. They they're an application layer sitting on top of
49. 7:23lots of other models. So, Perplexity is not. Frontier literally means they're on the frontier, the cutting edge of of the
50. 7:29things. MidJourney, you could argue, is for images, but but it they're a much more constrained company. Um, they're
51. 7:40also not publicly funded. They're privately funded. They're 100% owned by the CEO. It's pretty impressive. Um,
52. 7:48images and and MRIs are apparently not Frontier. Not according to [laughter] not according to the people that call
53. 7:56frontier companies frontier companies. Um, yeah. I mean, you know, frontier companies are the ones that are going to
54. 8:05be worth trillions, right? or already are. Okay. So, you take the three that are sort of half frontier or or not in Meta's case
55. 8:17and you just set them aside. So, right now we've got these two. There's a there's a cynical part of me. A lot of
56. 8:34me is cynical. There's a cynical part of me that looks at what Anthropic's been doing for the past three or four months. Daario
57. 8:46Amade started shifting his tone probably back in January, February. Um, and started shifting his tone to
58. 8:57things are dangerous. These things are dangerous. These things are dangerous. Got to be careful with these things. this thing I'm inventing
59. 9:07that I'm about to show the world. It's kind of dangerous. Um, it's possible that's actually genuine concern
60. 9:27and he's purely interested in safety and results be damned. But the cynical part of me just says if we make this thing, okay, you know Red
61. 9:42Bull, you know why Red Bull does extreme sports? Do you know what their initial marketing plan was? It just might kill you.
62. 9:58Red Bull's marketing was essentially, you want to be careful with this. It's such a powerful energy drink, it just might kill you.
63. 10:12So, let's call anthropic the Red Bull of the Mark of the AI space. It just might kill you. [sighs] So anyway, so they're out there, they're
64. 10:28doing all this stuff. They've got the Mythos model. They don't launch that because it's too dangerous. Then they
65. 10:33launch Fable. Then the government says, "Oh, it's too dangerous. You have to pull it back." Then it was three weeks away.
66. 10:39There's worse ways to go. [laughter] Um, and so then they gave us back they gave us back Fable and they said you can have it until the
67. 10:518th and then they said you can have it until the 12th and now they said you can have it until the 17th whatever whatever
68. 10:57the dates were. So 7 days 7 days 7 days. The reason they keep extending it is because Open AI launched
69. 11:06GPT 5.6 Six. So soul, not [ __ ] out of luck. Soul as in the sun, but it might be [ __ ] out of luck. Open AI is like uh
70. 11:24I guess they're like Dr. Pepper. Okay. Still independent sort of. They're not Google. They're not Meta, right? Coke and Pepsi.
71. 11:37They're Dr. Pepper. They got their own little secret sauce that seems to keep them slightly ahead of everyone else.
72. 11:43You got anthropic Red Bull. It's going to kill you. GPT 5.6 is good enough and stable enough that apparently customers are flocking from
73. 11:58Anthropic back to Open AI. So, [laughter] so they keep extending it the the access to Fable. So, so that's going on. Now, one of the rumors
74. 12:14for the new OpenAI 5.6 six model was that they were going to launch the everything app where they took their web browser Atlas
75. 12:26and they took Codeex, their coding platform, and they took ChatGpt and they rolled them into a single app. Close. Close.
76. 12:40Good call on changing it to it'll give you wings. It'll give you wings, right? It'll give you wings. You're gonna go to heaven
77. 12:49on wings if you drink this energy drink. [laughter] It's marketing, baby. [laughter] It's starting to look real as wrestling.
78. 13:01How close they are. Danielle, fun fact, they had to change it to wings after they got sued. Oh, that's hilarious. Um, okay. So um
79. 13:15so now we have the new chat GPT. It is not the everything app. It is the everything in the sync series of apps that are all different.
80. 13:31So what we're going to do tonight is I am going to try Uhoh. Google. Oh. Oh. Producer Brandon entered the studio. I guess he had to
81. 13:42leave. Okay. He's He's fine. He's fine. Okay. So, I'm going to start playing. And I Now, first thing I need to do is I need
82. 13:52to go get the chat GPT app because I thought I had the chat GPT app before I came on and then it said it was thrown
83. 14:00in the trash because it had malware in it. So, so I don't know what that was about. Uh, chatgbt. uh Mac download chat GPT and I'll share my
84. 14:17screen in a second here. Okay, where are we going? We're going. So, what we're going to do tonight is
85. 14:26we're going to go spelunking. I'm going to show you an image that I created using chat GPT to explain all of ChatGpt.
86. 14:40Um, Robert Scoble said it was helpful today. No screen share. I know. I haven't I haven't shared it yet. Um,
87. 14:51okay. Let me go do something here.
88. 15:06What's going on? Okay, we're installing chat GPT and applications. Yay, spelunking. [laughter] I was talking on the preview show about
89. 15:22remember if you're if you're Gen X, remember when they would just kick us out of the house? I I I remember three distinct times in my life
90. 15:32where I was kicked out of the house. That was every night, right? You were just like, "Just get out. Go. We're
91. 15:39watching TV. The news is on. Just get out. Go. What am I supposed to do? I don't just get out without the cave. Greatness." So, yes.
92. 15:53Yes. Spelunking without the cave. So, I remember three different occasions where we went and found caves.
93. 16:01One of them was a fox den. We crawled down in it. Ah, I found Kyle's OpenAI infographic. That's the one I'm going to be talking
94. 16:13about. Yeah. So, so beautiful. I don't know if there's a way to click on that link. Um, all right. Let me share my screen.
95. 16:23I'm going to share my whole screen. We'll see. We'll see how long this the sharing lasts. It might be shortlived,
96. 16:29but I I tried to delete some stuff off my hard drive, which I think has been the culprit. Okay. So, if I go here and go here. So, here
97. 16:43say copy image and go preview. Go new. Okay. So now we have this image.
98. 16:58Okay. Before the launch, chatbt was mainly a chat experience. You could access strong models there, but there
99. 17:05was no separate work mode. There was access recently to codeex in the Oh, that's web. Yeah, that's right. Web Web
100. 17:14was just web. That's right. Web was just chat, right? Desktop. There were effectively separate desktop
101. 17:22experiences. There was the chat GPT app and the codeex app. Correct. And they they renamed the codeex app chat GPT
102. 17:32retired chat GPT and then polluted codecs with a bunch of other [ __ ] which we're going to go that's where we're
103. 17:37going to go spelunking. And then mobile chat GPT mobile app focused on chat while Kodak Access felt like a separate
104. 17:46coding experience. Well, you could just send it messages. Overall, chat GPT and Codeex felt like parallel products
105. 17:52rather than one unified system. Okay, so the web now has chat mode and work mode. No codeex mode on web. That's not Oh, on
106. 18:03web, right? Okay. Oh my god, I'm I'm so confused just holding this together. Let me make sure that I'm actually still
107. 18:12sharing this. Yes. Okay. I'll get an SOS if if it's not. So, okay. So, the web experience, what used
108. 18:21to be chat GPT is now chat GPT and work, but no codecs. Got it. Standard chat. Okay. Here's the other thing. If you're
109. 18:37in chat mode, you can only access chat GPT 5.5. If you're in work mode Oh, no. No. Sorry, not true. Shh, calm down.
110. 18:52If you turn on reasoning in chat, it will upgrade you to 5.6 soul. And soul, Luna, Terra, and Luna are the three different sizes of 5.6.
111. 19:05Good lord. The desktop app. There's a new unified chat GPT desktop app that's got three modes. chat mode, which is
112. 19:15weird now, work mode, which is is the default, and codeex mode, which as far as I can tell, the only difference
113. 19:23between work mode and codeex mode is that codeex mode has pull requests, which if you don't know what they are,
114. 19:31you don't care about codecs anyway. So, chat is standard chat GPT conversations. Work is research, docs, spreadsheets,
115. 19:40presentations, reports, slides and work can use local files and desktop apps with permission and then codeex is
116. 19:48software development with files, repos, terminals and dev tools. So what is the difference between software development and and using local
117. 20:00files and desktop apps? because you could use the same apps in both work and codecs. So I think it's just one menu
118. 20:08item that makes the difference. And then the mobile app has got chat and remote which lets you talk to codeex or I
119. 20:19assume work. And then it says it's rolling out work. Good lord, it's a disaster. Okay. And what's this say over here?
120. 20:30The direction of travel travel interpretation based on current announcements. One umbrella open AI appears to be consolidating around chat
121. 20:38GPT as the main umbrella product. Chat as the everyday layer. Chat stays familiar through the conversational
122. 20:46front door. Work is the general purpose agent. Work becomes the place for longer tasks, connected tools, polished outputs. Codec
123. 20:55is codeex is the specialist. It remains dedicated to software development. Okay. Okay. All right. Good lord. Are we still sharing? We're
124. 21:09still sharing. It's a minor miracle. Okay. Let's see. Uh some questions. It's not clickable the link.
125. 21:19Uh if you go to my um if you go to my X profile, you can find that graphic. And I also posted it in the AI salon in uh I
126. 21:28think it was in community chat. Codex can do everything and work on your computer. I love it. Use it constantly.
127. 21:36I'm using it now. I like codecs a lot too. I still don't quite know the difference between work and codeex right
128. 21:41now. I I think I do and I'll tell you in a second. I like using chat within codeex. I can send the whole
129. 21:49conversation to a codeex project thread. Okay. Okay. Hang on, Charles. So, okay. It's going to be fun.
130. 22:04All right. My hair I'm I'm a little moist. I like using chat within codeex. I can send the whole conversation codeex. Okay.
131. 22:18I'm gonna do a series in the AI salon mastermind in the great repurpose. So, one of the things that I've been
132. 22:28talking about for the past week or two is reclaiming our digital selves. Figuring out how to work with these
133. 22:33agentic tools and giving them enough context about who we are, how we like to work, how we want them to work, and the
134. 22:40projects we're working on that it can be super useful for you. Right? I'm learning a couple of things. One is
135. 22:49my brain doesn't instinctively work the way these systems work. If you're a developer and you you're used to sort of multi-headed
136. 23:03software tools strung together like I've got github over here for my soft for my repository. I'm pulling models from
137. 23:11hugging face. I'm I've got a internet or interactive development environment, an IDE over here. I've got, you know,
138. 23:21tools to help me code. And if you're used to multi-tool chains strung together at, you know, with you as the orchestrator, then these
139. 23:34things will feel pretty good. They'll feel pretty normal. Um, if you're a human being, listen, I I know I just What I just said was I said
140. 23:47engineers are not human. I did it. I did it. I know it's triggering. I'm Gen X. We're excused. [laughter] Charles Adams, I have a brain for
141. 23:59modeling other people's psychology and writing. I've got like seven novels in production and all the plots are
142. 24:05perfectly clear to me. Yeah, that's actually pretty cool. Kelly Camp, what's happening? But tech is not my bag. But
143. 24:10you've got a systems thinking brain. This is something that I've been talking about with Andy is she's got a systems
144. 24:15thinking brain. I appreciate systems, but while I can conceive them and even design them, I don't naturally think
145. 24:23that way. And I don't use the systems once I create them. I don't like them. It's not my like once the systems
146. 24:29created, it bores me and I don't want to go into it. So, so my brain is not instinctively wired to think the way these tools think. So,
147. 24:40I'm doing a series of workshops on creating your digital self. And here's what I've come to learn other than I don't instinctively work that
148. 24:52way. Everyone has a slightly different approach to this. So, Mel is using it and talks about how she's using it.
149. 25:00Charles talking about how he's using it. Um, Gareth talks about how he uses it. Brent Peterson talks about how he uses
150. 25:06it. So, what I'm actually going to create is I'm going to do a couple of interstitial workshops where I lead them and then I'm
151. 25:15going to have experts come in and we're going to do an expert series. Um, so if you know this stuff and you feel like
152. 25:25you've got something to teach, it's going to be kind of like teaching an LOL, but I want to bring in experts in
153. 25:30this series because I think setting up agentic systems, there's no single way to do it. There's lots of ways to do it.
154. 25:37And so I think the more people we can get exposed to just looking at how their brain works, looking at how they've set
155. 25:44things up, understanding what works for them and what doesn't, I think it's going to be really important. So, I'm
156. 25:48super excited about that. Okay, [sighs] let's see. Kelly Camp, I've been building my first Codeex KODEX agent and
157. 25:58about to wrap up. Congratulations. I've always wanted the rules the rules lawyer in a DND game. That's really cool. Okay.
158. 26:13Um, what we're going to do is we're going to go play. So, let me see if Well, let's let's just go to I think the
159. 26:26first place to go to is let's go to chat GPT. So, like I said, I'm going to go spelunking. All right, I think you can see that.
160. 26:37Wait, let me come back here and pop out the chat so I can see the chat. Uh,
161. 26:51okay. We're going to go to new chat. We're on new chat. We're in chat. So, the first thing to notice about the web
162. 27:05is that at the top of the screen, you have two modes. You've got work mode and you've got chat mode, right? Create an image, write or edit,
163. 27:16look something up. Let's go to the plus sign and see what's changed there. Okay. Add photo. So, upload something,
164. 27:25create images, web search, deep research, Google Drive, which I think I've connected. Now, what I don't know.
165. 27:36So wait, let's look here. So I've got Google Drive, Google Calendar, Notion, and GitHub connected. Let's go to the
166. 27:43work side and see if it's different. Plugins. Google Drive, Calendar, Notion, GitHub, Gmail, Sales, Runway, Canva. Okay. Okay.
167. 27:54So chat has plugins that appear to be a different database than chat GPT work.
168. 28:08All right. So if you took all the time to set up all your connectors in chat GPT, it looks like you're going to have to reset those
169. 28:16up and work. And I would guess at this point probably reconnect them in codecs. Oh, here's my I I'll I'll save my theory for where
170. 28:30this is going until I open the chat GBT app because I because I have a I have a thought and oh, okay, over here. So,
171. 28:37we're in work and it says right here you can open the desktop app and you can't do that here, right? So, these are just ideas.
172. 28:47Messy mode the pin collection handwritten style cross-section. All right. So, this is trying to be a consumer app, right? So, if you say
173. 29:03create image, I should be able to get rid of that. Yeah. But if I say create image, this is trying to be a consumer app, right? You
174. 29:12don't have to think. You just want to let's see. You just want to create a tarot card. You click on it and it writes the prompt for you.
175. 29:28Create a tarot card of a unicorn in a popcorn basket. based off of what you know about me. All
176. 29:45right. [laughter] Okay. So, Chat GBT is trying to be a consumer app. That's important. It's not. Don't get me wrong. Chat GPT
177. 30:00used to be a consumer app. Chat GPT used to be the place that you can surf and just figure it out on your own.
178. 30:09It's it's it's now too complicated. It's it's officially bloatware. All right, that's doing that. Let's go look.
179. 30:22So, the library is things I've created or things I've uploaded. Okay. Scheduled. Okay. So, now scheduled.
180. 30:35Schedule a task. Schedule a daily brief. schedule an email monitor. One of the things that you can do in chat GPT work
181. 30:41and in codeex is you can create things that happen on a regular basis. In codeex they're called chron jobs as in
182. 30:49chronological. In work I don't I I don't know if they're called automations or scheduled. Let's let's see are they
183. 30:56different in work? If I go to work they still stay scheduled. Okay. So I guess they're scheduled. [sighs]
184. 31:10Huh. Pharma video market scan. So this is interesting. So So chatpt knows that I've recently started fly fishing again
185. 31:23and it knows that I do work in the pharma industry. So it it has dynamically suggested Huh? So, let's let's create one of these.
186. 31:40So, I hit plus. So, all of these pre-written things are basically just prompt front ends. You should deliver it at
187. 31:51when on Monday should I deliver it? Well, we'll do 9:00 a.m. How deep should each brief go? standard. What developments should get extra
188. 32:05weight? Competitive product launches. We'll go pharma workflow changes, sales and partnerships, AI video model advances. Next,
189. 32:24creating a scheduled t creating a scheduled task. [snorts] So, am I in chat GBT work? We don't know. I think I was in chat
190. 32:40Mondays. There's the There's the prompt. Weekly Mondays 9:00 a.m. End repeat never. So, this is an ongoing thing.
191. 32:51Brief me. 8 to 12 consequential developments in AI enabled video, pharma and healthcare marketing. MLR safe
192. 32:58content workflows customer-f facing video platforms that could affect storyines's product sales or positioning
193. 33:06give extra weight to pharma mlr workflow changes sales and partnership signals and ai model advances video model
194. 33:13advances for each item explained what changed why it matters for story mine so this is a pretty decent prompt
195. 33:22and happens I'm going to say can you run, let's say, run that right now and email me the results rather
196. 33:38than have me get them here. If you can email it, add it to the scheduled
197. 33:53event. Okay, let's see. Let's see how good this is. The thing that's got me weirded out
198. 34:08is that this is like a a chat GPT work kind of feature. This is like a codeex kind of feature. This is creating an agent. It's
199. 34:19well, it's creating an automation. At a certain time, it triggers an agent to go do some things, but it's in the chat. It's in the
200. 34:30web-based chat application. And I don't know if anyone can remember if I did this with uh if I was in work or chat mode when I
201. 34:42started this. Updating the scheduled task. Can I click on this? No. What can I do here? view sources, branch
202. 34:52into a new chat, read aloud. Okay, the other thing I've noticed about chat GBT is it hangs a lot for the past two
203. 35:02weeks. So, I would imagine that their their servers are getting slammed. Okay, while that's doing that,
204. 35:17let's go to a new window. Chat GPT. I'm going to go to work. I'm going to say open desktop app. Bang.
205. 35:32Always open this. Open codeex. Oh, that's interesting. There was a bug. Quality control didn't catch that one.
206. 35:41Update required. Update codeex. The version of the app no longer supported. Wow. Excuse me. Yes.
207. 35:56Yes. What have I done? [laughter] Excuse me. Angry Gen Xer reporting for duty. Opening chat GBT. So if you open chat
208. 36:10GPT, it says it's opening codeex. Update codeex. So wait, so I still have
209. 36:26codeex installed. Update codeex. Now I should be able to just go to chat
210. 36:42GPT the app and open it because I already have it. But they don't have that wired up correctly.
211. 36:57Okay, good time for a commercial. Okay, hang on one sec. Yeah, so there's chat GPT codeex is now the chat GPT app.
212. 37:08Okay, so they still don't have their [ __ ] together. Like within the web chat GPT work, you click on launch the chat GPT desktop app. It
213. 37:21takes you to codeex which says your codeex is old. Update it here and it turns it into chat GPT. Good lord. Okay. Um
214. 37:34I wonder if it downloaded the codeex app. I bet it didn't. That's okay. Okay. Doesn't [ __ ] matter. All right. So, we're gonna come back to
215. 37:42this and then let me just go look quickly here and see if it finished. I'll run the brief now. Email it. Oh, look. It's asking
216. 37:53always allow. Okay. Allow chat GPT to use Gmail. Send an email to kytovine.com. Storyine strategic intelligence brief
217. 38:01July 13th containing detailed strategic. So, yes, always allow that. And then let's go see if it sent us an email
218. 38:13and then I'm going to tell you about something and then we'll come back to this. I told you we'd be spelunking.
219. 38:20We're finding stuff. I've updated the recurring task. I'm now distilling the highest impact developments into an
220. 38:26emailed brief rather than patting it with marginal news. Oh my god, I hate it when chat GPT does that sort of
221. 38:33[ __ ] Okay, what's going on over here? I would love it if many AI salon members would complete the social profile links
222. 38:42in their profiles. I find most of them don't include those. That's really good request, Linda. So, especially if you're
223. 38:49an irregular here, and especially if you're an AI salon ambassador, either one of those two, do me a favor. Go into
224. 38:58the AI Salon into your account and and add your social channels into your profile. I think it's a really good practice.
225. 39:07That's a really good request, Linda. So, thank you for requesting that. That's good. Mr.'s here. What's happening?
226. 39:14You get the same message if you're downloading the chat GBT app for web. I took a bunch of screenshots. I know this
227. 39:21is probably obvious, but it feels like we're being led to agents eventually doing everything. Yes, Kelly Camp. So,
228. 39:27I'm going to get there in a second, but yes, everything will just be a simple prompt. Yes, Patty McKenzie. Oh. Oh.
229. 39:36Uh, Linda liked me requesting that you guys fill out your social channels. Listen people, how many people are here?
230. 39:44Hang on. We're going to come back to my little brief here and see. We We got to have a conversation. Listen to me. Listen to me.
231. 39:54Fill out your social profiles. Now, listen. If you go look at my AI salon profile and I don't have my social profiles filled out,
232. 40:06I will admit that I'm a bad man and I will do it. I promise. What is a social profile? Okay, I think it's Mary Carol. Hey MC. So, if you go
233. 40:19to your profile here, I can do it. I can show you right now. And I might need Andy's help on this or producer
234. 40:24Brandon's help on this. But if I go to my profile and I go to personal settings in here, account connectors, notification, language,
235. 40:39feedback, and info. Download the app. Show them my profile. Okay, but wait, where's where do you add them? Edit profile.
236. 41:00I'm incognito today. Only feeling partially like myself. Claire Jacobs not sharing. Oh, wait. I'm not sharing now.
237. 41:07Oh, right. That was on producer Brandon. Wait, I am sharing. Okay. Where are the social profiles? So, let's
238. 41:17go look for Linda Sherman. We'll go to Um, get started. Linda Sherman. Okay. Linda Sherman. So, here's Linda.
239. 41:36About me, personal links, LinkedIn threads, Facebook, Instagram. See all this? That. Okay. So, where are
240. 41:46the personal links? Where do you set those up? Anybody know? Andy, why the hell am I taking people out of a community and dragging them
241. 41:56back into the abyss? [laughter] That's a good point. [laughter] Why am I going back to fighting an algo?
242. 42:07I'm not. Okay, whatever. Um, I think it's a good idea, but Andy's got a really good point. We're already in the
243. 42:13community. But here's the important thing. We're in a community for a reason. finding each other in all those in all
244. 42:21the noise. It's getting noisier and noisier and noisier. I don't know. Have you noticed in X that it is impossible
245. 42:29to actually understand who's actually posting anymore? Like it's getting bad. It's going to get worse and worse and
246. 42:36worse. So, actually Andy's got a really good So, go ahead and enjoy the abyss. You hit personal links when you were
247. 42:42showing your profile edit. Okay, cool. All right. So, that's how you find those. So, if you want to share that
248. 42:47stuff, put that in there and then other people can find your stuff. To Andy's point, keep it in the AI salon. [laughter]
249. 42:56Okay. Um, let me talk about something. So, um, there is a lot going on in the AI salon. There is now a lot going on in the AI
250. 43:11salon mastermind. If you don't know about the mastermind, um, pop up that the graphic, Brandon. Um, yeah. Who's banning the bots on
251. 43:24LinkedIn? No one. Um, yeah, we have very few bots in the AI salon. That's because of Andy and Brandon banning them. Um,
252. 43:33okay. The the the mastermind. If you've not checked out the mastermind, go check it out. So, within the mastermind, we
253. 43:39have the AI salon practice lab, which is um a 10-week cycle that we just completed our third 10-week cycle of people
254. 43:50working on a daily practice around how they use AI. We've also got a tremendous amount of programming under the great
255. 43:57repurpose. Some of that is about um human development, like unhooking your identity from your job. Some of
256. 44:05that is about AI development like the workshop I'm about to do on um on creating your digital self. Um
257. 44:17these the human beings that are in the mastermind and the work that we do in these workshops is really quite
258. 44:24remarkable. Like people's lives are changing. People are changing their lives. They're discovering who they are,
259. 44:30what their value is. Um I regularly am [laughter] moved to tears in these meetings um because of the work that people are doing.
260. 44:42We're entering a time where the chaos around AI and what it's doing to jobs and what it's doing to the jobs market
261. 44:49is going to get noisier and noisier and more overwhelming and more overwhelming. What's happening inside the mastermind
262. 44:56is people deeply connecting with one one another, supporting one another, and understanding that who they are and what
263. 45:03their value is is more valuable than all that noise. And if you can take that and figure out how to amplify that with AI,
264. 45:13you're going to do remarkable things. And people in the in the mastermind are doing remarkable things. So, if you
265. 45:18haven't checked it out, check it out. If you have signed up for the mastermind and aren't attending these workshops or
266. 45:25aren't watching the videos on demand, please go do that. Really important stuff. Okay, beautiful. Does the
267. 45:33mastermind require a fee? Yes, you can apply for a scholarship. Talk to Andy about that. Um, it is a fee, but for
268. 45:41what you get, it is incredibly cheap. Like, it is it is stupidly cheap. Like we've regularly got people saying
269. 45:51it is changing my life. [laughter] Anyway, Mary Carol come to the mastermind. Mary Carol goes to all of them.
270. 46:06She is committed to figuring this stuff out, to figuring herself out. It's inspiring. Sunonny Yun started
271. 46:19when she got laid off from a 23-y year, 22 year, 23 year career in advertising and was completely demoralized when she started.
272. 46:30Figured out through Andy's workshop about identity, holy crap, I'm more than my advertising career. I'm this powerful woman. I'm going to go
273. 46:42do some [ __ ] in the world. And she's going and doing [ __ ] in the world. It's really inspiring. Charles Adams, I've been in my writing
274. 46:52rabbit hole for months. I should probably absorb some mastermind stuff. If if for no other reason, Charles, just
275. 46:58to allow your mind to like unwind from writing mode into a different mode, even for an hour, will probably unlock
276. 47:07insights for your writing projects. But yes, come come. Yay, Sunny. I know. Claire Jacobs is another one that's in there. Kicking ass.
277. 47:18It's like like what what what we're discovering what we're discovering in there is like that thing about ourselves
278. 47:35that's like, "Yeah, I'm that." Like for me, like I'm I'm really getting in touch with how much joy it gives me to be creative and
279. 47:48artistic and inspire other people to be creative and artistic and innovative and like being in the presence of people
280. 47:56that are creating like just gives me joy and light and in a weird way it gives me permission to be in the world and like own who I am
281. 48:08and use AI to like blow the doors off that thing. It's so cool. So, anyway, hello from Chicago. Can't believe I'm
282. 48:17watching this in a totally different time zone. Hey, Chef Kelly. What's happening? It was funny. Andy and I and
283. 48:23Andy and Daisy and I were in a meeting today and there was a woman in there and I'm looking at her. I'm looking at I'm like,
284. 48:30why does this woman look so familiar? She looks like she could be your sister, Chef Kelly. She looked very much like
285. 48:35you. So, it was it was cool. So, good to see you. Um, okay. Beautiful. I'm truly looking forward to joining. Awesome,
286. 48:42Patty. I'm excited to have you there. Okay, let's get back to spelunking with chat GPT. [laughter] What we've learned so far.
287. 48:53Let's see. Okay, done. I emailed a brief to kyle@storyvine.com with this with the subject story strategic AI video brief
288. 49:01July 13th, 2026. And I've updated the recurring Monday meeting. So, let's go look and see if it actually did that. Uh, I don't see it here.
289. 49:14Ah, here it is. I'm going to make it important so I get to see it. Let's see if it's any good. Okay, it's boring as [ __ ]
290. 49:27and it's too long. Okay, but what do we got? Google introduced Gemini Omni and pushed multimodal video creation. Yes,
291. 49:35it did. That's old. You should remind it ADHD. Great. Okay. The Verge. Okay. So, it can it can do this. All right. So, here's what we're
292. 49:46going to do. We're going to go back to chat GPT. Let's go look at the brief. Let's go look at the prompt. Research this. So,
293. 49:54what did it say? Prioritize. Okay. Give extra weight. Here we go. Email a completed completed brief to kyletstory.com.
294. 50:03Okay. So, what we're going to do is we're going to say to chat GBT, um, looks like a good good start.
295. 50:21I need the following changes. One, I am visual. Display some graphics from the articles you find
296. 50:37in the email. So, my brain has candy. Two, I loves me some ADHD.
297. 50:54Um, your email had my words. I need it tight and compelling.
298. 51:13Maybe with a little humor or squirrels. [laughter] Um, also
299. 51:28also I'm a creative director and your all nonbolded
300. 51:4312point text makes me want to break things rather than read them. Design this
301. 52:01as if you gave a damn. [laughter] Okay, let's see. Let's see how chat GPT responds to smarmy creative direction
302. 52:14generating a more detailed image. Hang tight. Stop. God damn it. I didn't ask you to create an image.
303. 52:29I asked you to update the design of the daily email and schedule.
304. 52:45Do you understand? Apparently not.
305. 53:00All right, that's off doing that. Let's go to the chat GPT app now. So now we're in the chat GPT app. Okay, so what have we learned? ChatGBT
306. 53:14and chatgpt work are two different modes in chatgbt.com. The chatgbt web experience in your browser.
307. 53:29It looks like you can mostly do everything in chatbt that you can do in work. Except I would imagine that the
308. 53:37difference is work can write files to your hard drive. Chat GPT probably can't. Except it's got plugins, so it
309. 53:47can certainly write things to Google Drive because it's got access to my email. So I don't I don't know if there's an
310. 53:55actual great distinction between the two. Okay. So that's first thing. Now we're in codeex. Well, no, sorry. Now we're in the Chat GPT
311. 54:10desktop app. It used to be there were two desktop apps, ChatGpt and Codeex. Codeex was for the nerds, the geeks, and Chat GPT was
312. 54:21just like the chat experience except it was always different than the chat experience. Make sense? Now they've been rolled together.
313. 54:31And I think what they did was they took the codeex app and renamed it chat GBT and they took the chat GPT app and they
314. 54:37retired it or the other way around. It doesn't matter. There's one app and it's sort of like what you can do on the web except different.
315. 54:50So we now have this thing get started. Okay. So let's make this bigger. Okay. So the one thing I know and I learned
316. 55:02this from HT snow day and office hours like last last week is that in the upper leftand corner it says chat GPT work
317. 55:10right like that's a label. It's also a pull down menu. So you can switch from chat GPT work to codeex. Now let's look
318. 55:18at the little menu over here. You got new task scheduled plugins sites. That's new and that's also available in chat
319. 55:27GPT. We'll go back to that. And then you've got chat. Chat has now been relegated to a left bar menu item. It's not a major pain.
320. 55:40It's not a major function. There's chat GPT work and there's codeex. This is where I think we're going. I think two things are happening
321. 55:50simultaneously. I think chat is going to be so when you use it, it it gives you a a popup window, a modal. A modal is a
322. 55:59thing that that floats on top of the actual experience, right? How do I put it away like that? Yes, you got chat GPT work and chatgex. Now,
323. 56:12when I click codeex, the only thing it seems to do is it adds this pull request. um thingy thingy,
324. 56:26but we're now we're in codeex and we have pull requests. So, a pull request is basically where you've done some work
325. 56:33on in a section of GitHub and then you want to pull your changes into the main branch of the code and push it to
326. 56:42production. It's it's a geek thing. So, here's what I think is going to happen. I think chat is going to remain a
327. 56:56like a support tool. If you want to do just a quick chat GPT chat, you're going to pop up chat. It's probably not going to have history.
328. 57:05I don't know. I I don't freaking know. I don't know. And you got work. Let's see. New task. Create a or build create a file or build
329. 57:17a site. Research and plan next steps. Get a briefing on recent work. Automate routine and recurring work.
330. 57:36If I'm a casual chat GPT user, this is a [ __ ] disaster because Kelly, to your point, so plugins are connectors kind of, I
331. 57:51think. So, so if we go to plugins, I'm pretty sure they are connectors. They just called them plugins. So, you've got computer use
332. 58:02So try it now. Is it is it on? MCP server skill one computer use is turned on. MCP servers one computer use.
333. 58:16So what the [ __ ] is the difference between computer use MCP server and computer use? Control lo local maps
334. 58:24through computer use for wait control local Mac apps through computer use for tasks that require reading or operating
335. 58:32app. Why doesn't this one have an ex? This is a [ __ ] disaster. All right. The vocabulary really throws me, but I
336. 58:49do look it up. We had plugins back in the beginning of chat GPT. Yes, they all work pretty interchangeably.
337. 58:59It's the It's the pretty in in that sentence, Charles, that is throwing me. Pretty interchangeably and interchangeably are two different
338. 59:08worlds. Like the fact that in wait, let me go back to chat GBT for a second for the to the website. The fact that in chat GPT if I do new
339. 59:23chat the fact that I've got different plugins enabled for chat than work or I've got more
340. 59:39says to me that they kind of are the same but they're not. And in this world, And in this world, chat is one of the
341. 59:51main PES and work is one of the main PES and CEX doesn't exist at all. And then I go to the app and and we don't have PES
342. 59:59anymore. We've got drop-own menus over here and it's chat GPT work and codeex. And chat is like a little featureette.
343. 1:00:11And in order to use chat GPT work, you need to understand that you're going to do things in projects,
344. 1:00:18you need to understand that you need to set up local directory files and context about who you are and how you like to work.
345. 1:00:29Good lord, this is a [ __ ] disaster. And I mean that in the nicest possible way.
346. 1:00:43Then if you flip to codeex, got a new project. Here's local work locally. New work tree usage remaining
347. 1:00:5591%. Yeah, cuz I haven't been using chat GPT 5.6 all that much. You know why? Cuz I'm confused. >> [laughter]
348. 1:01:11>> Oh, good lord. Schedule plugins more. Oh, now I'm back at Okay, wait. So, wait, do we have sites before you use sites?
349. 1:01:25Okay, by the way, sites is a brand new feature. Yet another feature on top of all the other features is you know how lovable
350. 1:01:36can make websites that are cool things. ChatgPT now does that. So ChatgPT now is directly competing with Lovable.
351. 1:01:45You're responsible for your site and anything visitors submit. If your site collects personal data, learn more about
352. 1:01:53your responsibilities. Open AAI removes sites that violate our policies. I guess people are making all kinds of
353. 1:02:00sites they don't want up there. Kelly Camp, as a technical writer and advertising major, the lack of naming
354. 1:02:05the tools properly drives me mad. The good news, Kelly, is it's been driving you mad for three and a half years. [laughter]
355. 1:02:17Remember when they had their four 4.0 and then 40? I just I would lose my [ __ ] on that on a weekly basis.
356. 1:02:26I'm excited. I'm part of the mastermind now. Welcome, Mr. It. Excited to have you there. Get you digging in deep and learning.
357. 1:02:36Peel back the onion. It's going to be good. Okay. What else do we have? What? Palm to forehead. [laughter]
358. 1:02:46The positive work in progress. Oh, okay. So, here's here's my this isn't a hot take. This is a This is more like chat TMZ. Okay,
359. 1:03:06I'm not even going to Well, no, I will. I'll look at my phone. I'm going to go to Chat GPT. I'm going to force quit it. I'm going to
360. 1:03:17relaunch it. I'm actually going to go to the App Store and see if there's a new update. Oh, you know what I can do? No, Kyle,
361. 1:03:27why don't you tell us? I can do this. Which one? This one. This one.
362. 1:03:43Oh, no. I can't. Never mind. Screen mirroring. Stop mirroring. I thought I could mirror my screen, but I can't because I'm
363. 1:03:52streaming. Okay, never mind. Hang on. Um, chat GPT open. Okay.
364. 1:04:11Okay. So, down the side we've got library, projects, schedules, plugins, and then remote. So remote I should be hooked up to Yeah. So I'm
365. 1:04:25hooked up to my my remote MacBook Pro. So this this tab that now says remote used to be called codecs. So it's
366. 1:04:42remote [clears throat] is basically I can send prompts to a remote machine that's running 24/7 that's running
367. 1:04:57codeex or chatgpt work. This is just like open claw. So open claw is this thing that's running and then in telegram you can send it
368. 1:05:12stuff. So it's kind of like that but kind of different. Okay. Then images and finances. You can also put in you can you can put in all of
369. 1:05:28your finances for chat GPT to publish to the world when they get hacked when they put out their good new model. Okay.
370. 1:05:36What? Remote equals codeex arg. Yeah, exactly. Okay. Three and a half names of angst over these naming disasters. Okay. Calm down.
371. 1:05:46Everybody calm down. Kelly camp has started a riot. People are insensed about the naming inconsistencies. I'm insensed about the
372. 1:05:58capability inconsistencies. We've got two of the three three things over here. Three of the things over
373. 1:06:04here. But the third thing is like half a thing. And we've got a thing that is just called something different here. That's
374. 1:06:10a thing that used to be a thing that doesn't exist anymore, but it sort of exists in the app, but you have to pull
375. 1:06:15it down from a menu that doesn't look like a menu. I understand
376. 1:06:29what a [ __ ] disaster this is. Okay. Here's the good news. I'm fairly confident where this is going is chat GPT work
377. 1:06:46will at some point just become chat GPT again. the chat GBT experience where you just ask chat GBT for something may
378. 1:06:54still live as a pop out window, but I think it's just going to there's just going to be a box again and maybe you can turn on
379. 1:07:05programming features or work features codecs chatgbt work but I think it's just going to go back to being chat GPT.
380. 1:07:17I think it does make sense to have a limited set of functions on the phone, but it sure would be nice if what the
381. 1:07:26phone primarily became was, hey, could you fix that thing? And then it's like giving you updates and letting you play with things and
382. 1:07:36review things here, right? because I think and then ultimately it goes to the smart glasses that we're wearing and just chat GPT is
383. 1:07:48off on your machines doing its agentic [ __ ] Um, but right now, like if you're if you're
384. 1:08:03if there's a bunch of people in here that are educators of AI to human beings that are like, "Hey, hey, Kelly." Yeah,
385. 1:08:14listen. I know that I told you you were you were the devil's spawn when I learned that you used AI about a year
386. 1:08:20ago, and I'm sorry about that. I probably it was probably a little aggressive of me. Um I just got laid off and I was so apparently
387. 1:08:30to be considered to work at my company I have to learn AI. It's just [laughter] you know it's the it's the devil's work.
388. 1:08:37I I'm going to stop saying that it's not the devil's work. It's just fine. It's fine. Any I was hoping maybe I could,
389. 1:08:45you know, I don't know if you've got like a spare 10 hours where you could just like show me what to do. It' be so good. It' be great.
390. 1:08:56I feel for you because where do you have someone start? We'll just go to chat GPT. Okay. And then what do I do? Okay. So, when you go
391. 1:09:06to chat GPT, there's two modes. There's there's chatgpt mode and then there's this other mode called work and it's
392. 1:09:13they're kind of the same but they're super separate and then but they each have plugins and plugins are like
393. 1:09:19connectors where you connect other things but it's like a plugin because you turn on a thing and then some of the
394. 1:09:23connectors are plugins and some of the connectors are connectors. So you have sometimes with something like computer
395. 1:09:29use like you'll turn one on and that's but it's an MCP version of computer use but then you also have the plug-in of
396. 1:09:34computer use that has separate configurations and you can configure them both however you want which is the
397. 1:09:40real powerful thing about chat GBT is it can do everything now and because it can do everything I JUST WANT TO [ __ ] scream
398. 1:09:51where do you Start. [laughter]
399. 1:10:06Here's where you start. I am so screwed. On July 21st, I am teaching a course called Back to Basics. There are no basics.
400. 1:10:25[laughter] You start where you want to be. Who cares about the tool? Literally, your message. It's true.
401. 1:10:33It's true. Yes, Andy. What you said. And that's maddening. It's maddening.
402. 1:10:47So anyway, July 21st, what we're going to start doing, so so I just like being the man on the YouTube and losing my [ __ ] sometimes. Um,
403. 1:11:00every third Tuesday, so the first Tuesday of the month is AI Salon Presents. We're adding back to basics on
404. 1:11:07the third Tuesday of the month. So there's 20 people in here right now. Um, think about people that you know who
405. 1:11:15want to get started with AI or if you're just getting started with AI, come on the 21st. I'm just going to talk about
406. 1:11:22it's going to be a two-hour session. We're going to talk high level about what AI is, what large language models
407. 1:11:30are. I might even do a thing on how image generation tools work because it's kind of interesting. I'm going to talk a little bit about
408. 1:11:36what chat GPT is. I'll try to calm my anger and I'll say, "You just come here. You just do this. Don't worry about all
409. 1:11:44the other buttons. Just ignore all that [ __ ] and just start talking to it. Write a poem. Go make a kids book. Make a picture.
410. 1:11:53I'll talk a little bit about image generation. I'll talk a little bit about songs. I'll talk a little bit about
411. 1:11:58video. And in two hours, someone will be able to say, "I'm an AI expert. [laughter] Making fun making funny pictures is the
412. 1:12:12elite. Exactly. [laughter] Wow. Looking back at spring 2023 and where we've arrived now. Yeah. Here's the here's here's my real like even
413. 1:12:24though I just I just went all drama. um where these apps are right now is actually a necessary step. Like the good
414. 1:12:39news here is that they're starting to collapse these. They realize that they had the chat GPT app and the codeex app
415. 1:12:46and they needed to roll them together. They also recognize that having a menu item like pull requests for consumers is stupid because
416. 1:12:57consumers don't know what pull requests are. But they also realized if they remove that for coders,
417. 1:13:05coders might be like, "Where the [ __ ] is my pull request tab?" Right? And then to Kelly's point, she's like, I
418. 1:13:16just have this sneaking suspicion that they're driving us toward aentic, which they are for a couple of reasons. One of them's
419. 1:13:25really sneaky. You as a human being can burn x number of tokens, right? Maybe it's a h 100,000 in a month
420. 1:13:37or maybe it's 200 or 500,000 in a month. you with an agent that runs 24/7 and wakes itself up at at multiple times of
421. 1:13:48the night telling ChatGpt what to do. You could burn 10 million tokens a month or half a billion tokens a month and at
422. 1:14:01some point you're going to want your things to keep running because one of them's making you money and nine of them are posting on your
423. 1:14:09socials and you're getting followers which also just happen to be bots following you but but it makes your ego
424. 1:14:16feel good. And so at some point you're gonna up your subscription from $20 a month to $100 a month because you keep
425. 1:14:23blowing through your tokens. So So the reason they're pushing us to agentic work is so they can sell you
426. 1:14:31more tokens. You need more tokens. You you become token hungry. It's wild. This is this is I go back to
427. 1:14:42three years ago Sam Alman saying I think instead of talking about universal basic income we should talk about universal basic compute.
428. 1:14:51And when I first heard it I'm like that's weird. That makes no sense to me. Give me the money and I'll pay you your
429. 1:14:5820 bucks a month to do my chat GPT. No. with Agentic stuff, we're going to need billions of tokens. Billions that cost.
430. 1:15:14So if they give you a bunch of compute, then you can use that compute for your agents to go start your business to give
431. 1:15:21you a check so you can make more money without having to work for it. Our new drug dealers totally. If you don't think that us getting
432. 1:15:35access to Fable is is the drug dealer model or us getting access to GPT 5.6 or GPT6 for 20 bucks a month,
433. 1:15:48Brent Peterson talked about the GLM 5.2 model, this big model that's supposed to be as good as Opus. He said if they run it on virtual
434. 1:15:57machines big enough to actually run it for their company which I don't know how many employees they have maybe 20 15 20
435. 1:16:07to run a local open-source model that is the quality of opus 4.6 not even fable or mythos not even 5.6 six three
436. 1:16:17trillion parameters. This is like a a trillion parameter or like 900 million parameter model. Opus 4.6 level
437. 1:16:28running in a virtual machine in the cloud. He figured for his company it would be $100 an hour to run a model
438. 1:16:41that's almost as good as Opus 4.6. They're now at Opus 4.8. 8 and they now have Fable 5 $100 an hour to run it just on a GPU rental service.
439. 1:16:56So we are we are absolutely be being given free drugs [laughter] to get us hooked and then when the little magic agents
440. 1:17:06start making us money, we're going to be like, "Oh, we love the agents." [laughter] Claude is killing me. It's forgetting everything.
441. 1:17:18That's a That's a setup problem, I think. Ivan, you got to get your memory dialed in there. But yes, I would imagine that's
442. 1:17:25frustrating as hell. One of the hypnosiserts. Let's see. FYI, my students from cohort one are presenting their finals
443. 1:17:35tomorrow. Can't wait. That's awesome. The new thing The new thing is teaching how to make infographics. Seriously, it's the new toy.
444. 1:17:44>> [laughter] >> Oh god. Well, Ivan, so I was talking about this earlier. I don't know if you were here, but I'm going to do the
445. 1:17:57series that I'm doing on creating your digital self, the the AI memory stuff. Um, I'm going to tee that that that
446. 1:18:07series up and then I want to bring in experts who just show us how they're doing it. Um because I don't think
447. 1:18:14there's a single way to do it. I think there's kind of an infinite ways to do it and that's a problem because it's very very complicated.
448. 1:18:25[sighs] How are we doing time-wise? We're getting there. Um who has any questions? Did Was there anything? Oh yeah. Let's
449. 1:18:33see if Chat GPT Am I still sharing? Hang on. Yes. Let's see if chat GPT fixed our our video brief. Yes, I
450. 1:18:47understand. I updated and scheduled the Monday email, not an image. The new format is 5750 words total. the big three at the top. Six to
451. 1:18:59six to eight compact features, then rapid quick hits, strong headings, two to four article graphics or thumbnails
452. 1:19:07when the email system supports them, brief humor, no beige wall of 12 12point text. It remains scheduled and we'll be
453. 1:19:16emailed. Okay. Um, send me a test email now. I want to see your work. I am not confident.
454. 1:19:31Period. You [ __ ] Okay, let me go see. Am I in 5.6? No, I'm not. Oh, wait. I am. Okay. Oh, that's weird. So, if you go
455. 1:19:43instant, you drop back to 5.5. Medium and high are on 5.6. Okay. So, I'm on high 5.6. Great. I'm sending a real
456. 1:19:53test, not another promise. Okay. Um, deep cleansing breath. Um, what is the best model to use less API tokens? Well, the the
457. 1:20:08best model to use less API tokens, Mr. IT, is a local model. So, have enough hardware to run a decent local model. I've got a 64 gigabyte
458. 1:20:20machine, which I wish I'd gone for the 128. The 128s now are like an an M a MacBook M5 Max with 128 gigs of unified memory is
459. 1:20:36now seven grand. It was five grand two months ago. And and a and a 128 will run a decent model, but it will not run a big model.
460. 1:20:46It will not run like a GLM 5.2, but the 64 gigs runs decently powerful. The Quen 3.6 model is essentially just the cost
461. 1:20:57of electricity. Olama. Yes, exactly. I'm setting a mini PC to house to all my agents. Told them you have to have Olama
462. 1:21:09like Candy Crush. We will have to watch ads to get extra playtime for the next model. Probably. Yes. Patty McKenzie. I
463. 1:21:16got a 64 GB too. Yeah, it's good. So, I just saw a post today that it looks like the M7 the M7 processors that are coming out in
464. 1:21:29like a year and a half are going to have up to 1.5 terabytes of unified memory. So, you know, two years from now, three
465. 1:21:43years from now, you'll be able to run something like the GLM 5.2 model on a local machine. Can you say that again
466. 1:21:51about the Mac? I'm about to purchase a new one. Okay. I did a graphic on this. I don't know where it is. Let me see if it's in chat. GBT.
467. 1:22:02Um, sent test. Okay, so it sent us the test. Let me do a search here for a second. Um, uh, what the [ __ ] was it? Um, local llm
468. 1:22:17infographic. Um, 128 gigabyte. Oh [ __ ] I said if infographic infographic.
469. 1:22:33No results. Agentic harness infographic. Is this it?
470. 1:22:52No, wait. Oh, no. That was No, that wasn't it. Uh uh let's see uh
471. 1:23:09Mac LLM infographic. I think I posted this in the AI salon. Hang on a second. Um
472. 1:23:25if I go get started and I go um local LLM info graphic. No results found.
473. 1:23:49Found it. Okay. So, so if you go to the AI salon, am I still sharing? Let me
474. 1:24:05make sure I'm still sharing. I am. Go to the AI salon. Um, go up to the top where it says get started. You have to be at
475. 1:24:12one of the top level things to search everything. It's it's in the community feed. So you can either go to the community feed and
476. 1:24:23search. Actually, let's let's do this. So if I go to the community feed and I search for local LLM, it's the second it's the second graphic.
477. 1:24:35Okay. So let me explain this graphic. What this talks about on the left hand side of it is Apple versus PC workstation.
478. 1:24:48The the big difference here is that the term unified memory is really important. So when you buy a Mac, the M series Max, and it's got 64
479. 1:25:04gigs of unified memory, both the GPUs and the CPUs can use the same memory. When you buy a PC, you've got RAM for
480. 1:25:16your CPU and then you've got graphics cards with VRAM for the GPU. You have to be able to run whatever
481. 1:25:27model on the amount of GPU RAM you have access to. So, if you've got like a really good CPU but small video memory
482. 1:25:38on a PC, that's that's not great. So, so the reason people like Max is that it's just s it's like one layer more simple. Okay,
483. 1:25:50screenshot plate, please. Got it. Thanks so much. Okay, so here are the here are the tiers. So So these were the
484. 1:26:01tiers I was considering. People that are buying like Mac Studios are buying Mac Studios with 512 gigabytes of unified
485. 1:26:10memory and then they're they're stacking four of those together or they're buying um 128 gig Mac minis and you know
486. 1:26:19stitching those together into a 512. But I have the 64 gigabyte one. I can run the Quen 3.6 model, the 31 billion
487. 1:26:29parameter model. If I had the 128, I could run like bigger 70 billion class models or higher quantization of of
488. 1:26:38these smaller models. Um, when you start getting into like things like GLM 5.2, you need like 512 gigabytes or, you know, a,000
489. 1:26:54gigabytes, 1024, something like that. Okay. But this this graphic is in community feed. Just look up local LL LLMs.
490. 1:27:08But this is the answer to Mr. It's questions. How do I not spend so much on tokens? Generate your own tokens. Right? Most of
491. 1:27:18the work we do does not require you to have fable level reasoning. Right? If what you're doing is little automations and a little bit of
492. 1:27:26orchestration, it's it's it's easy. Okay. Now, um let me go back to my inbox. It said it sent me a new email, but I
493. 1:27:42don't see it. Story Vine AI video brief. I don't see it. Let's see.
494. 1:27:59Is it time? It's time. Did it append the first message? No, it didn't. All right. Let me go back to chat GPT. Let me go back to
495. 1:28:15video brief. test signal. One limitation. The Gmail connection sends plain text emails and does not currently support embedded inline
496. 1:28:27articles. I compensated with strict hierarchy. There is no email in my inbox and
497. 1:28:43codeex uses Gmail and can send images. Get smarter. All right, that's going to go off and do that. Um, it's nine o'clock. I want to
498. 1:28:57get you guys out of here on time. Let's see. I have a supplier in Alibaba that can build me anything for a good
499. 1:29:04good price, but no unified memory. Um, on the PC side, you can look at the AMD. AMD has a new spec out that's a computer
500. 1:29:14that's 125 gigab bytes. You can look at the the Nvidia Spark machines. Um, and there's a there's a third one that
501. 1:29:25just came out. There's a bunch of people are going to be offering these like right now you're going to pay a premium
502. 1:29:33like like Max went up 25% in price in the past two months cuz there cuz the chips are not available like everything's being eaten
503. 1:29:45up by the by these big labs. Um so it's probably if you need to get a machine right now get a machine right
504. 1:29:54now. um because I don't think they're going to get cheaper. But I think what's going to happen is Apple in particular,
505. 1:30:02you know, Tim Cook is stepping down and the CEO who's replacing him is from the hardware division. So I think Apple's
506. 1:30:09going to lean in really hard into, you know, highmemory AI inference machines. Thanks, Kyle. Thanks, mods and AI peeps. Okay, cool.
507. 1:30:19Beautiful. Thank you, Kyle. Have a great night, everyone. Thank you all. Have a good night. I will see you Wednesday,
508. 1:30:28tomorrow. I don't know what's happening, but go check the events calendar in the salon. All right. Peace out, everyone. Have a good night.