8/12/2025 - ChatGPT 5's Secrets: AI's Evolution and The FUEL Framework: Mastering AI Prompting

Transcript

1. 0:00live. We'll do it live. We'll do it live. [Music] [Applause] [Music]
2. 0:16[Music]
3. 0:36freedom came my way that night [Music] just like a jet plate in and out of sight. I was hauling ass at a million miles an
4. 0:51hour, wondering how hard I'd hit when they came into the station. [Music]
5. 1:06They said I was bad beyond repair, but I got no qualms with my situation. say here I am.
6. 1:21[Music] So say Sheree Sheree Sheree won't you dare do say Sheree Sheree Sheree won't you dare
7. 1:37do say Sheree Sheree Shereehu yeah leave a message and your number please
8. 1:52take your time to want to satis satisfy me. Take all these old fantasies and send them care of me. [Music]
9. 2:08Oh yes, yes, yes, yes, yes. What day is today? Tuesday. It's Tuesday, people. It's Tuesday. Tuesday. What do we do on Tuesdays?
10. 2:22What do we do on Tuesday? Well, we often have AI salons. That's not tonight. What else do we do on Tuesdays? We reveal secrets.
11. 2:34I'm going to reveal secrets tonight that no one else knows about Chat GPT. I'm going to be like a magician like [Music]
12. 2:51in a westerly direction. This car is my train. I've been driving. I've been wondering
13. 3:05what it is I'm running from again. Feel like an old man holding on to 29 and up ahead on that horizon
14. 3:22[Music] is California life [Applause] [Music] of a head of trucks carrying a wide load.
15. 3:38Preab house cut in half. Cute little front door and two windows. My love ain't sure whether to cry or should laugh.
16. 3:51You see, I broke a home of myself once. As I stumbled to that door, I read a note by the dawn light said, "Don't you come around here
17. 4:09anymore. I had enough of this freedom of never was good with decision.
18. 4:24At least that's what I've been told. [Applause] >> Hi, Kyle. [Music]
19. 4:40[Applause] [Music]
20. 4:57I do hate gender up champ. But we do have an actual question on Tik Tok for you. >> Yes. Yes. >> Which I which I can kind of read uh
21. 5:04through your shirt. So if you'd be so kind to switch your cams. >> Flip flip my little cams. I can do that. >> Beautiful. >> Huh? That's
22. 5:16>> I would say while we're at it, I just want to make sure you have Miro up. Oh, I do have Miro up. I don't have the
23. 5:26um the Post-it notes in my view, but I now do. And I'm going to put the purple below the blue so because I can see them
24. 5:34both vertically off to the right here. Perfect. There you go. Champy sings the blues. He does. Okay. Um, hey Pate M,
25. 5:44what's happening? Side Hustle Mimi Joy Party in the house. got some some good folks. Professor Oscar G. is GPT5 or is it misunderstood?
26. 5:57Um, we're going to talk about that tonight. That's the secret of chat GPT5. Professor, Professor Oscar, you're jumping the gun.
27. 6:09Can't tell you the secret of Chat GPT5. Right off the gate. Right. Right off the gate. Out of the gate. Off the bat. Off
28. 6:18the bat. Out of the gate. Yeah, exactly. How dare you ask a question that I was going to give the answer to? [Music]
29. 6:32But I think you hit it. I think you hit it. Here's the secret of chat GPT5. [Music] I think everyone including
30. 6:48the people at OpenAI don't have a clue. I think that's the secret of ChatgBT5. We don't know yet. It's too early. We don't know how to use
31. 7:01it. They don't know how to present it. An hour before we came on here, they changed the interface again.
32. 7:10We are back to a model picker with eight different models in it as of like an hour ago. So the secret to chat GPT5 is we don't
33. 7:21know now. I think I think it's better. It's way better, but it's not a thing. Okay, let's see. What should we do
34. 7:33tonight? Image benchmarking video avatars research tools. Music night. Oh, no, no, no, no. We've got an agenda for
35. 7:41tonight. I was going to uh somebody asked for a poll of whether it should whether GPT5 was bad or just
36. 7:47misunderstood. And it it pulled a wrong poll. >> Oh, it pulled an old poll. Okay, good. >> But there is a poll coming. So,
37. 7:54>> okay, there's a poll on Tik Tok. >> On Tik Tok, there's a poll coming. So, ignore that. Um, if you're over there >> on YouTube, uh, and
38. 8:06>> what's that? >> We're doing it on TikTok. We're doing the poll on Tik Tok. >> Yeah, doing the poll on Tik Tok. If
39. 8:11you're on YouTube, um feel free to just comment if you have comments. Um and if anyone has any questions, feel
40. 8:21free to uh to pop them in either place, Tik Tok or YouTube. Um and we we'll put a poll up in a second here. Um better than a nose picker.
41. 8:32You can pick your friends. You can pick your nose, but you can't pick your friend's nose. [Music]
42. 8:51[Music] Good morning. [Music]
43. 9:21She came on him like slow moving cold fronts. Well, his beard was warmer than a look in her eyes. [Music]
44. 9:37Okay. Up in the uh up in the upper leftand corner. Oh, wait. Yeah, up in the upper leftand corner over there,
45. 9:45there's a there's a a blue box on an angle and pull that up. GPT5 is bad, misunderstood math. Give me my 40 girlfriend back. Um, so
46. 9:59yeah, go do a poll over there. All right. [Music] Alultman once said, "We will see a soloreneur hit a billion dollar
47. 10:10valuation with AI soon." He just said that on a a podcast the other day. He talked about that like a year and a half
48. 10:16ago. That that's a standing bet um amongst uh founders in Silicon Valley. It has been for a while. Um that's funny, Vicki. [Music]
49. 10:34[Applause] [Music] Thank you. Sad hustle. Mimi.
50. 10:50[Music] Standing between you and a hard place is insane.
51. 11:09Standing too near you and a fire makes it clear. [Music] You're trouble to me.
52. 11:24Real trouble. Can't you say [Music] leaving in close? Smell of your perfume scares me most.
53. 11:39Leaving away feels stronger every day. [Music] You're trouble to me.
54. 11:54Real trouble. Can't you see? All right, let's get this show started. Let's get this show on the road, people.
55. 12:06Um, that last song was actually a song I wrote. Most songs I sing I did not write. That last one I did. Uh, the poll
56. 12:14uh, five people took it. Chat GBT is misunderstood. Was 100% of the answer. Um, that's the big secret to GPT5 right
57. 12:22now. We just don't know. We just don't know. We don't know how to prompt it. We don't know how to use a model that's
58. 12:29going to route our prompts to different models. Um, and now I want to let me show you something they did here which I find astounding. Um,
59. 12:44[Music] [Music] okay.
60. 12:58There we go. All right. You can all see that, right? You can see that there. Fant Tik Tok pen. Two block tom. I
61. 13:06actually got some good historical fiction writing with thinking mode. I was impressed. Yeah. So, a couple of
62. 13:13things you should notice with five. One, it shouldn't hallucinate as much. They've got a what do they call it? A
63. 13:23I don't know. It looks at the answer before it gives it to you. Should hallucinate less. Um, look what they
64. 13:29did. I can't believe they [ __ ] did this. We are now back to one, two, three, four, five, six, seven, eight, nine different model choices.
65. 13:47Nine. Nine. Nine of them. The whole point of chat GPT5 was we're going to hide all these models from you
66. 14:00so you don't have to think about it. Then they launched chat GPD5 and everyone who liked chat GPT4 because
67. 14:08they were using it as their girlfriend freaked out on Reddit and so they added back in GPT4 and they're like we don't
68. 14:16know what it's doing. How are we supposed to know what it's doing? So now all this crap is back. It's unbelievable to me.
69. 14:28You don't have all those legacy models really. I don't want to revert to four because it will go away. And then the whiny
70. 14:37babies came out and Sam caved. I agree. Like I I feel like I feel like what I'm facing right here is So Vicki, do did
71. 14:46you So in order to see the legacy models, you have to go into your settings. You might not have been here
72. 14:51because I think you were traveling or or working on work stuff. If you go into settings and scroll down to the bottom
73. 14:57of the general page, you have to say oh and show additional models. Now they even changed that. That said show legacy
74. 15:04models as of yesterday. Now it says show additional models. So you have to turn that on. You have you have auto fast thinking and
75. 15:13legacy 4 model. Yeah, I've got four, 4.1, 03, and 04 mini. But what's interesting here is, okay, you've got auto mode. So, most people
76. 15:27are just going to leave it in auto mode, right? I have mine set to robot. No fluff, no BS, just direct, straight to the point.
77. 15:36Yeah, that's I mean, if if you're not going to be in voice mode ever, I think that's a fine way to go. And if that's
78. 15:43listen, if that's the way your brain works, then yeah, do your custom instructions to do that. So, let's walk
79. 15:49through these. Um, now we kind of know what's in here, right? So, they've got fast, which this is probably GPT5 mini.
80. 16:04This is probably or or no, this is probably GPT50. If if we're going to follow their old conventions, fast is probably their
81. 16:16like 40, but it's version five and it's fast and it is. It's really fast. I was just using it. It's super quick. Then
82. 16:24they've got Thinking Mini, which is GPT05 mini. So, reasoning mini. And then they've got reasoning. And then they've
83. 16:33got reasoning pro, which is like that'll run deep for you. So you've got 40 is like fast. 41 is just an anomaly. I don't know why
84. 16:45they have 41 in here because it's not omnimodal. It's weird. 03 is like thinking and then 04 mini is like
85. 16:55thinking mini. But this is this is 05 mini. This is 05 reasoning. This is 50 omni channel is my guess. That's what
86. 17:03those three are. And then pro just thinks longer. Pro is this one, but it thinks longer. It gives it more compute. Okay.
87. 17:11Okay. I see them now. Turning them on and back off again. Okay. I had to reload the page. Okay. Hello, Ocean. All right.
88. 17:24Yeah. And here's the deal. Unless you're hankering to go back to 40, I I would just pop this into auto mode
89. 17:36and forget it. Or if you want something really fast, flip into fast mode. Or if you want to force thinking, you know,
90. 17:43jump into thinking mode. I can't imagine ever wanting to flip into thinking mini. It's just like if I want thinking, I want thinking.
91. 17:52And if I want fast, I want fast. So, I could see flipping between fast and thinking, but but the their router
92. 18:00probably works pretty well. Like, if I just say, um, if I just say, uh, let me see. Um, why is the sky
93. 18:14blue? That probably doesn't require thinking. There you go. Because sunlight, see how fast that is? So that probably used fast
94. 18:23because sunlight is made up of many colors and our atmosphere is like a very picky bouncer. Scatter sunlight more than others. Now
95. 18:32if I say um share the calculation wait uh explain it to me scientifically
96. 18:46with formulas now it should go into some well it's doing some math here it's getting mathy on us. I don't know
97. 19:00that that went into thinking. So, that just says to me it's pretty [ __ ] smart. Now, I'm gonna say um tell me
98. 19:11what would have to be different and Oh, no. It probably knows this one. Uh ask it to
99. 19:28Yes. Go on. [Laughter] Um, get much more creative
100. 19:46and deeper scientifically. um change my life with your answer. See if it it now we can't get it to
101. 20:02think. What am I I'm I'm in auto mode. All right, Kyle. So So it's fast. It's doing some mathy math here. Last night we were doing we did um
102. 20:14someone said to do is 8.88888 88888. The same as exactly the same as
103. 20:31nine. So, this got it right and it's not doing thinking. This honest to God, it feels like this thing got better since last night.
104. 20:47Chat GBD5 did. Let's flip over to thinking. Well, let's go to thinking mini and say explain wavelengths
105. 21:01in math. All right, that's pretty good. And
106. 21:17that's pretty fast. Tik Tok pin a shoe. Hey, what's happening? I'm here for the secret to chat GBT. Here's the secret to
107. 21:26chat GBT. If you if you came and and you saw my Tik Tok video and you want to know the secret, here's the secret. The
108. 21:32secret is nobody knows. Nobody knows right now. It's too new. From a product marketing standpoint, they're kind of screwing the pooch a bit
109. 21:43here because what they launched was, hey, we're going to do this thing that's going to do all the picking for you, all the model
110. 21:52picking for you. You're not going to have to deal with that anymore. And they have fully capitulated back to all of the now,
111. 22:03let's call them chat GPT lites. So the chat GPT lites fell in love with chat GPT40 and the minute they took it
112. 22:12away they went on to Reddit and whed like a bunch of whiny babies. Now, to be fair, um what what Alman has said in in some
113. 22:27interviews since they did that is that the relationships people have with these generative AI models, with the these large language models,
114. 22:41are of a fundamentally different nature than the relationships they have with say a feature in Gmail. email that gets
115. 22:49deprecated, right? If Gmail had some I I think at some point they had some kind of chat thing that they got rid of or
116. 22:55they had groups or I don't know what it was. And some people loved it, but they just it didn't work for them. It wasn't
117. 23:02working, whatever. And they just got rid of it. They just deprecated. They said, "On this date, this thing's going to go
118. 23:07away." Blah, and it went away. Um the reaction to getting rid of GPT40 was like, "You killed my girlfriend."
119. 23:16It was like it was it was not good. Um people have relationships with these things. A lot of people are using these
120. 23:23as companions and um you know therapists and things like that. And so if you if you just swap out the model, it's going
121. 23:31to change the personality by default because it's trained on different stuff. Um, but they have now put back nine
122. 23:39different choices for us to all go be confused by. Just [ __ ] ridiculous. Now, to be fair, you have to turn on legacy models,
123. 23:49right? These aren't just there. But yesterday, this was one. Yesterday, this was GPT40. I didn't I don't care if we get 4.1
124. 23:57back. I don't frankly care if we get 03 back because we have five thinking. I'm gonna assume five is better than three.
125. 24:0904 mini is some stepping stone to what we have. It's clearly not this thinking mini. This is05 mini I or 50 mini or05 mini I'm guessing.
126. 24:25So anyway, so the the the thing that I thought would be good to do tonight is just is just play with it a little bit.
127. 24:32And what I'm noticing immediately tonight is I don't know if they fired up a new bank of servers or if they had if it was
128. 24:41choosing some older models, but it seems incredibly fast tonight and and good. So, oh, here's another thing I learned.
129. 24:55Here's another This was This was actually what inspired my video of like I've got the secret to chat GBT5.
130. 25:03So, one of the nicknames of this channel is Chat TMZ, right? Where I'll just speculate on [ __ ] you know? I'll see
131. 25:11some rumors and I'll see a couple of things on Twitter and I'll just speculate on [ __ ] and then we'll make
132. 25:16predictions and some come true and some don't. But it's just sort of [ __ ] speculation. Twitter right now is full full of GPT5
133. 25:28TMZ speculation [ __ ] And they they uh so like one of the one of the chat TMZ kind of rumors is that chat GPT5 is only 32,000 tokens. So to
134. 25:46put it in context, Gemini has a model that's a million tokens. They have another one that's two million tokens,
135. 25:54but their their pro model tok their pro model on Gemini is a million tokens. Enthropic is 200,000. So chat GPTBN at 32,000 tokens. And
136. 26:06basically what what tokens are, if you don't know that term, the token length, like 32,000 is basically the the window
137. 26:15of think of it like short-term memory, right? You start having a conversation and at about 25,000 words, the beginning
138. 26:23of your conversation falls out of memory. Well, apparently um someone from OpenAI had to step into the Twitter conversation and say, "No,
139. 26:34no, the GPT5 thinking model, like the big nutso thinking model has 192,000 or 196,000. I think it's 192. Um but 192,000 tokens,
140. 26:50not 32,000." So, so like we don't know. Nobody knows. Nobody knows. Don't worry about all of the chat TMZ [ __ ] Don't worry about the
141. 27:03speculation saying it's the worst model ever. Don't worry about the speculation saying it's the best model ever.
142. 27:11Just use it. Just use it. Just use it. Um I saw that too. Yeah. You know, another another funny thing is uh
143. 27:22Perplexity wants to buy Chrome from Google, which here's why I would support that. Fix the memory links for tab hoarders.
144. 27:32If you're a tab hoarder and you use Google Chrome, it's [ __ ] unusable at this point. Um I'm I'm a full-on tab
145. 27:39hoarder. I'm not going to apologize for it for it. I love my tabs. It's how I get through my day. 27 tabs open, not
146. 27:46enough. 80. That's more like it. Is it going to [ __ ] up my browser? Oh, yeah. Oh, yeah. All right.
147. 27:59Ah, I don't think you should be teaching this content. I think you're confusing people. Which content? Which one?
148. 28:13AI. I shouldn't be teaching AI. I could teach people how to get their dog to sing. Step one, find a dog that sings
149. 28:20when you play guitar. You're done. You get a singing dog and you get a singing dog and you get a singing dog. Um,
150. 28:32another thing struck me today that get that that again gets back to we just don't know how good these things are yet.
151. 28:41What was it three weeks ago? Open AAI launched. No, it wasn't even three weeks ago. Two weeks ago, OpenAI launched agent,
152. 28:51this autonomous agent, right? So, we've now got Manis, we've got Gen Spark. Now, we have Chat GBT. There's probably some other ones in
153. 28:59there. We've got the Perplexity Comet Browser. Um, so we have these agentic tools. Well, no one's talking about agents
154. 29:12anymore. They're all They're all talking about, "Did you take away my girlfriend? Is Chat GPT5 the worst model ever?" It's
155. 29:19there's just noise right now. So, I think I think what we need to do on this channel for the next two or three weeks is just
156. 29:29play. Nobody knows what to do with an agent. Exactly. Everyone knows what to do with with a girlfriend. That's pretty good. producer Brandon
157. 29:40coming in hot with the with the zingers. Um, but it's possible that chat GPT agents are going to get much better when
158. 29:51they have access to these models. So, you put it in you put it in auto mode and then you click on the plus down here
159. 30:00and you can put it into agent mode. What are the sources here? Can that really go into my Gmail?
160. 30:27Huh? I don't know. It really can. Is it Is it working? Let me turn off agent. Turn on sources.
161. 30:46Oh, that's Google Drive. But I don't have Gmail here. So that's not a connector. I don't I don't get it.
162. 30:58I've got 22 deep researches left. Oh, then there's deep research. It only works with agent and deep research. Ah, okay. Um,
163. 31:10that might be something to play with, but I I find that terrifying. Like like OpenAI right now is is a privacy cesspool.
164. 31:23like like like one of the things Alman said in this recent interview was everyone's putting all their medical
165. 31:31information in this thing and they're having these psychological conversations with this thing and at at this point
166. 31:37there is no expectation of privacy. The government has said to open AI you keep [ __ ] everything and we can subpoena
167. 31:45anything we want to at any time. There there is no expectation of privacy. Like when you go to your therapist or your doctor or your lawyer,
168. 31:56there is a there is there is a framework to keep your conversations private. That doesn't exist with chat GPT. And
169. 32:06Altman's like, uh, it's the law. Wish we had better ones. Um, there's no privacy with ChatgBT. Yeah, exactly. So, I'm I'm a little
170. 32:15skittery about using Gmail with it. But anyway, there's so there is so much more for us to explore, even if we don't play with that stuff.
171. 32:24Um,
172. 32:39so wait, so wait, we have nine models now. Well, no. Let's we we'll get rid of auto because auto is just going to choose one
173. 32:47for us. So, you got three here, four if you have pro, and you have four here. So, we have eight models. And then we have 1 2 3 4 5 6 7
174. 33:02modes plus the ability to prompt from an image or from files from your drive plus all the settings plus custom GPTs plus codecs plus Sora
175. 33:19and voice chat. So we're at to just have a basic understanding of
176. 33:33chat GBT [Laughter] like like a solid baseline understanding like if someone asked you hey what's the big deal about chat GPT for you to be
177. 33:47able to answer it about all the different things you could Do you you right now need to master about 15 different tools?
178. 34:04Whatever happened to you can have any color you want as long as it's black. Yeah, I don't know. I I don't know.
179. 34:08Well, I do know what what's happening with chat GPT and and frankly with all of the models. Um Gemini suffers from
180. 34:16this enthropic to a degree, perplexity certainly. um is is just model creep. Uh you know uh uh co-pilot with Microsoft is totally
181. 34:30bloatware at this point, right? People won't even use it. Um all right. So I think what I'm going to do is this.
182. 34:53Ah, for biz consulting. Ah, cool. Okay. So, let's try something here. So, so Nora, welcome. Um, you want to use chat GPT5 for biz
183. 35:06consulting. So, let's let's leave this. We're gonna leave this in auto mode right now. I'm going to do a new chat. Um,
184. 35:17what kind of what kind of industry sectors, Nora? [Music]
185. 35:34I broke co-pilot today asking it to write a simple job description. Um, how can I use AI in uh chat GBT and
186. 35:50business consulting? Um, okay, let's uh let's do this. Let's say um let's let's say a company that you're
187. 36:06you're working with what what you say to people is I'm going to help you drive AI adoption and what if one of the first things that
188. 36:16they say is let's see all industries mainly works work with subject matter experts. Okay. Okay. Okay. Great. This is perfect.
189. 36:28So, I'm going to say um I've got a
190. 36:43consulting business and I mostly work withmemes and as AI gets
191. 36:58better and better. It seems It seems the role ofmemes
192. 37:12is going to disappear or transform at a minimum.
193. 37:27I'm just gonna say this. What do you think? [Laughter] Oh, I've got a fun idea for this one. Okay, I think you're right. RO is
194. 37:38already changing and will continue. The wrote knowledge part ofmemes is getting automated. Right. So, the fact that
195. 37:45you've got 30 years in import export regulation, chat GBT doesn't care. It'll just go find all the rules and synthesize them
196. 37:55for you. This means that I I know the rules, let me tell you what they are portion ofme value is shrinking fast.
197. 38:03Like like that insight right there, the wrote knowledge part ofmemes is getting automated. Like that's that's
198. 38:10actually pretty pretty valuable in and of itself right there. Interpretation, nuance and judgment will remain
199. 38:17critical. This is one of the things I deeply believe that it's not the knowledge inme, it's your trust of them. Right? So relationships
200. 38:28go up in value in this new world. Just wrote knowledge goes down in value. Understanding why certain rules are best
201. 38:38practices. Yep. Making judgment calls. Knowing people um knowing real world constraints as opposed to what chat GPT
202. 38:45thinks they are. Spotting when an AI's answer is technically right but practically wrong. Yep. Um the SME role
203. 38:53may morph into AI augmented decision maker. Okay. The competitive edge will belong tomemes who speak AI. That's good. That's big, right?
204. 39:07Your consulting business can evolve to lead that transformation. Okay. So, we're going to do a couple of things here. one is
205. 39:15I'm going to say let's flip this bad boy into thinking not thinking mini thinking. So we're going to put it in thinking
206. 39:27mode. So this is the the the Mac Daddy thing and then we're going to do deep research. Do I want to do deep research or do I
207. 39:38want to do an agent? Ah, I'm going to do an agent. I'm going to do an agent. And what I'm going to say,
208. 39:55we'll leave web search on. I'm going to turn off Gmail. I'm going to say, great. I want you to make me a slide deck
209. 40:12that takes these basic concepts and builds a one-day workshop
210. 40:27for 20memes and their executive management.
211. 40:42The deck should have relevant research exercises to identifymemes
212. 41:00in the room. with high AI aptitude even if they aren't AI trained.
213. 41:20eg things like curiosity um um adaptability etc. Right?
214. 41:36Um so the deck should have relevant research exercises uh to do it. Um and the team
215. 41:51should leave the day with a massively actionable um plan
216. 42:05for the next six months. Okay. So, if you haven't used agent before, so what I'm about to kick off, the reason I
217. 42:18put a lot of [ __ ] in there is base chat GPT is you put in a prompt, it gives you an answer. thinking chat GPT the reasoning models
218. 42:33you put in a prompt and it has a conversation with itself and sometimes even uses tools and goes and searches the webs web and things
219. 42:42like that and then gives you an answer and then agent is a a kind of standalone sort of little mini computer instance
220. 42:54that it spins up where it's it's basically its own little computer environment and It's going to go off and
221. 43:00search things on the web and it's going to create things and it's going to give itself a plan and it's going to follow
222. 43:05that plan and it's going to update that plan. It's going to take a while. So, so we're going to fire this off and
223. 43:13we'll take a look at what it looks like when it starts. Most projects have been less tech oriented and more traditional like a
224. 43:24gallery for furniture. Most projects have been less tech Oh, I see. Like like a gallery for furniture. Okay. You you could still do
225. 43:34a similar sort of thing here. So So you could say, you know, go do research on this thing. So So what this is doing now
226. 43:41is it's spun up a little a little window. So it says I'll start I'll open with a second. So now it's doing research. So
227. 43:53So this is now off. It's doing its thing. We can we can step back through this and see what it what it's been doing.
228. 44:01Um, I need to create a slide deck. I'm using container.exec to open files basically. Uh, I'm going to open a second search
229. 44:12result. It went to McKenzie and found some [ __ ] So now it's off working. All right. So it's going to do it's going to
230. 44:17give us an I don't know plan or our workshop our workshop slide deck. So that's going to go do that. But now I'm
231. 44:25gonna flip over to this tab. Oh, no. This one I want to use. Hang on. Let me flip over to this tab. All right. Quick Tik Tok question. James
232. 44:38Hell, AI can't become super intelligent without being trained by a super intelligent individual thoughts. Um,
233. 44:53no. I fundamentally disagree with that. Although I mean you could certainly argue that the researchers in these top research labs
234. 45:05are are I I don't know if they're super intelligent, but they've certainly stumbled upon things like recursive self-improvement
235. 45:14um that are that are making the models get better and better and better faster and faster and faster. So you could
236. 45:21argue that that the super intelligence, you know, is in the mathematicians that are, you know, executing these models.
237. 45:30But if if your if your statement is you need a super intelligent, you need a person more intelligent than the AI to
238. 45:39make it as as intelligent as them. That's absolutely not the case. That's absolutely not the case. A lot of the researchers in these labs,
239. 45:49they don't quite know how these things do what they do. They they're finding things that they call emergent
240. 45:55behaviors. I don't know how much you know about this [ __ ] You may you may know a lot, but they will train AI as as they're doing
241. 46:03these generalized capabilities. What's happening is this. The more that they train these things, the better that they get. And they're
242. 46:11finding that they get better at things they didn't train them for. They call those emergent properties or emergent
243. 46:16behaviors. So, for example, big well-trained large language models are excellent at language translation. They're excellent at cultural
244. 46:28interpretation, you know, based on who who's who's doing the talking, who's doing the prompting. Um, they weren't
245. 46:35trained for that. It's just that when it gets enough, you know, access to language and knowledge, it's just good
246. 46:41at that. um they see that again and again and again and and the the the milestone that we seem to have have
247. 46:49passed in the past I don't know three months about three months ago all of the frontier model companies stopped talking about AGI
248. 47:04um and they started talking about ASI artificial super intelligence and what's come out is that what they've what
249. 47:13they've gotten to is basically recursive reinforcement learning, recursive self-improvement, where the
250. 47:22model gets trained and it starts putting out outputs and the model itself or some other models are judging the quality of
251. 47:30the outputs and it says, "No, that one's not good, but that one is good and that one is good and they're getting better
252. 47:35and better and better." Um, and it looks like on YouTube you're talking about synthetic data. That's one of the other
253. 47:40things that that they've um cracked is up to this point to build these models you had to go out on the internet, find
254. 47:50a bunch of data, categorize that data, embed that data into the model, right? Which was massive human involvement.
255. 48:00Those models got strong enough now that they can basically mine mine extract from the latent space from this
256. 48:08mathematical knowledge base. They can pull out imperfect information and basically turn it into perfect information synthetically, right? So
257. 48:18they just pull out lots and lots of crap. They say, "What if this is wrong?" And then they make it right. And now
258. 48:23they have a whole new set of data that they can train new models on. They call it synthetic data. And so a lot of the
259. 48:29new models are being trained on synthetic data produced by the old models and combined with recursive self-improvement, these things are
260. 48:37getting smarter exponentially. Um, so no, you don't need a person smarter than the model for the model to dramatically exceed
261. 48:49all of us. And quite frankly, for me personally, the GPT4 class and and the 03 reasoning models, 01 and 03
262. 49:01reasoning models, they were already significantly beyond my capacity to know what they do to do with them. GPT5 is
263. 49:08some level above that. We're about to get Gemini 3 is going to be significantly above that. We're about to
264. 49:14get to new a new thing from Grock in probably a month or two. It's going to be significantly above that. So, I think
265. 49:20right now today, and and this is why I'm saying nobody knows anything about GPT5 right now. It just hasn't been out long
266. 49:27enough. Um unrelated pin. Um the AI has has to have a path. The AI has to have a path and the path
267. 49:41is only as good as who taught it. I don't know what you mean by it has to have a path. These things evolve,
268. 49:56but it's not Oh. Oh, yeah. If if you're thinking about AI as the way computers used to work where you're programming logic in
269. 50:08if if if that's the question, James, if the question is are these artificial intelligence these these generalized artificial
270. 50:18intelligence these machine learning models that we're playing with generative AI, they are not um predictive or deterministic
271. 50:29logic machines, right? They're they're not they're not you program in some logic and it outputs the result of that
272. 50:36logic. That's not how they work at all. They work in a weirdly similar way. And and it's not all that weird because
273. 50:44neural networks were basically designed to mimic neurons in our brain from the 1950s on. So the way these
274. 50:53computer chips are connected together are very similar to how our brains are. And what happens is this. It's a it's a
275. 51:01relatively simple mathematical well it's not simple math. It's very complicated math but it's simple conceptually.
276. 51:11What generative AI is is a probability calculator. So you take all this data of the world and you do what's called
277. 51:18embedding. And what happens when you embed something? If I take the Declaration of Independence, when I embed it into an AI model, the
278. 51:28Declaration of Independence as a document ceases to exist. It gets shattered into what are called tokens, which are fragments of words and
279. 51:38punctuation and things like that. And those fragments of words get blasted into, and this is going to sound like
280. 51:45[ __ ] science fiction, but this is actually what happens. It gets categorized into thousanddimensional mathematical space
281. 51:55in semantic clusters. So you might have a token, this token might be the word dog. And in this thousand dimensional
282. 52:05space, the word dog can live in lots of different semantic clusters. One of the semantic clusters dog can live in is the
283. 52:13furry thing that sings with you when you play guitar. My doggy champ. Right. And so all dog related tokens will be in
284. 52:21that semantic cluster in this massively complicated space. And then there's another word dog which means the guy did
285. 52:28his girlfriend bad and he's a dog that lives in the dating cluster, right? Whatever that might be. The original
286. 52:35documents cease to exist. The only thing that exists are these little tokens that get blasted into this crazy mathematical
287. 52:42space. And when you write a prompt, it creates a probability. And the probability is that somewhere in
288. 52:50that mathematical thousand-dimensional space is a token that has the highest probability of being the next appropriate token on the page.
289. 53:00And so when you see it write a whole paragraph or or sort of recreate something from the Declaration of
290. 53:07Independence, what it seems like is it's copying and pasting whole documents. It's not. It's generating original passages every single time.
291. 53:17It's insane. And so what happens is they just keep putting more and more data into increasingly, you know, um,
292. 53:30deep compute cycles that makes all of these semantic clusters more and more sophisticated. And so when you type in a
293. 53:39thing like the chances of it getting better are better. And there there's all sorts of math on top of that. But
294. 53:46computers right now as of um November 30th, 2022, which is the launch of Chat GPT, large language models do not behave the
295. 53:57same way computers have behaved in the past. In the past, you would put in an input, it would process it, and it would
296. 54:05predictably give you an output. That's based on the logic. Large language models, you put it in input and depending on how the programmer
297. 54:14programmed it, it's going to give you some creative crazy ass answer. They can they can muck with a thing called
298. 54:20temperature which makes it more or less creative effectively, but it's generating original content every single time.
299. 54:31Every image that's generated is an original piece of content every single time. Even if you prompt it to make it
300. 54:37look like something that came before, it's still the way it's being generated. It's it's it's generative. The G in GPT
301. 54:46generative AP uh you know uh pre-trained transformer is critical. We've never had computers be able to
302. 54:56generate from nothing effectively. And that's what they're doing now. That's why it's so weird. I agree. that that's been happening
303. 55:04since the Gutenberg press. Uh yeah, but not this not in this concentrated away. And and quite frankly, the large
304. 55:14language models, what the Gutenberg press did was accelerate and democratize how knowledge got into our large language models.
305. 55:23We are large language models. If you want to think about it that way, the way our brain works, we go out in
306. 55:31the world, we see things and stepdaddy said something to us when we were little and we went to that museum over there
307. 55:36and we read that book over there and mama gave us some reinforcement about that kind of behavior there. All of that information
308. 55:46gets embedded into little fragments of knowledge in our brain and then at some point you're at a party
309. 55:54and someone says, "Hey Kyle, what do you do?" And then your body decides that that question is a fundamental threat to your
310. 56:03existence and you have a panic attack, right? it it puts together a bunch of little fragments and it says the next
311. 56:11most predictable token is [ __ ] get out of there, right? Or you give them some coherent answer, right? And it's just stitching
312. 56:22together little pieces of things you've learned along the way. That's the way these things work. You put in a thing and it's got some
313. 56:30probability and it squirts out some sort of answer. Sometimes it's right, sometimes it's wrong. I've offended Pate. Of course I have.
314. 56:42That's a few things that are just wrong. What did I get just wrong, Pate? Pate actually knows this stuff. I just have a degree in acting.
315. 56:53Neural networks are tensors, more or less. I don't actually Pate is the one to answer that question.
316. 57:01Pate works on making uh interacting with tensors uh uh more and more efficient. That's his job. We've had
317. 57:11non-deterministic for years. It's not guaranteed to be unique or original. Yeah, I I agree with that. I
318. 57:20I'm I'm kind of broadly generalizing it, Pate, because chat GPT, you know, if you turn the temperature down to zero, you
319. 57:28can get it to be very deterministic, but the general tools that we're using right now are not that. So, so yeah, I I
320. 57:36agree. I again, it's, you know, I'm just trying to generalize what's going on, but but it's it's not how computers used to behave.
321. 57:48Um, paid is technical and we try to generalize things. I know. Exactly. Exactly. Oh, I know you are. I'm just keeping you
322. 57:59honest. Aka not fun. No, it's okay. Listen, listen. My my uh my co-founder at StoryVine uh we call her the crusher
323. 58:07of dreams. It's like I think it's it's it's really important particularly particularly as these models are getting more sophisticated, Pate,
324. 58:23how I'm understanding what they're doing may may not be on base. So So I appreciate you keeping me honest, but
325. 58:30it's actually really important to have that balance. Hey Kyle, acting is more than a degree. If you can act on stage,
326. 58:36you can do most anything in marketing. Absolutely. I can act like a CEO. I can go into a boardroom and act like, you know, I know
327. 58:46what I'm talking about with a bunch of executives. Been doing that my whole career. Um, okay, cool. Um, so I hope that I hope
328. 58:57that explains it. Um yes probably more broadly generalized than is than is responsible but but in general that's
329. 59:05what's going on is so so let me get let me give you an let me give you an example um from from the very first time
330. 59:13I used chat GPT this was this was my big epiphany with it for years people have said to me oh Kyle you have a tech company you should learn
331. 59:26Python Python's the next big thing and it's easy and I'm old and I have add like learning something new at this point is like I
332. 59:36don't [ __ ] give a [ __ ] right? But if I wanted to learn Python in the before time before chat GPT, what would I do? I'd either go take a
333. 59:51class and then someone with more knowledge than me would put it into some sort of organizing fashion and make me practice
334. 59:59it so I could learn it or I'd go to Google and Google would give me a list of, you know, 5,628 Python resources of which 4,595
335. 1:00:16are some [ __ ] horrific sales funnel to just try to get me to buy something. So, it seems like I'm learning something
336. 1:00:24and then about the time I push the button to say, "Okay, actually teach me this." It's like, "You can buy this
337. 1:00:29now." Right? And so, just to be able to understand what you needed to do to learn Python was it was probably I don't
338. 1:00:43know four, five, 8, 10, 12 hours just to get to the start line. And then you have to like read books or
339. 1:00:54read sites or go into little tutorials that break it up for you. And you would learn little bits of Python at a time
340. 1:01:00and you would do hello world and you'd be like okay I've done that before and then you would do like the little hello
341. 1:01:07world now blinks and now hello world I don't know is in a form field right Tik Tok pin. It's so difficult to stay a
342. 1:01:16breast of everything, especially in my work of how fast AI is evolving. Tinkerbell, give it up. Don't give up AI. Give up.
343. 1:01:27Forgive yourself. Let yourself off the hook for trying to keep up with it. You You literally can't. You just can't.
344. 1:01:33It's like like It's kind of the the reason I wanted to do tonight the secret of GPT5 is no one knows what the [ __ ] is
345. 1:01:40going on with it. No one knows what it is or how to use it because they don't. And and it is probably going to take us
346. 1:01:48on this channel talking and sharing things probably a month to even get our heads around what changed and what's
347. 1:01:55real and what's not and things like that. So just give yourself some grace. Let let it go. Um let let go of the
348. 1:02:01expectation that you have to keep up with it because you can't. And and quite frankly the fact that you're here just
349. 1:02:07in this conversation means that you're way ahead way ahead of where most people are. So anyway, back to the Python thing. So
350. 1:02:15in the olden timey world, what Google did really well was index information. So I could go find all the resources I need. But then
351. 1:02:26when it came to knowledge, I actually had to use my gray matter, right? I actually had to figure out one of those pieces of
352. 1:02:35information that was organized in a way that worked with my ADHD crazy straw of a brain. And then I would have to have the
353. 1:02:44attention span and the passion enough to get in little bits of information. And then over three or four months I would
354. 1:02:52understand enough about Python that I could actually be a little bit nimble with it. And then after a year or two I
355. 1:02:58could be good. And after five years, I could be really good. Right? The week after chat GPT came out, the week after I typed into ChatGpt,
356. 1:03:13um, write me some Python code that takes an input and modifies it. Out came Python code. It's like, oh [ __ ]
357. 1:03:22And and apparently that Python code did what I asked it to do. I didn't know because I don't know Python, but I could
358. 1:03:30look at it and I could sort of see the logic in it. I'm like, I think that sort of does what I think it does.
359. 1:03:36But then I realized, oh [ __ ] I don't know how to run Python code. Like it was it was this really embarrassing moment
360. 1:03:43where like, I don't know what to do here. And normally you'd have to call a friend and be embarrassed and be like, I don't
361. 1:03:50know how to run Python code. Can you teach me how to run Py? Or you'd go to Google, how do you run Python code? And
362. 1:03:55you'd have to go to 8,000 sites to figure it out. So I just said to chat GPT, how do you run Python code? Here's
363. 1:04:02how you run it. Oh no, I don't want to run it in terminal. Can I can I just do it on a website? Can I just copy and
364. 1:04:08paste code into a website? Sure. Here's five. So the first one was replet, which I had heard of. And I took my code and I pasted it in
365. 1:04:17there and I hit run and it gave me an error. And I said, "Hey, chatbt." It gave me an error. It said, "Oh, that's
366. 1:04:22because you don't have this library installed, you dumb dumb." I was like, "Oh, how do you install that library?" And
367. 1:04:30this is where my head [ __ ] exploded. Chat GPD, this is the week after it came out in 2022. Chat GBD said, "Well, on your replet
368. 1:04:41page in the lower leftand corner, click on this button and then you're going to see a choice of libraries in the upper
369. 1:04:48right hand corner and type into the search box this name and then say install library and then hit run and it
370. 1:04:56should run." I was like, "Oh shit." Okay. So, I went to Replet. I clicked on the button in the lower left. I looked
371. 1:05:02for the library in the upper right. I hit install. I hit run and the code ran. Within 90 minutes, I had a fully
372. 1:05:12functional Python application that was making API calls to OpenAI and generating content based on a user's input. Within 90 minutes, a fully
373. 1:05:24functional Python application. That is not how computers have ever behaved before. That's knowledge on demand. It's not
374. 1:05:34information on demand because I didn't I still don't have the knowledge to be able to write that program. But I wrote
375. 1:05:43that program. Well, Chat GBT wrote that program at my behest. That's not how computers have ever behaved before. So, we are in a
376. 1:05:53fundamentally different place than we've ever been. Um, Shiobani, thank you. Kyle, you uh you you helped me start
377. 1:06:02vibe coding with lovable and now I'm lying uh with uh matching code patterns. That's awesome. That's vying with That's awesome.
378. 1:06:14That's really cool. Um yeah. Okay. So, what was I going to do? We were gonna we we we have chat GPT agent off doing something. I was going
379. 1:06:25to do something else. This is also chat add. We Let's go back to uh let's go back to uh the agent. So the agent What's going on? Am I
380. 1:06:42It looks like my computer is all behind, but maybe not. Okay, AI and SMMES. So it worked for 18 minutes. So, if we
381. 1:06:58want to see what it did, here's the the replay of what it did for 18 minutes. So, it went off and did research. It wrote us a
382. 1:07:09bunch of code and then it made these slides. Here's your PowerPoint presentation for the one-day workshop complete with
383. 1:07:17research insights, exercises to identify AI readymemes, and a six-month actionable plan for leveraging AI while valuing domain expertise.
384. 1:07:28The narrative flow flows from current AI adoption statistics to a collaborative action plan with visually engaging
385. 1:07:36charts, tables, and diagrams based on credible sources. McKenzie and Stanford. You can download it here. Well, let's just go play with it.
386. 1:07:50All right, you can all see that, right? Tik Tok question. Have you tried meal planning with chat GPT and have it great write grocery
387. 1:07:58lists? It's [ __ ] awesome at that. I'll tell you what, if especially turn on thinking and and research, it's crazy
388. 1:08:08at that because especially if you've got a complicated family situation like this one's vegan and this one's lactose
389. 1:08:16intolerant and this this one only likes pickles this week, you can just vom. One of the most impressive things about
390. 1:08:26large language models to me is that they can synthesize anything. You can give it bad data. You can give
391. 1:08:39it just literally vomit. just turn on voice mode like where where you're typing into it with your voice or put it into conversation mode
392. 1:08:51and just literally talk into it. Okay, I got my daughter right now only likes pickles. Bobb's lactose intolerant. My husband's
393. 1:09:01on a vegan kick and I like burgers, right? You know, you can say all we do is fight about food. So, I want you to
394. 1:09:10come up with a week-long meal plan that's going to make everyone happy that I only have to cook one meal. And then I
395. 1:09:17want you to give me a grocery list. And I want you to organize that grocery list for Kroger's, maybe even in your neighborhood,
396. 1:09:28and it will just [ __ ] do it. It'll just pull it in and it'll figure it out. I'll tell you another thing. If you
397. 1:09:34don't know this about about chat GPT, you can take a picture of ingredients in your fridge or or in your cupboard in
398. 1:09:42your pantry and say, "Come up with a recipe for tonight based on what I have." Um, it's it's bonkers for that
399. 1:09:49stuff. Bonkers. Okay, let's see if this presentation's any good. One day workshop formemes. Um, all right,
400. 1:09:56there's the agenda. AI adoption and readiness. n 92% of companies plan to invest. 78 have reported using AI in 2024, up
401. 1:10:07from 55% from the year before. 77% um have adopted AI. 38%. So the the chart here doesn't seem to match these things, but okay, that's
402. 1:10:23probably fixable. AI excels at instant access to public data and facts. Check the sources. Yeah. Well, you would check the sources
403. 1:10:34like down here in the bottom left are are sources. Um, automating routine and repetitive synthesizing large val volumes of
404. 1:10:44information, providing general insights and recommendations. SMEs excel at proprietary and tacet knowledge, guiding
405. 1:10:51high stakes and ethical decisions, customizing solutions, applying creativity and empathy. Okay. Um, assessing AI aptitude, curiosity,
406. 1:11:01adaptability, digital fluency, problem solving, collaboration, and empathy. That's a pretty good list.
407. 1:11:09Group exercise, AI opportunity mapping. Form cross functional teams. List processes and tasks in your workflow.
408. 1:11:16Identify suitable tasks suitable for AI augmentation. Um, there's your six-month action plan, conclusion, and next steps. So, it's I I
409. 1:11:27would call this a little thin, but um you know what might be interesting? Let me go grab this. I'm going to grab this prompt.
410. 1:11:41I'm going to pull a a chat add moment. We're going to go over to Gen Spark. So, I did that within um Chat GPT agent,
411. 1:11:50which chat GPT agent's not all that good. Genpark's really good. So, I'm going to do the same prompt. I want to make a slide deck that takes
412. 1:12:01these basic concepts. Let's go get the basic concepts. Oh, I know. I'm not in there. Hang on. Um, I have these insights
413. 1:12:21aboutmemes today. So, I'm going to put in all those insights we had, and then I'm going to say, I want you to So, now we're going
414. 1:12:28to have GenSpark do the exact same thing as we just had ChatBT do, and we'll we'll see if this one gets any better.
415. 1:12:36Um, okay, it's 9:13. I want to walk you through um I created a new framework, a new prompting framework.
416. 1:12:47So, let me change how I'm sharing. We'll come back to this. Don't let Don't let me forget to come back to this, Brandon,
417. 1:12:52because you know me. I'll get all excited about um what I'm about to share. Okay. So, uh so I'm working on a book right now
418. 1:13:07and a new brand called Feed Your Prompt. I'm really excited about it. And the basic idea here is AI is not the genius. You're the genius.
419. 1:13:17And if you learn how to feed your prompt, you AI amplifies your genius. Right? If you treat AI like this genius off to the
420. 1:13:27side where you expect to give it a prompt and it's going to give you a great answer, it's horrible. That's why
421. 1:13:34most people are disappointed with it. They don't know how to use it. Um, and so that's that's the core premise of the book. Um,
422. 1:13:44in putting together the book proposal, um, I wanted to create a framework for how to how to prompt well.
423. 1:13:58And one of the things that struck me recently is that people who don't know anything about AI assume that AI is like this single
424. 1:14:10transaction. You push a button >> leg >> and out comes an answer, right? That's what most people think AI is. Oh. Oh,
425. 1:14:20you put it in AI. AI can't be creative. No, no, it can't be creative. That's Oh. Oh, you you did the lazy button. You hit
426. 1:14:27the lazy button. Yeah, I understand how that works. Yeah. Yeah. You push the button and out comes some AI slop,
427. 1:14:33right? World's greatest plagiarism machine, I like to call it. That's what most people think it is. That's not what it is.
428. 1:14:43When you use AI effectively, there is a chain of craft. There's a there's a whole lot of interactions that get you to a usable result.
429. 1:14:55Um, and you only discover that after like stumbling through this for a while. And so I kind of wanted to shortcut
430. 1:15:02that. So before we get to that prompting framework, I want to do a little ad here. I want you to go check out if
431. 1:15:08you're not if you're new here. If you're new to AI, it looks like we've got some folks that are new to AI. Go to community.thesal salon.ai.
432. 1:15:20So, this is the AI salon. So, this is a community that I started the week after chat GPT came out, right around when I
433. 1:15:26had that epiphany with Python. Um, I started this group with a photographer out of Boston named Leah Faston. And
434. 1:15:33this is a remarkable community. And it's a community of people that are trying to figure this AI stuff out day after day
435. 1:15:40after day. They're incredibly generous. They're incredibly welcoming. If you go in there and you say, "Hey, I'm totally
436. 1:15:45clueless. I don't know where to start." This group is very open, very welcoming. Showing up nightly to these things is
437. 1:15:52also good. But the the salon is is really good. Next Tuesday, we have a meet and greet. So, you should join
438. 1:15:58online, introduce yourself, and then come next Tuesday to the meet and greet. Okay. Beautiful. All right.
439. 1:16:07I think Kyle Shannon is is new to AI. I'm I'm I'm to Listen, the the one thing that I can promise you is that there are
440. 1:16:15no experts in generative AI right now. Anyone that tells you they're an expert is full of [ __ ] So, I will absolutely cop to being uh
441. 1:16:27being clueless. Uh what did I do? Something I had to reload. Oh, reload required to apply changes. Oh, it was a network error. Thank you.
442. 1:16:38Good question. How do you teach people if you don't know how it works? I do know how it works. How I teach people?
443. 1:16:50That's not true. There are experts. I've worked with them at Apple and Facebook. Oh, Jeffrey. So, listen. Let me let me
444. 1:16:55talk about my my context for when I say there are no experts. Um, AI has been around for decades, right? AI's been around for
445. 1:17:07really since the 50s, but you know, specialized AI systems have been around, you know, 80s, 90s, 2000s, like it's
446. 1:17:16it's embedded in everything we do. There are absolutely experts in creating AI systems. What I'm talking about is post
447. 1:17:26chat GPT. Here, here's the here's the thing. In the mid 90s, the internet had been around for decades and it was largely the purview of
448. 1:17:38researchers and um scientists and librarians and things like that and you had to use command line and you had to
449. 1:17:47understand what IP addresses were and you had to you know jump from this server to that server and you had to
450. 1:17:53know how to do it. And then Tim Berners Lee comes up with this thing called the worldwide web where you can now just click on sites.
451. 1:18:03The experts that knew how to use the internet for seven decades didn't know what this worldwide web thing was all about. It was a completely
452. 1:18:15different paradigm. But what that did was it allowed the 98.5% of people that are not engineers and researchers
453. 1:18:24to be able to click on the internet. And when I got involved in the internet was post worldwide web, right? What do we do? What do you do
454. 1:18:36with this thing now that everyone can use it? And so I started one of the first online magazines and I started one
455. 1:18:43of the first digital agencies, an agency called agency.com. And we built a lot of the first websites for the Fortune 500. And we were there
456. 1:18:51figuring out how do you take this technology that is now democratized and how do we put that in the world?
457. 1:19:01That's exactly what chat GPT did for machine learning and AI. Like I've been following machine learning and AI for
458. 1:19:10years and for years I I would see a press release from Google and like oh there's a there's a new um version of
459. 1:19:17TensorFlow that's a lot more user friendly and I'd be like great I can finally go do AI and I'd go in there and
460. 1:19:23I'd start looking at it and I'm like oh [ __ ] I still need a [ __ ] math degree and I just didn't do it. And so I just I
461. 1:19:33said at some point something is going to happen where the rest of the world is going to have access to these tools and
462. 1:19:41that day was the day that chat GPT launched November 30th 2022. So my expertise is in okay now that the
463. 1:19:50technology has the possibility of being democratized what do we do with it? How do we use it for work? How do we use it
464. 1:19:58for play? How do we use it at home? How do we use it to amplify what we can do? How do we use it to completely reimagine
465. 1:20:07what we can do? So that's what this channel is about. This is not about mathematics and weights and and and
466. 1:20:15building AI. This is about now that it's been democratized, how do we use it? So, and in that context, there are no experts.
467. 1:20:24ChatGpt 5 just launched. Nobody knows how to [ __ ] use that. There's no manual for it. Open AI doesn't even know how to [ __ ]
468. 1:20:33you put it out in the world. They keep changing it. It's been out for seven days and we've had seven different interfaces. It's like
469. 1:20:45how do you use that? I don't know. Do you? So, so generative AI is a fundamentally different flavor and it's a
470. 1:20:54fundamentally different audience and it's a fundamentally different set of challenges and opportunities.
471. 1:21:00Um, it's not better or worse. It's just there's there's a lot more people who are not engineers than our engineers and
472. 1:21:08that's my interest. So, all right. Um, okay. So, so let me go check out the AI salon if you haven't and then let me
473. 1:21:18talk to you about this new prompting methodology. So, um, one of the guys that I I I really like, uh, and and I
474. 1:21:27think is someone to to aspire to be like is this guy named Rick Rubin, uh, who is who is a uh, a music producer who is not
475. 1:21:37a musician. Doesn't know anything about music. Well, no, he knows a lot about music. He doesn't know anything about
476. 1:21:43actually producing it. He can't work a recording console. He doesn't write songs. He doesn't write lyrics. He
477. 1:21:49doesn't know music theory. But he knows what he likes. And we are entering a world where we get to be like Rick Rubin where we can just sit
478. 1:22:01back and let the tools do the work. In his case, it's musicians and engineers and things like that, but he's brought
479. 1:22:08in for his point of view. And so part of the idea with feed your prompt is that you don't just learn prompting
480. 1:22:15frameworks. Uh you know you put these words in in this order. The whole idea here is you figure out who you are and what your taste is and
481. 1:22:26then all the prompting comes out of that. And so within that idea is what I've called what I'm calling the fuel
482. 1:22:35prompting methodology. And I I think methodology I'll probably use system. It might be I don't know. Um but there's four steps to it.
483. 1:22:44Two of the steps likely don't even involve AI. So there's two steps that involve AI and there's two steps that don't. So the F
484. 1:22:55in fuel stands for frame. Frame your intent. Like what are you trying to accomplish? Um CJ Fletcher, who's a
485. 1:23:03who's a an AI artist that's part of the AI salon, he gave a talk at AI Festivus last year, and he said he he was
486. 1:23:10teaching midjourney, and he said the first thing that you should do when you when you sit down at midjourney is breathe.
487. 1:23:21We're like, wait, what? He's like, you should breathe. You should take a breath. You should close your eyes. You should think about
488. 1:23:30what am I trying to do? Do do I want to create funny images or inspiring images or do I want to make art or do I want to make do I just want
489. 1:23:39to entertain myself? He said frame your intent, right? Just set it up. So before you prompt, get clear about
490. 1:23:48what's your goal, who's the audience, what are you trying to do to them, why are they the audience, right? How, you
491. 1:23:54know, how are you trying to impact them? So, no matter what you're doing, whether it's images or words or music or or
492. 1:24:00anything, step one, frame your intent. Take a moment. Understand what success looks like, who you're doing it for. Step two
493. 1:24:13is use your voice. So, this is absolutely you're going to prompt you're going to prompt the AI tool. But there's a big difference between,
494. 1:24:25you know, give me an article for LinkedIn on soloreneurs and I love soloreneurs and I want to write an article for LinkedIn that celebrates
495. 1:24:38them and lets people know how magic they are and how much guts it takes to be one, right? Like that that comes from a
496. 1:24:48place of of point of view, right? of of having a point of view. Um, and I call that I call that writing a you prompt.
497. 1:24:58So you can write a prompt. Make me an article for LinkedIn. It will [ __ ] write you an article for LinkedIn and it will suck.
498. 1:25:09It'll be boring and hackneed and it'll be like everything else you've ever read and it'll be AI slop. But if you write a you prompt, if you
499. 1:25:18take that moment and say, "What's my intent here?" And then the prompt that you write is laced with your point of
500. 1:25:24view, the result you get back will be marketkedly better. And you could stop right there. Right? And again, I think
501. 1:25:32this is where most people think AI is, is you figure out the perfect prompt, you put it in, and you get back brilliance.
502. 1:25:41AI is not the genius. You're the genius. AI is there to amplify your ideas. So use your voice. Frame your intent. Use your voice.
503. 1:25:58All right. He uses uh stable diffusion. Sorry. CJ Fletcher uses stable diffusion, not mid journey. Similar idea. Okay.
504. 1:26:08Step three. Edit, iterate, and refine. Now this is where the magic happens in AI. Um,
505. 1:26:23I'm trying to think maybe 3% of the time when I'm using AI, I'll put in something, get an answer out, and
506. 1:26:33use that answer. Maybe 3% of the time most times. And and what's funny is when I talked about talk talk to people about using AI,
507. 1:26:47one of the questions I'll get is, well, what happens if it gives me a shitty answer? The answer is you tell it it gave you a shitty answer.
508. 1:26:57So step three is about the back and forth. So So in step two in use your voice, you're you're lacing the prompt
509. 1:27:05with with your intention, your personality, your ideas. And rather than treating it like a transaction, I'm going to put in this
510. 1:27:14prompt and get out an answer. Think of it more like starting a collaboration, starting a conversation, starting a a creative back and forth.
511. 1:27:24And step three is where you do all that back and forth. This is where the magic happens. AI's first answer is usually
512. 1:27:30crap. You shaping it is what is what gets it there. So, I want to show you something. So um this this is something I worked on with chat
513. 1:27:44GPT. This is the extended version of the fuel prompting methodology. Right? And so this is in a canvas here
514. 1:27:53and um you know I've got frame your intent and I've got you know examples and then I've got use your voice and
515. 1:28:00I've got examples and prompt prompting examples for business creative and storytelling. I didn't just say, "I want to come up
516. 1:28:10with a a four-step framework with an acronym, write it, and it generated this and I'm done." No. Let me show you something.
517. 1:28:20The back and forth to get to this simple framework that I'm I'm walking you through right now probably took me off and on about six hours.
518. 1:28:33I'd go in and I'd ask it to do something and then it would do it and then it would be sort of right and then I'd do
519. 1:28:39something else and I I think the original acronym I picked was fire and then at some point just fire didn't
520. 1:28:47feel right and I went back and I looked at well what other ones did it give me and fuel was there and I was like yeah
521. 1:28:51fuel's better and I liked something about it but to create that short little document that like one or two page
522. 1:29:00document explaining this framework was like six hours of back and forth and creative collaboration. I think most people don't know that
523. 1:29:12this is the work of AI. It's what I call chain of craft. That that when you get good with AI, when you become a
524. 1:29:19professional is when you understand that you've got it's this back and forth and it's you who are the one that's saying,
525. 1:29:27"No, that's not good enough. No, that's not good enough. No, you [ __ ] that up. I want it better. I want it better. Oh,
526. 1:29:33that's really good. Amplify that. And then it does. It's really good at responding to what you give it, but it's
527. 1:29:39not really good at doing it in an instant. When people try to give or sell you prompts, um, is that an annoyance or you
528. 1:29:49mean for me personally, I just don't Here's my philosophy on on on prompt libraries and prompt recipes.
529. 1:30:01Um, anyone can give you a prompt that will work. Any prompt will generate something in midjourney single word word
530. 1:30:11prompts can generate really amazing results like really amazing. You can use incredibly complicated prompting systems that include JSON
531. 1:30:22structuring and and things like that. I've used those before. You could also just talk to it. I I mean, one of my
532. 1:30:30favorite things to do with chat GPT is just throw it into um into voice mode and just talk to it. Just talk to it like a person. And it's
533. 1:30:40it's writing. It's transcribing your whole conversation. Um so, no, I mean those those prompt libraries don't
534. 1:30:48don't bug me. What what bugs me is how they're positioned. How they're positioned is I'm going to give you this library and it's going to
535. 1:30:57solve all your problems. No, all they did was basically give you a new thing to Google. Right now, I've got a if if
536. 1:31:05it's a comprehensive prompt library, they just use chat GPT to write all those prompts. They came up with a
537. 1:31:11framework and said, "Write a bunch of prompts for all these different use cases, threw it in a PDF, and sold it."
538. 1:31:17And so now you're you're digging through that thing thinking the answer's in there. And what I'm here to say is the
539. 1:31:24answer is not in there because the tools keep changing. That prompt library may or may not work well with GPT5. It might
540. 1:31:31have worked well with GPT3, maybe GPT4, maybe not with GPT5. We don't know. Nobody knows. What I'm trying to do with the this fuel
541. 1:31:42prompting methodology is make it tool independent and prompt independent. Um, so so pop the slide back up again if
542. 1:31:50if I'm still sharing my screen. Am I? Yeah. Um, so the third one is iterate. So that's it's the back and forth. It gives you an answer
543. 1:32:03and then at some point you're going to get it to a point that it's good enough. And and the fourth phase of the fuel prompting methodology
544. 1:32:12is launch. And again, this may or may not involve AI. It likely doesn't. Launch is like you taking the output and going
545. 1:32:25Okay. Now, where where do I want this to live? Is this going to go into a word document? Am I going to use Grammarly to
546. 1:32:32go, you know, fix up the grammar? Am I going to read through it again and and fix any little noobly things? You're
547. 1:32:40going to polish the output because this is something you're going to put in the world with your name on
548. 1:32:46it. And back to the the the consulting question earlier with subject matter experts, um those subject matter experts, even if
549. 1:32:57they're using AI, they need to be trusted. And so what they put back in front of a client, shouldn't just be
550. 1:33:04whatever gets vomited out of an a AI session. They should actually look at it, package it, put it together, refine
551. 1:33:10it, things like that. You might use AI to help you do that. Maybe add some images, things like that. But you might also just
552. 1:33:19use your brain and a pencil. That works too. So anyway, so that's the new the new methodology. So frame your intent,
553. 1:33:30use your voice when you're doing your prompting. Iterate, edit, refine it, back and forth, back and forth, back and forth. And then once
554. 1:33:43you get something usable, then polish it before you share it with the world. All right. Should I learn Python for AI? It
555. 1:33:51depends, Carl. If you want to be, it probably doesn't hurt you to know it. I don't think you necessarily need to
556. 1:34:01learn it to execute it. If you want to work behind the scenes in AI, if you want to work on the IT side and the
557. 1:34:10implementation side and the operations side, probably not a bad idea to know it. But what I would say is you damn
558. 1:34:18well better be good at vibe coding, right? using cursor or or replet agent um or depending on where you are in the
559. 1:34:28tech stack maybe lovable if you're if you're closer to the front end um you should understand software development
560. 1:34:35and you should understand the process um I think learning Python's not a bad thing but I think learning to program
561. 1:34:43with vibe coding tools and being um nimble with that I think is going to serve you better YouTube question. Do
562. 1:34:53you tend to start at 30,000 foot level and refine or break such a project into smaller chunks and work on them
563. 1:35:00individually? That's a great question. Um, I tend to personally, if I'm not using AI, I am very much a start with the three
564. 1:35:14acts and then break the act into scenes and then break the scenes into beats and then break the beats into moments.
565. 1:35:22Um, I I am very much I start at 50,000 feet and come down. Um, AI is weird. I've like one of the most successful things I did
566. 1:35:38was I took I took an incomplete outline where I where I had written most of act one in the outline, most
567. 1:35:49most of the bullets of act one in the outline and just a handful of bullets of act two and act three. This was for a screenplay.
568. 1:35:56And then I I had gotten excited and I wrote like the first six or seven scenes of act one. So I had these two documents. I had this
569. 1:36:07incomplete 30,000 ft view and then I had some scenes that that like down at tactical level, right? I put both of
570. 1:36:18those into chat GPT and into it was 40. I don't know when it was. This was probably November of last year. And I told it to
571. 1:36:31Oh, wait. This was 01. This is right when 01 came out. This was the thing, the reasoning model. I had it um
572. 1:36:41I said, "I want you to complete the outline first." So, I I uploaded both. Here's my scenes. Here's my incomplete outline.
573. 1:36:50And I said, "Finish the outline." And one of the things it recognized was that I had started the outline in a in a particular
574. 1:36:59it was called um it was from the software company called Mariner and it was oh [ __ ] was it called? It began with a C. Anyway, it was a
575. 1:37:11particular kind of outlining software and it recognized it and it finished the outline in the structure of of that
576. 1:37:18proprietary thing. It was called Contour I think. Um, so that was really impressive. So it finished the outline and then I said,
577. 1:37:28"Take the most dramatic scene that you created new in the outline and based on the script that I gave you, write write
578. 1:37:37that scene." And so it picked this pivotal turning point scene in at toward the end of act two and it added
579. 1:37:46characters and it maintained characters from the the ones I had written and it was good. It wrote like an eightpage
580. 1:37:54scene. So I kind of feel like with AI, one of the things we talk about in the AI salon is you should play first.
581. 1:38:04One of the things AI is really good at is just throw [ __ ] at it. Just throw [ __ ] at it. Just throw [ __ ] at it. See
582. 1:38:09what it does. It It will likely surprise you how good it is. Um, Tik Tok question. Lol. He's an actor.
583. 1:38:24I do have a degree in acting. Um, anyway, so that's the fuel prompting methodology, right? Frame your intent,
584. 1:38:31use your voice, iterate, iterate, iterate the edit phase, and then launch it. Give it a polish. Um, so anyway,
585. 1:38:41hope you had fun tonight. Oh, let's go look at We've got one more thing to look at. Let's go back to um, Gen Spark and see how it did.
586. 1:38:57So this was we we did this with um with chat GPT and this this is the theme AI transformation workshop a strategic
587. 1:39:08one-day workshop for subject matter experts. Okay. Um agenda is more nicely formatted. One of the things I noticed about GenSpark
588. 1:39:19is it's it's just more nicely presented. This looks better. A window for SM transformation is closing rapidly as AI
589. 1:39:28adoption accelerates. 70% of knowledge work activities potentially automatable. 30% of theme digital transformations succeed.
590. 1:39:40AI adoption has le leapt from 33% to 72%. That 72% was in the the open AI one. Generative AI can automate 60 to
591. 1:39:4970% of knowledge work. That's a good slide. It was from McKenzie. Five core insights exploring the changing theme landscape.
592. 1:39:59Wrote knowledge automation. What's at stake? So, GenSpark, you'll notice here it took the five ideas that I gave it,
593. 1:40:06it incorporated them into this, which chat GPT didn't. Test at knowledge, the last human frontier. I don't know what's supposed to be
594. 1:40:19there. Oh, I guess these are not done. Maybe it crapped out. There's 15 slides, though. What happened to those slides? Oh,
595. 1:40:27they're in here. Okay. It just It just the the display didn't show. Interactive AI, aptitude, self assessment. Where are you in curiosity
596. 1:40:40and learning? Adaptability and new tools, system thinking, comfort with ambiguity. Really good. Build your six-month
597. 1:40:48transformation template. Yeah, I would say GenSpark did did better. Now, it didn't do perfect because it didn't
598. 1:41:04it says five core insights. Insight one, insight two, and then it moves on to the aptitude test. So, it missed insights three,
599. 1:41:13four, and five. So, again, not perfect again. And and listen, this is in in that fuel framework. This is the edit
600. 1:41:21phase, right? It it gave us a really good start here. It's missing three slides, right? So, you got to go ahead
601. 1:41:28and tell it, "Hey, you [ __ ] this up. You're missing three slides." Anyway, all right. I'm getting out of here. So,
602. 1:41:35tomorrow at 400 p.m. Mountain time, 3:00 p.m. Pacific, 6:00 p.m. Um, wait, is it 400 PM or is it 3 p.m.? It's
603. 1:41:484 p.m. Yes, 4. Uh, 6 6 PM Eastern. Um, we've got the AI Readiness Project podcast with Ann Murphy and myself. Um,
604. 1:41:58we have a really good guest who's a Tik Tok AI influencer. I'm really excited about it. Um, so you can check that out
605. 1:42:05at a readiness project.com and you can see where we stream. We stream on YouTube and a bunch of places.
606. 1:42:12Um, our podcast is now also available, the AI readiness project on Apple podcast and Amazon and Spotify and
607. 1:42:22wherever you find those things. So, go check that stuff out. And, uh, yeah, I'm gonna get on out of here. Hope you had fun tonight.
608. 1:42:32Hope you people had fun tonight. [Music] Final thoughts, questions, questions?
609. 1:42:49Yeah. Yeah. Come back tomorrow. We'll see how many new versions of Chat GBT are available. I I just here here's what
610. 1:42:59here's my thought is you should keep coming back here at least for the next two weeks. I'm just going to keep
611. 1:43:05playing with chat GPT and and trying to figure out what it's good at. Like the agent experiment we did tonight. It's
612. 1:43:12kind of shitty at slides. We know that now. All right, but let's go find out. Maybe there's things it's good at. Um, it looks
613. 1:43:22tonight like it got a lot better than it was for the past three days. Why? I don't know. We're back to having eight models, though.
614. 1:43:34It's so annoying. It's so annoying. Oh, man. All right. [Music] All right. I know.
615. 1:43:50you aren't into open stuff stuff, but Olama Turbo is pretty cool. Elong Ma, it's not that I'm not into open source
616. 1:43:58stuff. It's that I don't have the bandwidth to or the hardware, frankly, to install it all and and just it it's
617. 1:44:08Yeah, I'm just at this point I'm a prepackaged guy. I'm in the prepared food section of the grocery store, not the raw ingredients
618. 1:44:19where the chef shop, but the open source stuff is getting really really amazing. All right. Um, it case it makes fake case law and would
619. 1:44:34not correct it. Yes. Yes, it will hallucinate. Now, GPT5 should be better at hallucinations than GPT4 would. Um,
620. 1:44:42but yeah, this is the thing. AI at this point is not good enough to be left to its own devices. You have to be there.
621. 1:44:49That's the whole reason I put that fuel framework together. That third section, edit and iterate, that's really
622. 1:44:56important. If you're trying to do legal cases, you're going to have to check all of what it produces because it will
623. 1:45:01[ __ ] hallucinate. And it, to your point, you you tell it to fix it, it'll go, "Oh, you're absolutely right. I did
624. 1:45:09that wrong." And then it'll do it wrong again. It's maddening. But yeah, that's that's just the state-of-the-art right
625. 1:45:15now. That's where we are. All right. Hope everyone has a good night and uh I will see you come to the AI readiness
626. 1:45:20project tomorrow at 4 PM Mountain. AI readiness project.com. And uh I will see you tomorrow night 8:00 PM. Same bat
627. 1:45:29time, same bat place, same shitty home office. Peace.