7/18/2025 - AI Agents: A New Frontier in Human-Computer Interaction

Transcript

1. 0:10Woohoo! [Music] Find a night
2. 0:24every [Music] like me. [Music]
3. 0:40One of life's little mysteries. You all right there? You're choking out. You're choking out. You gonna be all right.
4. 0:52Somebody didn't do their vocal warm-ups, did they? [Music]
5. 1:07So tonight I ask the stars above. How did I ever hear your love?
6. 1:22What did I do? What did I say [Music] to turn your rain jealous my way? [Music]
7. 1:43I'm the guy that never I learned to dance, never even got one second glass. [Music]
8. 1:57across a crowded room that was close enough. I could look, but I could never touch
9. 2:11[Music] the stars above. All right, good people. Happy Friday night. Date night. Friday night. Date night. We are Sans's producer Brandon
10. 2:25[Music]
11. 2:45[Music] in a westerly. direction. This car is my train.
12. 3:02I've been driving. I've been wondering what it is I'm running from again. Champ, you going to sing? I feel like an 80-year-old man
13. 3:18holding on to 29. Up ahead on that horizon is a California line. [Music]
14. 3:37Up ahead of trucks carrying a wide load. Free house. Good night. Cute little front door into two windows. My love
15. 3:49ain't sure whether a cry should last. You see, I broke a home on myself once. This I stumbled to that dog. I write a note by the dogs light.
16. 4:05Said, "Don't you come around here anymore." Well, I've had enough of this freedom of
17. 4:21Oh, losing Champ. Champ's walking. He's had it. He's had enough. He's like, I can't take this. Learn a new [ __ ]
18. 4:30song. learn a new [ __ ] song. It's what he said. That's what he said. I can talk dog. You know how I can talk dog? I
19. 4:38got I got access to the new dog talking AI. You just put the chip in their head, a little uh injection in their neck,
20. 4:47staple something to their ear, and now you just talk to him. And boy is he pissed. He does not like me.
21. 4:55You know, you know what's hard for dogs? They they uh they want to not do things, but because they're dogs and the Pavlovian thing,
22. 5:07they want to eat against their will. They'll come sing with us. Yeah. Anyway, I don't know. I got access
23. 5:15to that, so I'm kind of a big deal, you know. Jeff Fran Flanigan getting a shout out from Ocean. Hey, Ocean. What's
24. 5:24happening? Haven't seen you in a while. Hope you're well. [Music]
25. 6:11Oh man. Happy Friday night date night everybody. Hope you got your nachos going, your dirty microwave hot pocket.
26. 6:28[Music]
27. 7:00[Applause] [Music]
28. 7:15[Music] Oh, yes. I've already made a preemptive cheese run. Trying out mild drinking for Friday night. I like it. No hot pockets,
29. 7:30though. That's a shame. That's a shame. Jeff Flanigan realized Ocean said hello. Felt like a heel because he had stepped
30. 7:39away from the computer. Wait a minute. You all are not watching this like like like a new episode of Seinfeld in the '9s. Come on, people.
31. 7:51[Music]
32. 8:15I don't know if I'm feeling philosophical tonight or demoy or ornery or hopeful or pessimistic. I think tonight will be a reflection of
33. 8:31all of you. Well, hopefully every night is a reflection of all of you, but sometimes I have an agenda. I just
34. 8:40I'm feeling agenda free. So, ask away. Ask if you're new here. Hello. My name's Kyle Shannon. This is my
35. 8:48trusty guitar. Uh I got a dog over there named Champ. Got a family elsewhere. And then uh we talk about AI stuff here. Uh I am wholly wholly
36. 9:02100% wholly unqualified to teach anything but I do it anyway. Think of it as as as mansplaining as a service. Welcome.
37. 9:18Like chat GBT. I will confidently answer you even if it's wrong. The other thing I like to do is entertain myself. Uh, not really for
38. 9:34your entertainment, really just for my own personal pleasure. So, it's a lot of stupid jokes. It's a lot of repeated
39. 9:39jokes. A lot of jokes, a lot of things I find funny, I'll just share with you because I think they're funny. And if
40. 9:48you think they're funny or not doesn't really matter. The people that tend to stay here tend to be people that like the same stupid
41. 9:55[ __ ] I do. But if you have questions about AI, pop them into the comments. Totally unqualified. That's it. [Music]
42. 10:09Only have to be 98% right. Yeah, if I'm 98% right, I beat Chat GPT. Kyle Kyle Kyle was actually watching you live on
43. 10:16YouTube on my big TV screen since some months. I'm sorry about that. Yeah, I don't know how you explain that to
44. 10:23friends when they walk in and they're like, "What are you watching?" But hopefully you have no friends and
45. 10:30you don't have to deal with that. You can just put me up on the jumbotron. But I spent last night at a Holiday Inn.
46. 10:43It's Holiday in Express. Gota You got to get your You got to get your marketing memes right there, Brian.
47. 10:52Oh my god, that's funny. Um, all right. Jeff Jeff Flanigan, I'm 99 and 4100s% wrong. [Music]
48. 11:12[Laughter] I like it. I like it when my friends watch with me and say, "Is he qualified?" And I say, "Just wait. You'll see."
49. 11:24That actually happened last week. That's hilarious. Yeah. Like, why are you watching this guy? Hey. Hey, Winston.
50. 11:32Yeah. Listen. No, I No, I listen. I love coming over to your place. Um, the guys and I were talking. We were just uh we
51. 11:42were wondering the I I know you gave up on basic cable years ago. Uh this fell you were watching on a the
52. 11:52what was it the about the AI stuff. Yeah, I know. We know you're very technical and you enjoy the technical uh aspects. Uh uh
53. 12:02we're concerned for you for your for your health for your mental health. We're Yeah. Are are you okay? Because he it
54. 12:13I I mean Yeah. No, you're good. Okay. Or no. Hey, we just wanted to check in. We just wanted to check in. Make sure
55. 12:21because we've never seen uh television quite like that before. It was It was It was uh It was interesting.
56. 12:29All right. All right. We're going to just go then. We'll uh Yeah. No, we can't do it next week, but maybe uh
57. 12:35maybe in August. Maybe in August we'll come back and you know, you'll have moved on there. No, you won't. Okay. All
58. 12:41right. No, that's good. That's good to know. Good to know. All right. H. Hey, Champy. What's up? You want out?
59. 12:48You want out? You want your cheese? Hold, please.
60. 13:10[Music] I did a demo advanced voice at a family
61. 13:25gathering and one of my grandkids said, "That's just creepy. You should show your grandkid the the waifu from Grock.
62. 13:35They'll be like, "That's just creepy. Can I go talk to her?" Hey. Hey, Grandpa. Can I borrow your phone again?
63. 13:52Hey, Danielle. What's happening? Oh, what's going on? I was watching TV. Uh, no. I think I think you're it was the
64. 13:59wiser move was the watching TV part. I think you were good before. Uh nothing's going on yet. We haven't really started.
65. 14:05I'm trying to figure out what the hell I want to talk about tonight. Um happy to show stuff. Happy to answer questions.
66. 14:14Happy to wax on philosophic. [Music] We are definitely entering a new uh We're definitely entering a new
67. 14:32Is era the right word? No, it's not an era. This whole thing's an era. We're entering a new phase. We're entering a new phase.
68. 14:46[Music] the the uh the tools like Manis and Genspark and
69. 15:00the new OpenAI agent thing. Um, shit's about to get weird. Shit's about
70. 15:15to get weird. because these tools are um
71. 15:29they're very raw right now and they don't really work, but if you look at what they're doing and what's actually going on under the hood,
72. 15:41they're I think they're going to start to improve quite quickly. If you've been paying paying attention to Gen Spark in particular,
73. 15:48Gen Spark launches a new feature probably about on average every two weeks. Um, and they're just getting better and
74. 15:57better and better and better. If you haven't played with Gen Spark, it's really worth you going and playing with
75. 16:01it. Paradigm. Uh, I don't know if we're in a new Well, I guess we are in a new paradigm. The new paradigm is we're
76. 16:08we're in a we're in a we're entering a phase where how we interact with with
77. 16:22computers and data and the internet is not going to be like we've ever done it before. And I don't think any of us
78. 16:32really know what that means. I don't think we we know how to use these tools. I don't think we know when to use these
79. 16:38tools. Um I don't know if we know when to trust these tools. Um I think a lot of people are going to make a lot of mistakes and share [ __ ]
80. 16:49with these tools they shouldn't. Um yeah, I don't know. It's wild.
81. 17:06We're entering a paradigm where intelligence isn't the bottleneck. Yeah. Yeah. And and our relationship with
82. 17:23Yeah. I don't know. Anyway, if you don't know what the [ __ ] I'm talking about, I'm happy to go back and explain it. If
83. 17:30you do know what I'm talking about, Hume AI had an update to EV3. I I actually Oh [ __ ] You know what? I got a uh I got a
84. 17:40LinkedIn request from the CEO of Hume to go check it out and I didn't do it. Maybe that's something we can play with
85. 17:45tonight. Sorry, you probably addressed this. I wasn't there. What's the thoughts about YouTube stop paying for
86. 17:51anything with AI? Is it fair? Um, yeah, I think it's I think it's absolutely fair and I think it's I think it's so
87. 17:59there's there's some nuance. So, if you don't know, about a week ago, no, two weeks ago, YouTube said they're going to
88. 18:07demonetize any video that is basically AI slop. So, if it was just sort of generated out of a machine, um,
89. 18:20you're not going to be able to monetize it. And so what that looks like is channels that, you know, spew out
90. 18:27programmatic iterations of faceless videos like, you know, 10,000 videos a week or whatever. YouTube knows which
91. 18:35ones they are, right? What they also said though is we're not demonetizing all of AI videos. We're just demonetizing that crap. And
92. 18:47if you have handcrafted, if you have a a a chain of craft, right, where where you've thought about stuff and maybe you
93. 18:54went to midjourney and you made images there and you made this and then you made a song over here and you stuck it
94. 18:59all together and you made a video that tells a story, that's still monetizable. If you make a music video that's really
95. 19:05good, that's still monetizable. Now, clever programmatic AI assemblers will be able to replicate well-crafted AI content, but I think
96. 19:19YouTube will still know if there's, you know, a regularity to regularity to uploads, things like that. I assume you
97. 19:25can appeal it. But what it starts to do is the the YouTube decision starts to say there's actually a difference in in wellthoughtout
98. 19:41crafted AI productions and [ __ ] And I think that's I think that's great. I mean I think that's something that we
99. 19:49all need to discern. We you can look at Tik Tok right now and go, "Okay, that's AI news. That's AI influencer. That's
100. 19:59the, you know, crap crap crap crap." Oh, and then you see Kelly Bosch and it's all AI with AI music and it's [ __ ] art,
101. 20:08right? You can tell the difference. And so I think over time that's just going to be part of the part of the
102. 20:16noise of the internet is there's going to be a bunch of AI crap and then there's going to be stuff that rises
103. 20:21above it and we'll gravitate to the stuff that rises above it. The stuff that moves us, the stuff where we have a
104. 20:27relationship with the creator. No, no different than it is today. Um, this is new to me. Interesting. Yeah,
105. 20:36cur curation's important. Um Netflix said today they've got one of their movies that's live on Netflix right now
106. 20:45has AI generated scenes in it. I that shouldn't come as a shock to anyone. Hollywood studios are looking for people
107. 20:57that know how to prompt [ __ ] into existence. So, if you're getting good at prompting V3 and putting together midjourney images and
108. 21:08runway ML stuff and going to LTX Studios and making things, there's there's stuff to do there. Which, by the way, if
109. 21:15you're on LTX Studio or even if you're not, they're doing a contest through April 6th, I think. Um, where first prize is five grand and
110. 21:25a trip to Germany. [Music]
111. 21:40Is Digital Gods here? I don't know. Winston's here. I don't I haven't seen Lord Digital Gods, but I haven't been
112. 21:46looking at TikTok here for a minute. April, did I say April? August. August 6th. Uh, Daniel, Daniel, any of my irregulars
113. 22:00here? Yes, lots of irregulars here. Um, the videos that just do AI voiceovers. Yeah, those those take no talent. Yeah.
114. 22:10Now, you could argue, dude, what? Go respond. Go respond to who? To what? April fools.
115. 22:27[Music] Digital gods usually watches on YouTube. Oh, the CEO of Hume. No, I did respond to him. I said I said, "Hey, uh, thanks
116. 22:42for connecting with me. I love your [ __ ] I've loved your [ __ ] for years. I think what you're doing is great." And
117. 22:48he's like, "Cool. Go check out our shit." and then I didn't. So, so I will I will uh I did respond to
118. 22:57him, but but maybe we'll go play with it tonight. Oh, the other thing if you work for a large company or if you know people that
119. 23:08work for a large company, um I was actually accepted into the Washington Speakers Bureau. So, if you go to
120. 23:14Washington Speakers Bureau, I now have a profile page up on on their um if you're looking for speakers. So, if you want
121. 23:23someone to talk about AI, I do not bring champ. [Music]
122. 24:07I was watching network chuck. setup NB. Oh, N8N. It looks complicated but cool. Yeah, I I have not played with
123. 24:17N8N. It's one it's one of those there are there are many things in AI that um I keep meaning to go try
124. 24:32and then I look at it and I'm like I don't have the energy right now. Nad is one of those where it's like I know it's
125. 24:39it it's just like make and it's just like Zapier and I've made lots of Zapier automations and they're jankier than
126. 24:45[ __ ] and I I think NAN's a little slicker, but um it's just another [ __ ] thing to learn. It really is
127. 24:53just another [ __ ] thing to learn. And then I look at something like GenSpark and I look at something like OpenAI
128. 25:00agents and I sort of play them out for a year or two and it um we're not going to we're not going to
129. 25:10need automation tools like that. I don't think um I think we'll just be able to ask for [ __ ] Hang on. My my my light my $20
130. 25:21million production studio lighting package needs to be plugged into USBC. Um, what are we doing today, Kyle? I don't
131. 25:33know, Mr. I think we're going to go check out Hume. I'm just kind of chilling right now. It's Friday night
132. 25:40date night. We're kind of hanging out. We got some folks here. We got some folks on YouTube. We're talking about
133. 25:46how people are inviting their friends over to watch me, which I think is just a bad call on on on many fronts. Many fronts. Many fronts. Um,
134. 26:00I appreciate it. I love being up on the jumbotron. I'm I'm a CEO who doesn't mind a jumbotron shot. I think soon Chain of Craft will be all
135. 26:15AI handled as well. You know, Shane saw I'm glad you said that. I when I think about the YouTube demonetization of automated crap, right?
136. 26:30You can also automate well-crafted. You can design a chain of craft and then automate it. I mean, one of the things that I do a
137. 26:45lot is I'll go find something that I like. I'll go to Chat GPT. I'll have Chat GPT analyze the thing I like, break
138. 26:53it down into a series of prompts, and then I say, "Now, give me 20 completely different ideas inspired by that."
139. 27:03Right? different characters, different locations, different styles, whatever it might be. But I'll use it as a jumping off point
140. 27:11and then I'll do something really cool with it. That process that I do, that's this sort of it's just a way I work and it's like
141. 27:20the manual version of a way to get interesting work out of AI. I could absolutely automate that, right?
142. 27:35Kelly Bosch probably has enough knowledge of her process that she's probably got four or five major chains of craft that she repeats. Right?
143. 27:49There's one where she starts at midjourney, then she goes to um to uh Luma Labs. She does a start and an end frame. She
144. 27:59goes to Luma Labsa, right? That's one chain. She's got another chain where she's just going in using midjourney
145. 28:06video. She's got another chain, right? And she knows how to do her settings to get the prompts the way she wants them.
146. 28:16That could be automated. So then the question becomes, if you take a really interesting chain of craft and you automate it,
147. 28:29does that immediately invalidate the automated version of that? I don't think it should. It still had to be crafted. But
148. 28:40someone could take that automated chain of craft and just automate it themselves and then go spew out 10 th 10,000 of
149. 28:46those a week. I don't know. We're in weird territory. We're in weird territory, people. Ton Todd is checking
150. 28:54in. Check. You're in. Congratulations. Checking out. All right. Good day, sir. Yeah. you didn't miss anything. So, uh,
151. 29:08have a good night. Have a good Friday night if you're indeed checking out. That's awesome. But thank you. Hello. We need GPT agents.
152. 29:20Nah, I don't believe it. James Dutder, what don't you believe, James? What's the topic for tonight? Cam
153. 29:27Katkin, it's up to you. I I I basically have punted on having an agenda tonight. I don't have a producer. I'm just kind of hanging out.
154. 29:37What happened with the Tik Tok love clicking? Uh, I think we got Mr. It in here. It looks like there's some tapping going on. I
155. 29:47don't know if he's using his tappy tap machine. [Music]
156. 30:31Um, let's see. Can you talk about the recent announcements from OpenAI? Sure. Um, I made a Champion video in
157. 30:42irregulars. Okay, Dr. Jay, let's go. We'll go watch We'll go watch Dr. Jay's uh video in irregulars and then Oh, I never have moved my my
158. 30:53little thing over. Producer Brandon would have yelled at me for that. Actually, all you people on TikTok
159. 30:58should have yelled at me for that, but that's okay. You just had the backstage view for like what's it been an hour?
160. 31:04Oh, no. I started late. Half an hour. Um, I used Manis to design edit workflows apparently so I could feel so overwhelmed I
161. 31:15couldn't start to use them. Okay, archetypal architect, I think that you're on to something. Actually,
162. 31:28this is what I was saying. I don't think we understand. I don't think we understand what's in our hands right now.
163. 31:42And I won't speak for you. I don't think I understand what's in our hands right now. When I had a chatbot that you put in a
164. 31:52prompt and it gave you an answer, I could get my head around that. And when it got really good and started,
165. 31:59you know, making images and doing code better and doing all that stuff, it was good. I got it. I want a program that is
166. 32:07an emulator for this kind of video game. Bang, out it comes. Cool. Makes sense to me. Then they come along with these reasoning models,
167. 32:18which sometimes you talk to people, they're like, "It's not really reasoning." Like, "Okay, calm down there, Skippy.
168. 32:25But what it's doing is it's doing multi-step internal prompting before it answers you. So you give it kind of a goal and
169. 32:33then it goes, "The user asked me for this and I went and found this and now I think I should do this and then I should
170. 32:39do that and then I should do that." And now we've got these things that do that multi-step reasoning combined with
171. 32:49tool use. And the tool use includes things like surfing the web on our behalf and taking actions on our behalf.
172. 33:00And I think we're going to I think we're going to end up in this weird situation where someone like archetypal architect is
173. 33:08going to go, "Oh, I wonder if I could use Manis to create NADN automations. Hey, Manis, could you go make me some
174. 33:15NAND automations?" And and Manis goes, "Sure." And it goes out and it reads all the documentation and it asks, you know,
175. 33:24what do you want to do? Uh, I want to do this, that, and the other. Okay. And it goes off and it makes these automations.
176. 33:30And then he's sitting there like, uh, uh, did I really want that? Did I Now I got to go install these things and set
177. 33:39So I I think I'll just take a nap. [Laughter] I think most humans don't have really complicated
178. 33:55like hundredstep problems to solve that if it were solved in 15 minutes they would know what to do with the answer because like if it goes
179. 34:06off and does all that research and make makes all these reports what are you going to do with them? Are you going to
180. 34:13publish the report? Are you going to use the software? You going to try to sell the software? If you create a brilliant piece of
181. 34:21software and you want to sell it, well, now you need to make it bulletproof, right? That's a whole separate
182. 34:29activity than making the software in the first place. And then if you launch it live, then you've got to market it. And then you've
183. 34:36got to deal with tech support. And sure, you could use NAND to to create automated tech support, but then that's
184. 34:42another [ __ ] project to do, right? So, I just think we're entering this world where our capacity to make [ __ ] is
185. 34:51going to be so significantly beyond our capacity to imagine what we might want to make and then deal with the fact
186. 35:00that it just got made for us. I noticed that when 40 switches to web search or reasoning, it often loses
187. 35:11context. Yeah, that makes sense. I get my jollies from having problems not solving them. Exactly. I think a lot of
188. 35:17people are like that to be quite honest. All right. Um Okay. So, let me go to Ireulas. Let's go to irregulars, people, shall we?
189. 35:33Hey, irregulars. What you got going on? I'm an irregular. I'm an irregular, too. You're an irregular? Yeah, I'm an
190. 35:42irregular. Are you really? How did you get in? I just showed up. Me, too. That's so funny. I thought you had to fill out a form.
191. 35:54There's no form. You don't have to fill out a form. You just show up and you're an irregular. Who knew?
192. 36:02Apparently, they frown on being late, though. However, very strange. Very strange. It It just seems to be a Tik Tok channel that
193. 36:13doesn't always start on time and yet everyone's concerned about being late. It It's very concerning. I I I don't
194. 36:24think it's normal behavior. It's irregular. I'm starting to get it. I'm starting to get I think I just might have become an
195. 36:31irregular. Oh my goodness. Oh, this is very exciting. Very exciting. All right. Um, that little skip was skit
196. 36:46skip that little skit was brought to you by uh old SNL. Um, all right. What am I looking for here? Urgent. Spend your
197. 36:55Friday night with Manis. Credits inside and invite codes live tonight. Manis founders versus open AI agents
198. 37:03exclusively for the AI collective. Just came across this. Of course, your first stop should be the AI learning lab with Kyle Shannon. Oh,
199. 37:11that's very sweet. Um, holy [ __ ] The Okay, so Cam Ken, I'm coming to you. I'm coming back to I will circle back to um the OpenAI
200. 37:23announcement because Manis and Gen Spark are [ __ ] trolling them hard and I'm I'm here for it. So, I'm liking this a
201. 37:31lot. But yeah, there's a there's a Dr. J Champy video inspired by Steo's post. How do I Where is it? Oh, here we go. Canva. Okay. Oh, nice.
202. 37:48Doggy date night. I like it. Okay. Wait, did did this make a new thing? It did. Hang on a sec. Copy. Go back to here. Go back to here. Go
203. 37:58back to here. All right. Now I'm sharing it, right? Yes. Okay. [Music] It's doggy date night. Who will champ?
204. 38:14It's doggy date night. Okay, this is the kind of Listen, this is the kind of automated dril that YouTube will not let you monetize.
205. 38:29You can't just make a doggy date night video with cute dogs in it and expect that to be considered narrative. This is
206. 38:36not narrative, but I'm going to watch it again because it's cute. >> It's doggy date night. will choose. It's doggy date night.
207. 38:52>> Wait, you can't see. I thought you could see. Can you hear this? I think I'm sharing correctly. You can see. Yes,
208. 39:07I love doggy date night. Yeah, but you can't monetize this. You'd have to put another scene in it in
209. 39:14a in a title. We can see it and hear it. Okay, good. All right, one more time. It's doggy date night. Who will champion choose?
210. 39:28It's doggy date night. Okay, that is so stupid. It's brilliant. Okay, let me go to X X.com,
211. 39:46formerly known as Twitter. It's doggy date night. Uh, it's doggy date night. Okay, I What am I looking for? I am
212. 40:02looking for Oh, who did that? Winston. Those are beautiful. Thank you very much for that gift. Open AAI um agent.
213. 40:16Okay, let's go watch a video. So Cam Katkin. So yesterday OpenAI launched agent OpenAI agent. And I'm going to find a
214. 40:32video here in a second. Yeah, this is it. Yeah, this is a >> before the whole >> it's a fascinating video on a lot of levels.
215. 40:46All right, so we're going to play the marketing video. And mind you, this is a market. My AI comedy club vid is less entertaining.
216. 41:01Oh my god. Okay. So, what we're gonna do here Oh my god, I'm sleepy. I'm sleepy, people. It's Friday night date night. There's no
217. 41:14yawning in AI. Good lord. All right. Um All right, let's watch the video. It may be a while for the whole world to
218. 41:27evolve to a AI agent ccentric worldview. And so I think we should do what we can to meet the world where it's at.
219. 41:36>> Okay. So, so you pop into agent mode, right? So the little toolbar at the bottom, this will be available to plus
220. 41:44users in the next three days, like Monday or Tuesday. That's me, too. I've had two naps today and I'm going to bed. I can't stay awake.
221. 41:55Who is that? Is that Vicki or or Brandon that just posted the AI salon post there? All right. Okay. So, Monday we should
222. 42:10have this this new thing called agent mode. It's me. It's Vicki. Oh, cool. Yeah. Uh head head off. Take take a nappy. I I had I had um
223. 42:22I didn't have a nappy at work. I have I have napped at work before though. But I had a nap. I got home. I had to go take
224. 42:29photographs of Gabby's art show and then she wanted to hang out there and so I came home and took a nappy and then I
225. 42:36went and picked her up and then I came here. So that's why I was late. Um Oh, that's okay. I don't mind you posting as
226. 42:41a salon. That's cool. Um where's the agent link? So the agent link will just be within chat GPT under your tools menu
227. 42:52in the prompt bar. It'll just be a new mode called agent mode. Okay. >> Meet the world where it's at. [Music]
228. 43:03My name is John. I work on the deep research and agents teams at open. One great use case that comes up a lot is
229. 43:10you have some kind of budget file and whenever you do that it's kind of a pain. It takes maybe four to eight hours
230. 43:15and that's kind of your day. I'm going to show you an example where the agent >> Okay, first of all, you you we all have this scenario that
231. 43:24comes up all the time. We've got like this budget and we've got to spend four or eight hours working on that budget.
232. 43:31Really? Does that really come up for a lot of people? A lot of times, I mean, maybe. And then the example that
233. 43:40he uses, well, you'll see, is not one like it's his budget. It's like the city of San Francisco's budget
234. 43:47>> sources information on the city of San Francisco's annual budget, expenses, and revenues for the past five years, and
235. 43:54it's going to compile that all into one nicely formatted spreadsheet. It >> Okay, so one of the things this says to me is
236. 44:03that the people that are making these tools don't actually have a handle on how they're going to be used.
237. 44:13They use two examples when they did the announcement. They used two examples. At work, budgeting is annual. Not all the
238. 44:19time. Exactly. Exactly. I've been posting AI salon vids in YouTube and forgot to change accounts. Oh, that's
239. 44:26cool. You're cool, Vicki. Vicki, I'm I'm I'm cool with you posting either way. Um I thought it was kind of cool. Um,
240. 44:36the two examples that they used in the demo were book me a vacation. Okay. I I mean, yeah, that that's a use case.
241. 44:50Every single time one of these companies that does these multi-step things, you know, says, "Here's what you can do with
242. 44:57it." They're like, "You could book a vacation." you I know that use case. Could someone come up with a [ __ ]
243. 45:03interesting use case? So he's like, you do budgets all the time. And and so now I'm going to show you the San Francisco
244. 45:10budget. Okay, whatever. So So anyway, >> goes on by itself. I usually just close my laptop, go grab a coffee, maybe I
245. 45:17have lunch that expenses and revenues for the past the city of San Francisco's annual where the agent sources eight
246. 45:25hours. When whenever you you have some kind of budget file and whenever you do that it's kind of a pain. It takes maybe
247. 45:31four to eight hours and that's kind of your day. I'm going to show you an example where the agent sources information on the city of San
248. 45:38Francisco's annual budget expenses and revenues for the past 5 years and it's going to compile that all into one
249. 45:45nicely formatted spreadsheet. It goes on by itself. I usually just close my laptop, go grab a coffee, maybe I have
250. 45:51lunch. Okay, let's talk about this. This is a marketing video. The promise here, the promise here is, "Oh, once you have
251. 46:09agent, you just say, "Hey, I need a budget whipped up there, Skippy." And I and I hit the button and I close my
252. 46:16laptop and I go off and have a latte with the rude barista. Okay. Is that how anyone you know works? I guess that could be the future. We
253. 46:31We're just there to sort of initiate a thing and then it does the thing and then we're going to go relax in between.
254. 46:38I don't know. I don't quite get it. But I get it, but I don't get it. But they're basically saying it's going to
255. 46:44take 15 minutes for this thing to do all this [ __ ] So you might as well go get a coffee. >> So first it needs to find the data. So,
256. 46:50it probably does a web search to figure out where it can find this San Francisco city budget information. Once it finds the San Francisco
257. 46:58>> Okay. So, so here's where I want to stop. So, what you're looking at on screen here is this this little window. Notice that
258. 47:05they've got the same background for this window as the advanced voice little circle when you talk to advanced voice
259. 47:12and it's got that little sky thing. It's got similar branding here. So, this little window is an agent
260. 47:21within a normal chat context, right? So, you're having a chat thing and within the chat thing, you're like, I got to
261. 47:27come up with a budget thing and I'm going to do it this and a that and oh, why don't I just have this agent go off
262. 47:33and do all this research on the budget, right? And so this is if you've used GenSpark or Manis, it's essentially an identical
263. 47:42interface where what this window does is shows you the steps that the agent is taking. It's off surf surfing the web.
264. 47:51It's off researching websites. It's it's writing code. It's switching to image generation. It's switching to generate
265. 47:58um a spreadsheet. So this window is now this playable um history of all the steps that this agent took along the way but it's in the
266. 48:07context of a larger chat GPT chat that I think is actually quite significant. GenSpark and Manis both are these kind
267. 48:15of standalone operations that you go there, you kick off this agentic behavior, it does all of its [ __ ] and
268. 48:22then it gives you a bunch of stuff. Chat GPT, this is all going to be within the context of a search. So I think I think
269. 48:30more people are going to experience this than experience Manis and GenSpark because those are these specialized
270. 48:36tools that you actually have to think about how will I use this, this, and that. I think this might be uh something
271. 48:43a little um a little more intuitive for people. Um anyway, let's keep watching this. Go city government website. It
272. 48:53will try to access the PDF files. So, it has its own file system and everything. Then it needs to extract maybe 200 numbers from each
273. 49:03>> Vicki. Right. If I'm working, my manager will want me at my desk, not randomly grabbing coffee or out for lunch. Yeah.
274. 49:09They they see out at the cafe. They're like, "Hey, Vicki, what's going on? Aren't you supposed to be working on
275. 49:14that that thing?" "Oh, I not to worry. My agent's taking care of it." Okay. So, Vicki, why do we need you? Let me get
276. 49:23back to my desk. Exactly. Like, I just don't I just don't buy this utopian little vision they have here. But
277. 49:30>> PDF and finally, it will have one command that will generate the entire spreadsheet all at once. If you go back
278. 49:36to the chat, you'll see the final response. And let me just open it now. Yeah, I think it got 98% of the
279. 49:43information correct. It also formatted the Excel workbook as I instructed. >> Wait, wait, what did you say?
280. 49:53What did you say? Goodlook marketing dude >> command that will >> I don't know if he's good looking.
281. 49:59>> Generate the entire spreadsheet all at once. >> If you go back to the chat, you'll see the final response. And let me just open
282. 50:05it now. Yeah, I think it got 98% of >> Let me just open it now. Yeah, I think it got like 98% right. Wait, what?
283. 50:20Which 2% didn't it get right and you just opened your laptop and immediately noticed it got 2% wrong? How How did you know that? Well,
284. 50:36I would assume that when you get back after your double Zippo latte with the friend that you don't really like at
285. 50:43work, your colleague that you pretend to like but don't really because they're passive aggressive and they don't have a
286. 50:49point of view that's their own. I assume when you get back to to the Zippy Latte with that person, you sit
287. 50:58down at your desk, you open it up and you look at all of the [ __ ] that it did and you're like, "Oh my god, that's a
288. 51:06lot. Do I have to look at all this and then you look at the spreadsheet that it made?" How do you know those numbers are right?
289. 51:16How do you know where the source is? I assume you as a human need to dig through some [ __ ] right? So, the the four or eight hours
290. 51:30it would have taken you to make this on your own. Now, you've probably got one to two hours. If if you actually want to see if
291. 51:38it's 100% right, you actually need to go look at every number in this spreadsheet and figure out how it got to that.
292. 51:47Now, is it possible that it wrote up a whole description of how it did it and maybe it's got areas of where it's not
293. 51:53confident about those number? Maybe. I don't know. But this is where I'm what I'm saying where I don't think that we are prepared for
294. 52:06what our role is with these things. If like if I'm doing something like creative ideiation with one of these
295. 52:13tools, then it doesn't [ __ ] matter. Like it's either good ideas or not. But if it's something like a spreadsheet for a [ __ ] budget,
296. 52:21oh, it got it 98% correct. Great. Is everyone going to check to make sure that it got 100% correct? I don't know.
297. 52:32So, how many of these reports are going to be put into annual reports that someone's going to actually do the
298. 52:39math on and go, um, where did you get this number? Ah, chat GBT information correct. It also formatted
299. 52:55the Excel workbook as I instructed it to. In this case, the revisions were small, so I just made them within Excel
300. 53:02cuz it was just a copy paste. But absolutely, you can make them in chat GPT. I would say just try it out. If it
301. 53:09can do 90 95% of the actual time consuming part of the work, uh, that's going to save you a ton of time.
302. 53:16>> Yeah, that I agree with. Right. If it's going to do 90 or 95% of the work, it's going to save you a ton of time. Current
303. 53:22chat GPT maybe gets you to 80%. Maybe these new agents get you to 90 95%. But you still have to deal with the output,
304. 53:31right? You still have to use your critical thinking. Is there going to be a point in the future where you don't need to use your
305. 53:38critical thinking? Probably. And that's the point at which they probably don't need you for that job. So, make no mistake, as cool as these
306. 53:49things are, and as much as we want them to be rock solid, the minute you have a tool that can do your job as well as you do it,
307. 54:01why do we have you here? And that's something that we're all going to have to figure out because I think there's an answer to it. The
308. 54:11answer is I'm going to use these tools to take, you know, what I do for my job to the next level. I think who ends up staying in jobs are
309. 54:22the people that go, "Oh, okay. If all that shit's automated, rather than heading to the cafe with with Betsy from accounting,
310. 54:32I'm going to reinvent how we do something and deliver more value to the client or save the company money and I'm
311. 54:38going to go tell my boss about it. I I think I think that's how you use these tools to to, you know, amplify
312. 54:48yourself, your work, your livelihood. Um, now really interesting thing. So, this thing's out. It uh let me let me talk a little bit
313. 55:01more about it. Um, so for Cam Catkin and for anyone else who's interested in the way these things worked work um
314. 55:12chatbt will never be able to retire as well as I do. Exactly. Um, so what OpenAI did with with Agent is
315. 55:28they realized that they had two different tools that were both very powerful but limited. They had deep research, right? If you've used deep
316. 55:38research, you turn on the deep research button and you say, "Hey, go find all the marketing claims that my competitors
317. 55:45make." and it goes off and it researches websites and does stuff and it writes you up a report. And then you have
318. 55:52operator and what operator does is it can surf the web visually and it can go look at web pages and click on interface
319. 55:59elements and fill out forms and analyze pictures and whatever it does. Deep research can't see the pictures, can't navigate
320. 56:10um complicated interfaces. operator is slow and can't do quick analysis of lots and lots of websites,
321. 56:19right? So, they both have strengths and weaknesses and they realize they're kind of complimentary. So, what agent is is it's basically
322. 56:28a virtual machine wrapper that its primary job is to navigate which tools to use. So, it might go, oo, I'm going to go use
323. 56:39deep research to to find a bunch of [ __ ] Then I'm going to go to this site at the Gap and find the latest in men's
324. 56:46fashion and describe them. Then I'm going to write a Python tool to bring it all together into a little interface.
325. 56:54Then I'm going to get this spreadsheet that I got from my deep research and and turn that into an Excel spreadsheet.
326. 57:00Then I'm going to put that all together into a presentation with charts and graphs. So it's got access to lots and
327. 57:05lots of different tools. And its job, what it's been trained on is how to use which tool when. Right now,
328. 57:16it's which is cool. Like, it's okay. [ __ ] awesome. They're now not doing the reinforcement learning just on the core model. They're
329. 57:27now doing the reinforcement learning on the navigator, right? the the thing that is going to figure out which of all of the tools I
330. 57:36have access to am I going to use um agent also has access to APIs. So you're going to be able to do things
331. 57:44like connect it to your Google Drive, connect it to your Gmail, connect it to anywhere where you've got a connector,
332. 57:50which is a [ __ ] privacy and security nightmare that we won't talk about right now. But it's but it's it's this very
333. 57:57powerful coordination layer figuring out all of the different tools it needs to use to accomplish your goal. That's what it is. Cool.
334. 58:07If you've used GenSpark or if you've used Manis, you've used Agent. They're the same thing. And I would argue that
335. 58:16Manis and Gen Spark are way ahead. And what we've seen in the past two days since this announcement is that they know that Manis
336. 58:33Manis AI let's see um here they're bragging about millions of real world sessions.
337. 58:47Here are the lessons we learned about context engineering for AI agents. Um, let's see. Maybe it's in replies.
338. 59:11I don't see it here. Um so so when when OpenAI launched Agent um Manis did a tweet that said welcome to the club and then GenSpark also did
339. 59:26one and then GenSpark today announced a million um a million dollar contest. Let's see. Genspark.
340. 59:43[Laughter] Milliondoll Gen Spark side by side comparison show AI showdown. $100,000 on the line. Same prompt, one
341. 59:57shot, side byside results. We think we're pretty good, but we're hunting for more use cases where chat GPT might have
342. 1:00:05an edge. Help us make Gen Spark better. So, they are [ __ ] absolutely going after um going after OpenAI's agents. Now, here's
343. 1:00:20the deal. They should be scared. They should be scared. GenSpark and Manis. Most people haven't heard of them. If
344. 1:00:31you're in AI, if you watch this channel, you've absolutely heard of them. You've probably used them. Um,
345. 1:00:38hang on a sec. Get black bar going there for the people on the Tik Tok. Uh, Kyle, soon you're reaching 15,000 likes
346. 1:00:47in Tik Tok again. Very nice. 15,000 likes. Beautiful. Already done. 25,000 likes. Um, okay.
347. 1:01:05It's hard to overstate the advantage that chat GPT has. um big corporations, big companies,
348. 1:01:21um even ones that are Microsoft shops and have access to Copilot often also have ChatGpt subscriptions, enterprisewide
349. 1:01:33OpenAI chat GPT subscriptions because nobody likes C-pilot. People don't like it. Like Microsoft is having trouble selling what they sell
350. 1:01:45because the companies that they're going into are like, "Oh yeah, our people use chat GPT. We get we got them licenses.
351. 1:01:51That's what they use. They don't they don't they can't figure out how to use co-pilot." So the fact that this agent experience,
352. 1:01:59the manis and genspark-l like experience is going to live inside a chat GPT chat, there's a competitive advantage there
353. 1:02:08that's going to be really really hard for the Gen Sparks and the Manises of the world to overcome. Not to mention
354. 1:02:18the security and privacy issues. Manis is a Chinese company. Gen Spark I don't think is, but I'm not 100% sure on that.
355. 1:02:25But Manis is a Chinese company. Both of these things, as well as the Perplexity Comet browser that's, you
356. 1:02:34know, can can access all of your tabs and all of the things you've logged into. All of them are security nightmares,
357. 1:02:43right? because you're like, I want you to go look at, you know, all of my tweets on my Twitter account and and
358. 1:02:52post responses to them. It'll go do that, but you've got to give it access to your Twitter account with full write
359. 1:02:58access. And then if you want to like book the travel thing like they they demo, you've got to give it access to
360. 1:03:04your credit cards, to your travel accounts, to your right, all that stuff. And it's going to go buy that [ __ ]
361. 1:03:10that's going to ask you permission. So, so I think it's I think it is smart for them both to um sort of kick OpenAI in
362. 1:03:24the teeth while they're just getting started. Um I don't think that advantage is going to last all that long. Chat GPT
363. 1:03:30is incredibly wellunded. They're incredibly smart and they're going to figure this shout [ __ ] out incredibly fast. But right now,
364. 1:03:39if you want to use OpenAI's agent, you don't need to. Just go to GenSpark. If you've not played with GenSpark, go to
365. 1:03:48GenSpark right now and just play with it. So, um, so that's that's my take on this stuff. I think agent is a is a a strong initial
366. 1:04:01um attempt at doing what GenSpark is already doing really really well and Manis is doing really well. I think the chat GBT
367. 1:04:13version is going to really suck initially. It's going to be slow. It's going to be janky. I think it's essentially what Gen Spark says. In
368. 1:04:22fact, they did. Let me see. Um, Jen Spark prompt chat GBT agent. Let me see if I can find this.
369. 1:04:55One of the one of the Gen Spark engineers. One of the Gen Spark engineers. I can't find it right now. Is this it?
370. 1:05:12Does this guy work at GenSpark? No, I think he I think he just copied what the Gen Spark guy did. But basically, um, they took the exact
371. 1:05:23prompt that that the open Oh, wait. I'm not sharing my screen. Damn it. That's why I need a producer. [Laughter]
372. 1:05:35I can't believe I've been I've been talking about all this [ __ ] without share sharing my screen. That's okay. Um, so GenSpark um
373. 1:05:45took the exact prompts that they did in the in the OpenAI demo and they're like, "Oh, look, our thing did it way faster
374. 1:05:53and way better, right?" So, so I think that um I think that the the OpenAI one's going to suck for a while and and that's okay. But again,
375. 1:06:05this is not about where we are in the present. I look at these tools and I and I'm like this is what some version of what these do is where
376. 1:06:18everything is headed. So, so our whole idea of you need to learn prompt engineering and you need to learn this and you need to do that and
377. 1:06:28you need to understand all these different tools. We're not going to need to understand all these different tools. If you want to build
378. 1:06:36applications that leverage all this stuff, then yeah, you're going to need to know some of that [ __ ] But even
379. 1:06:41then, you're going to get assists. But just as a general enduser, we're just going to ask for [ __ ] and
380. 1:06:50these tools are going to go off and make it. So, if if one of my strong recommendations and one of my if if if
381. 1:06:57I've got weekend homework for you, even if you've tried GenSpark and Manis in the past, I would go back and really
382. 1:07:06start to explore what are all the different ways I personally might use something like Manis or Genpark
383. 1:07:14to do [ __ ] for me. Like, think about things you might use it for. start to understand how you would use
384. 1:07:22that differently than how you use chat GPT. All right. I think it's a big big big big deal. I you know like this is the
385. 1:07:32the new phase that I think we're entering is is going to be centered around these things. Genpark uses GPT
386. 1:07:39models via API. So OpenAI gets paid when GenSpark does it. That's funny. We're at 20,000 likes already. Good.
387. 1:07:48Fantastic. Only a matter of time before model makers eat any company. Uh but needs the model to run. Yeah.
388. 1:07:56Listen, I think I think Genpark and Manis are going to be my instinct is they're going to be successful companies. Um I don't see them,
389. 1:08:09you know, I see them being becoming more niche successes than really broad successes, but we'll see. Or they get acquired, right?
390. 1:08:18like Gen Spark's doing what they do so good. It wouldn't surprise me if OpenAI goes, "Yeah, here's a billion dollars.
391. 1:08:24Bring all that IP over here. We'll take it." Um Sam told us there'd be days like this, we have you here to check the missing
392. 1:08:382%. Yeah, exactly. Um, all right. Questions that make that
393. 1:08:54all make good sense. I know it was a lot. I rambled for a long time and I thought I was sharing my screen when I wasn't.
394. 1:09:04I didn't have producer Brandon yelling at me. Is it possible to make sub projects in each projects folder in chat GPT?
395. 1:09:13Yeah. Well, you you mean a separate a separate agent? So So the way it's going to work is this. For 20 bucks a
396. 1:09:22month, you get 20 agents per month. So So basically like an agent a day, right? Think of it like that. Or you know on
397. 1:09:34the weekend you do, you know, seven or eight of them. How I can imagine it happening is you're working in chat GBT on something
398. 1:09:44and then you say, "Oh, I've got this more complicated thing. I've got I I need an output as part of this project I'm working on."
399. 1:09:52And that would be a perfect thing for me to to generate an agent to go do that task, right? and you're like, "Okay, go
400. 1:09:59find, I don't know, the latest public art installations in Denver, how much the commissions were, who who issued the
401. 1:10:08commission, and how long it took from um application until granting of the thing, and then turn that into a report for me.
402. 1:10:17So, I'm putting together a thing for some art installation I want to do in the city of Denver. And one component I
403. 1:10:23want is a little report to know what's the lay of the land. So I could see working within chat GPT having that
404. 1:10:31realization and go let me spin up an agent. So you spin up the agent that little window is now in the context of your chat
405. 1:10:41and you keep chatting. That agent is off doing its thing. It may take 15 minutes, right? And and you're still chatting doing
406. 1:10:50other things. So I think within a single chat or within a group of chats like in a project, you could have multiple
407. 1:10:55agents off doing things. So yes is the answer. Again, I haven't used it, but that's just based on what I'm seeing and
408. 1:11:04knowing how they've done other things. I think that's how it's going to work. Workflow hack from Josh Groves on on X.
409. 1:11:17Drop any deep research report into Gem Spark AI slides. Yeah. Beautiful, perfect, coherent slides in less than 10 minutes. Yeah. Um,
410. 1:11:28GenSpark slides are really good. They're really good. They're welld designed. They're well put together. I think that
411. 1:11:36hack is a good one where you have deep research, go do the deep research to make sure you got the research part of
412. 1:11:42it right. Because what I found is if you just ask GenSpark to go make you a slide deck on whatever topic, it'll just make
413. 1:11:51[ __ ] up or it'll like it's I think it's better to have some sort of coherent vision for what you want the deck to be
414. 1:11:59and then yeah, let it go do that. That's a that's a really good hack. Is that agent function what's coming on Monday? Yes.
415. 1:12:07Yeah. And in fact, let me go just check right now and just see on the off chance I have it. I doubt I do. New chat. No, I don't have it yet.
416. 1:12:26I wonder if I have it over on Teams. Uh, search connected apps tools, deep research connectors. Nope, I don't have it. And I was gonna I was
417. 1:12:41gonna like reup for a pro GPT account for a month, but I'm like, ah, [ __ ] it. I don't care. Like, it's I've got access to Manis. I've got
418. 1:12:54access to GenSpark, and I don't really use them. Again, I'm I'm still in this place where I don't really know what to use these things for.
419. 1:13:10I'll figure it out. We'll figure it out. I think it's important to understand what they can do. I think when we figure
420. 1:13:21out what these things can do and get good at them, I think we are going to occur to other people who are not using them like wizards,
421. 1:13:35which I think is cool. I think we should all be like wizards. I don't have it yet either, Kyle. I need
422. 1:13:42a cartoon avatar persona. All right. So, Mr. The way you do that is go to um any of the image generating tools, chatbt midjourney ideoggram,
423. 1:13:58generate a cartoon character, and then um put your script together in chat GPT. Use 11 Labs to pick a voice to to to
424. 1:14:12make an audio track of that thing talking. And then go to Hedra Hed and upload the audio track of your script and your character.
425. 1:14:26And you have it. I've done it in here many times. Video avatar to promote my product. Yep. Make an image,
426. 1:14:38write a script, pick a voice in 11 Labs, have it read that script, upload the the audio file and the image
427. 1:14:48file into Hedra and say, "Make me a video." Done. Done. And just like that, Kyle caused everyone
428. 1:15:03to yawn. What did I do? Oh, because I yawned. But seriously, Cam Catkin, go play with Gen Spark. If you want to if you want to
429. 1:15:18see what's coming with agent, go play with Gen Spark. I think with Gen Spark, you get like one agent a day for free.
430. 1:15:33Let's go do how we doing on time. We'll do we'll do one one quick little thing here. Let's do the the Josh Grove um Groves little hack.
431. 1:15:46So, I'm going to go to chatgpt. Oops, I got that wrong. Let me change this. All right. So, I'm going to go grab deep research.
432. 1:16:01I'm going to say um find me the
433. 1:16:17five most unexpected unexpectedly awesome
434. 1:16:34sites to see in Denver and five remarkable um
435. 1:16:50things to do. All right, that's kind of shitty, but it should happen pretty quick. And so now, because we're doing deep research, this
436. 1:17:01thing's going to kick off to tailor this. Do you? Yeah. Quirky and offbeat. Quirky and offbeat. Yes. And no. Meow Wolf.
437. 1:17:16Been there. Okay. Are you interested in art? Um, art history. Um, you pick topic.
438. 1:17:31Will you be traveling solo? We'll say this is with family. All right, there we go. Okay, so now
439. 1:17:46we've given a little context. How's the agent with Excel? Can it perform complex tasks? Yes. Yes. So,
440. 1:18:01so agent. All right. So, this is off starting research. And now, if you haven't seen deep research, see where it
441. 1:18:07says starting research here? I should be able to see what it's doing once once it gets started here. I should be able to
442. 1:18:14open a window and we can watch what it's doing. So, what we're going to do here, let me give you a little transparency
443. 1:18:20into where my crazy straw of a brain is. We're going to have this go research cool things to do in Denver. I'm going
444. 1:18:26to then take that output. We're going to go over to GenSpark and say, turn this into a slide presentation. All right.
445. 1:18:34Um, okay. Looking at sources. How do I You used to be able to go look at this. Why can I not look at it?
446. 1:18:52Oh, there we go. You had to click on looking for sources. So, all right. I'm piecing together a list of things.
447. 1:18:59I looked at at atlasobscura.com. Inspecting unique places, 80 nostalgia store, mortuary t um turned restaurant.
448. 1:19:12That's That's old. I've been there. Okay. Um, do you come on live every day at this time? Every weekday? Yeah, weekdays
449. 1:19:21at 8:00 p.m. Mountain time. And sometimes I'm a little late because life, you know, life happens. Um, but
450. 1:19:27yeah, I'll be back here Monday. Monday at 8 Mountain. Be careful. 100,000 likes are coming. Um, but Sam, what you should also do is
451. 1:19:39if you have not, you should go to um that URL, the salon.ai, and click on join our community. So, this is a community of AI
452. 1:19:51creators and entrepreneurs and tinkerers, people that are trying to figure this [ __ ] out. A lot of people
453. 1:19:56that are at at the AI salon come here as well. Um we call them the irregulars that kind of hang out here on a regular
454. 1:20:02basis. So, so go check out the salon um and uh and come hang out here with us. Okay. So, our deep research is doing its thing,
455. 1:20:22which is cool. All right. Questions, thoughts, thoughts, thoughts, thoughts, thoughts, thoughts. What are you making?
456. 1:20:31So, what I'm making right now is a bit of the theme tonight. So, one of the nicknames for the channel is Chad
457. 1:20:39Add. So, the chances of me ever completing a sentence in an evening are kind of low. Um, the chances of me
458. 1:20:47staying on topic are highly unlikely. Um, but the chance that I'll stumble into something interesting, pretty good. Um, the
459. 1:20:58what we've been talking about tonight is OpenAI's agent. And if you want to get a essentially a fancier precursor
460. 1:21:07or or a fancier example of what agent is and is going to be, go use GenSpark or Manis. Manis.im or Genspark I think.ai.
461. 1:21:19Um so what we're doing right now is um Josh Groves on X said, "Oh, here's an idea. take it take a deep research
462. 1:21:27output and put it into Manis that generates slides and have it generate you a slide deck. So what I'm doing is
463. 1:21:34I'm doing a quick deep research project on Chad GPT. I'm going to take the output of that and then upload that into
464. 1:21:40GenSpark and turn it into a deck and we'll see it. I've done it but literally don't even prompt GenSpark. Just
465. 1:21:47dropping the research and it'll do its thing. Oh, that's cool. Awesome. Okay, I'll try that. We'll see how it does.
466. 1:21:54Sea slug of doom is late. Unbelievable. It's Friday night date night. You You can't be late. What? Tardiness is not an
467. 1:22:04option. This is AI we're talking about. Things move faster than that. Good lord. What next? You tell me you went out and
468. 1:22:14had a lovely dinner. What about an overcooked hot pocket in a dirty microwave? That's the tradition here. Oh my god. I feel like no one
469. 1:22:26understands the rules. No one understands the rules. They just they just show up. They show up whenever they
470. 1:22:31want. Willy-nilly Mr. K all day. Deep research chat GBT3 wrote me 60-page proposal document to make myself That's
471. 1:22:40cool. To make yourself what? Um a job I want. Oh, that's very cool. That is so cool. essentially to be a
472. 1:22:50traveling tech coach for districts that don't have one. You would be so good at that, dude. Hey, Mr. K.
473. 1:22:59I I have I since you're here, I I can I beg you for something? What are you doing next Wednesday? We were supposed
474. 1:23:06to record You and Ann Murphy and I were supposed to record something last Sunday or the Sunday before and we missed it.
475. 1:23:13It didn't happen. What are you doing Wednesday at 4? You want to do something live with me anyway? Ping me
476. 1:23:22if you can do it. 4 PM my time. Um, next Wednesday. If you can do it, it would be great. You'll save my butt. It's It's
477. 1:23:30her birthday anniversary. It's her anniversary birth. I would love to. Okay, cool. So, I will I will uh I'll
478. 1:23:36ping you on LinkedIn or on the on the AI salon, Mr. K. That'd be great. Cool. Perfect. See slug of doo. By the way, Kyle, when
479. 1:23:45can we expect to see you on Jumbotron with your HR manager and the salon logo clearly visible? Listen, I said this at
480. 1:23:52the beginning of the show. I am I I am a CEO is proud to be on a jumbotron. Put me on a jumbotron anytime.
481. 1:24:01Oh my god. That one that one is just like what what was what was the best thing about that was you know you got
482. 1:24:10the contrite letter. I've made a mistake. There was a lapse in judgment. D, but he was such a [ __ ] CEO. He had
483. 1:24:17such the arrogance of a CEO. He goes, but I am very disappointed that, you know, this was done without my
484. 1:24:25permission. Like, he had to be a douche. He just had to be a douche. Like, on top of being a douche. He had to be a douche.
485. 1:24:38Okay. Uh, quirky and offbeat family stuff. Here we go. Um, can I download this? Let's see. Sources, share,
486. 1:24:52download PDF. Okay, beautiful. Look, OpenAI actually has like decent decent exporting functions now. Who knew? Okay.
487. 1:25:11The International Church of Cannabis. Immersive Art Venue. Little Man Ice Cream. Been there. 5280s Retro Toy and
488. 1:25:20Pop Culture Shop. Buckhorn Exchange. Remarkable family-friendly activities. Candy Factory Tour. This is cool.
489. 1:25:29Here's a bunch of [ __ ] All right. So, we got a PDF. So now what we're going to do is we're going to go to Gen Spark, Gen Spark.
490. 1:25:47And so when you go to Gen Spark, you can all see this, right? Hang on. Let me get this thing. That's not part of the interface. Let me make
491. 1:25:57this bigger. Plus+. Wait, I'm not there. Um, sign in.
492. 1:26:19[Music] Now look at all these things that Gen Spark has. AI slides, AI sheets, AI docs, AI pods as in podcasts, AI chat,
493. 1:26:36AI image, AI video, deep research, call for me. You can have you can have Gen Spark make phone calls to all your favorite
494. 1:26:48restaurants and ask them if they've got Prince Albert in a can or whatever and it will do it. Uh it's got a
495. 1:26:57download thing. It's it's got all agents. So what we're going to do is we're going to do AI slides. We're going to click on AI slides.
496. 1:27:06And then let's see. I click on the little browse local files. There's our quirky and offbeat Denver things. Open.
497. 1:27:22And I'm say like uh giving a presentation in five minutes. Make it pretty and make it snappy
498. 1:27:39and It should look like Denver. No. No. Let's just say make it pretty and make it snappy. Let's see. Let's see
499. 1:27:52if it can figure out what the [ __ ] it should look like. To Josh Groves's point. Chef Kelly. Chef Kelly loving the hot pockets.
500. 1:28:05Chef Kelly. Now, actually, this is really interesting. As a chef, you could probably take a hot pocket, preheat the
501. 1:28:13oven, put it in an air fryer or the oven, but you just got to admit that it's just not the same experience as
502. 1:28:20microwaving it in a dirty microwave in the little sleeve, right? The crisping sleeve. You know what I'm saying?
503. 1:28:28I think even as a chef, you've got to acknowledge that there's there's ways to do this. [Laughter] Okay. So, um,
504. 1:28:42so GenSpark. Now, what did it do here? Time to build this. So, they created a cover slide using a Denver skyline. Cool.
505. 1:28:54Nothing like the toxic sleeve. Exactly. It just gives it a different flavor. Right. If you've got it in the chemical
506. 1:29:01sleeve in the dirty microwave, right, that's got the leeching radio fields. There's just something about that. You
507. 1:29:10can't replicate that in the oven. If you were to cut it up in chunks and, you know, sauté it, it's just not going
508. 1:29:18to be the same. It's not a hot pocket. It's some sort of pastry delight. Sure, it's not a hot pocket. All right.
509. 1:29:32Some sort of pastry delight. Chef Kelly, I'm gonna need you to whip up a pastry delight for me. Which, by the way, Chef Kelly, since
510. 1:29:43you're here, I should probably show the good people. She put together a um an avatar of herself, like a 3D character
511. 1:29:52avatar talking about food additives, dyes in foods, and it's really good. Kudos on putting that together. I
512. 1:30:00thought it was really well done. Jen Spark's decks are mediocre, but Gamma is good. Yeah, well, Gamma is more specific to design. Um,
513. 1:30:12yeah, Gamma Gamma is quite a good product as well. Um, do we have previews yet? Oh, we do. Look, there's Denver in the background.
514. 1:30:25There's our first slide. Quirky and offbeat family adventures in Denver. Beyond the typical spots. Why go off the beaten path? Very cool.
515. 1:30:38Nice. There you go. Love it. This is looking great. You can watch it write the code.
516. 1:30:59We got placeholders for all that [ __ ] So anyway,
517. 1:31:18look over here. It's absolutely brilliant. This this the section divider is fire. The Colorado family adventure
518. 1:31:26backdrop. So, it's going and searching for actual photos of these locations, selecting the photos to put in there.
519. 1:31:39That's pretty [ __ ] slick. Like, this is [ __ ] you would do, right? How specific was your prompt, Kyle?
520. 1:31:46Sorry I'm late to the party. It was not. It was Well, okay. So, what I did, Corey, was I went to Deep Research and I said, "I
521. 1:31:55want five quirky sites and five offbeat locations or some whatever I don't know, activities in Denver." And so, Deep
522. 1:32:04Research, I did that in Chat GPT. Deep Research went off and just wrote me a report of here's some interesting places. I took that PDF,
523. 1:32:16uploaded it to GenSpark, and I said, "Giving a presentation in five minutes. Make it pretty and make it snappy. That's it."
524. 1:32:27And then uploaded the the the PDF. And so it's now going and finding all this stuff. It's writing the code
525. 1:32:39to generate these things. Are we done? View and export. We're done. Play slides. So, this is essentially Jim Ross did this.
526. 1:32:54Jim Jim Ross was was asked to jump in um to a keynote tweet was supposed to happen, the guy dropped out that was supposed to
527. 1:33:10do it and they reached out to Jim Ross and said, "Hey, can you do this keynote for us?" And he said, "Sure." And he did
528. 1:33:16what I just did here. He did in front of the room. He said, 'I want to do, you know, a presentation to this audience.
529. 1:33:26You know, I I assume he had a PDF there based on this PDF I've got. And it generated a a keynote that he then gave.
530. 1:33:36So, in front of the audience, he had one of these tools generate the slides for the keynote presentation that he then
531. 1:33:44just delivered. I thought that was [ __ ] genius. And then of course at the end of the presentation, what did
532. 1:33:50people want to talk about? Nothing about what what his topic was. They were like, "How did you do that?" Anyway, beyond
533. 1:33:57the typical tourist spots, discover hidden hidden gems and memorable experiences the whole family will love.
534. 1:34:03Um, why go off the beaten path? Unless unleash your family's curiosity. Denver's packed with surprises outside the usual hotspots.
535. 1:34:13International Church of Cannabis offbeat attraction. The Money Museum. Oh, I've been to that. That's cool. Um, Dinosaur Ridge.
536. 1:34:22Um, the Museum of Transportation. The iconic Big Boy locomotive at the Forny Museum. Look at this. Right. It
537. 1:34:32went and found the picture. It gave it a caption. Home to over 600 transportation artifacts. Unconventional family activities.
538. 1:34:46Rhino Art District. That's where I work is in Rhino. I know that that mural actually. Plan your own offbeat adventure. There you go.
539. 1:35:01Boom. Done. Crazy crazy crazy. So,
540. 1:35:16so this is the kind of stuff this the kind of stuff where um what do you do with capability like that? I don't know
541. 1:35:34where are the B you know like a lot of what this channel for a lot of the way I think about this channel is that my job
542. 1:35:45or the job that I've assigned myself I don't know that this is my job the job I've assigned myself in this channel is
543. 1:35:52find stuff I think is interesting go play with it and figure out where the boundaries are, right? Figure out where the edges are of
544. 1:36:05what works, what doesn't, what's janky, what's not janky, so that as the tools evolve and something new comes along, we can
545. 1:36:17look at that and immediately know, oh, that's better. That's good. Right? because we know what it was before.
546. 1:36:27All of these tools like GenSpark and Manis and and OpenAI agent are really janky right now. There's lots
547. 1:36:35to there are lots of edges to discover where they're not good at [ __ ] But I have a feeling they're going to
548. 1:36:42evolve very very quickly. And I think exploring different ways to use these automated multi-step tools
549. 1:36:53that are not the same [ __ ] everyone else has been doing, I it feels to me is really really important. So there's your um homework for the
550. 1:37:07weekend. go to Manis, go to GenSpark, play with them over the weekend, and that way when OpenAI agent shows up next
551. 1:37:16week, it should show up Monday, but it might be, you know, next week sometime, have a good sense of what it can do and
552. 1:37:25and compare it to what Gen Spark did, what Manis did, and start to get your head around these things are different. What can I do with
553. 1:37:35them? What can I do with them? What can I do with them? And my request would be if you have good ideas and breakthroughs
554. 1:37:40like Josh did tonight, um, come share your ideas. Um, super sweet. You can also fact check it, check it after it
555. 1:37:48finishes and edit the slides. Cool. Yeah. Like one thing I would do with that is I would tell it I want it to
556. 1:37:54make a slide per destination. It kind of consolidated some of them in the here's some [ __ ] to do with your family. Like
557. 1:38:00break those all out as separate things. But yeah, that's that's a really good point. Yeah, none of this stuff that it
558. 1:38:06makes is final form. You can say, "I don't like the design. Go make it look more hip and colorful," and it'll go do that, right? So,
559. 1:38:15absolutely crazy. All right. Um, so let me Oh, you know what? Let me see if I can go find Chef Kelly's thing and then I'll share that.
560. 1:38:32Chef Kelly since she's here. Kelly Anderson, right?
561. 1:38:50Videos introducing my alter ego. Okay, hang on a sec. Let me change my sharing. Share screen. Okay. Um,
562. 1:39:09so if you haven't seen this, um, Chef Kelly's an irregular. She comes to these things regularly. Get it? and um
563. 1:39:25has basically taken her brand, her ideas, learned enough of this AI [ __ ] and this is her first kind of alter ego video, and I got to say,
564. 1:39:41for for a first video out of the gate, like this is pretty [ __ ] remarkable. It's really [ __ ] good. So, um so I
565. 1:39:47won't play the whole thing because it's four minutes, but go check it out. She's uh Kelly Anderson on LinkedIn and the
566. 1:39:54the name of this post is introducing my alter ego. All right. >> Can you believe that red dye number three has been linked to cancer since
567. 1:40:05the 1980s? Somehow it's remained in your food, your drinks, your medications for over 40 years. Why? Because the FDA, the
568. 1:40:16regulatory government system that's supposed to protect us, was never really designed to do so in the first place.
569. 1:40:23This is the story of how easy it is for dangerous additives to get FDA approval, stay legal, and end up in your kitchen.
570. 1:40:32>> Like good editing. I assume, Kelly, that you've done video editing before because the the editing's good, the B-roll is
571. 1:40:38good, the use of the music is good. It's just really well done. >> Yes. Source Camp has shown this to several people. The like
572. 1:40:49one of the things that I like most about this is this kind of output for me starts to become an incredibly strong counterargument
573. 1:41:04to all of the [ __ ] Like I got a lot of hate on LinkedIn when I when I put out um a LinkedIn post about chain of craft. Meaning if you
574. 1:41:17if you have a creative process where you know you know you know you have a story you want to tell and you put together you
575. 1:41:27know a series of activities that involve generating images and music and this and that and then you stitch it all
576. 1:41:34together. you can end up with something good. And almost all of the the feedback that came back in to that was, you know,
577. 1:41:42you're a loser for daring to say that that AI could be in any way creative. And what those people are are are
578. 1:41:50missing the boat on it was the whole point of the article. They're missing the boat that the ability to push a button and
579. 1:41:59generate an image is not 100% of the process with AI, right? It's never that. It's like Kelly had this story she
580. 1:42:09wanted to tell. She she had to create the character. She had to think about the script. She had to create the
581. 1:42:14script. She had to create the B-roll. She had to create, you know, the voice. She had to edit it all together.
582. 1:42:22>> Dies began. And if if someone were to watch this, if what Chef Kelly never said was, "I used AI to make this." There's nothing
583. 1:42:35in this that I can see that immediately goes, "This is AI." And quite frankly, I don't think it's anyone's [ __ ]
584. 1:42:43business what tool she used to make this, right? Like, this is just a good video. cold tar derivatives literally
585. 1:42:51the byproduct of burning coal in the early 20th century hundreds were used in every >> right like there's a good example like
586. 1:43:00>> literally the byproduct of burning coal >> like is that is that stock footage should did she just have one of the
587. 1:43:09video tools generate that I don't know does it matter no right the the the chef is talking about burning coal. We see we
588. 1:43:20see that it brings that idea to life. >> Hundreds. >> And now we move on to the next part of the story. That's just good storytelling.
589. 1:43:30It's irrelevant what tool she used or if she got that from a clip library or if she generated it. It's irrelevant whether she generated
590. 1:43:40this character with AI and did the voice with AI or hired a 3D animation studio out of Hollywood and spent $50,000 creating a rigged model.
591. 1:43:53Do we care? No. Like it's compelling. Like just good storytelling. I got a good character here. There's clear branding
592. 1:44:03in the kitchen. Like I know what the I know what everything is about here. So this is a really good example if if you
593. 1:44:13want to share with people um the future of AI. The future of AI is we stop talking about AI, right? We stop talking about AI. We just
594. 1:44:25talk about, hey, I saw this cool video on on on dyes. Did you know that that they've known these things cause cancer
595. 1:44:32since the 80s and we're still sticking them in our bodies? That's what to talk about. How did you use a Hey, Kelly, what tool
596. 1:44:43did you use for that? Oh. Oh, you use Mid Journey. Yeah. Yeah. No, I've used Mid Journey. Yeah. No, it's I you know,
597. 1:44:48I think I I tried it once. I didn't like it. That shouldn't be the [ __ ] conversation. Period. End of story.
598. 1:45:01It's not meltdown Mondays, but I'm taking this opportunity I'll put on my pink [ __ ] bow. I'll do it.
599. 1:45:16All right. Well, literally over half the posts on LinkedIn today were created by AI. People just said don't say they used it.
600. 1:45:29Yeah, exactly. They [ __ ] hypocrites. By the way, I think this is amazing. I absolutely agree. If a company like
601. 1:45:38Pixar did the exact same thing, people would think it was amazing. Well, I but the my whole point about Kelly's thing
602. 1:45:44is I think that thing's just really good. I think it's amazing as is independent of how she made it.
603. 1:45:51And that's that's kind of that's kind of my point of where we need to be headed with all this stuff. Like even with the Manis and and
604. 1:46:01GenSpark and and AI agent thing, like what are we using them to do? If what we can do is we can use those in a way that
605. 1:46:09it outputs something that you can put in the world and people go, "Oh, wow. That's really good. Who cares how you got there? You're the
606. 1:46:18one that manifested it. You're the one that learned the tools well enough to generate a something that they can then look at and just enjoy.
607. 1:46:29So anyway, yeah, there you go. If this came out two years ago, people would have thought I was an animator. Exactly.
608. 1:46:38They still may. Do you know, Kelly, that I would I would argue that probably more than half of the people that see that
609. 1:46:46video have no clue, zero clue, that how you made that is even possible. that they look at that and you are
610. 1:46:59effectively you've hired someone that used to work at Pixar. We're we're in a very weird place in history. It's also I
611. 1:47:12personally feel this is a very magical place in history. It's why I come go live five nights a week because we're never going to be in this
612. 1:47:22moment in history again. Do you know that three years from now, Kelly, or maybe it's probably three three years from now,
613. 1:47:33AI will have saturated the world like the the internet took, you know, whatever it was, probably 15 years, 20 years maybe, to saturate
614. 1:47:45the world, right? And now everyone knows what the internet is. They know what the worldwide web is. They know how to
615. 1:47:50communicate digitally. They know what social media is. They know how to connect data sources and share and right
616. 1:47:57that's all it's all just fabric of communication now three or four years from now people will know that if you want something in the
617. 1:48:06world you literally just ask for it and out it comes but we're in this weird gap time where some small percentage I still think it's
618. 1:48:17less than 10% some small percentage of people actually know what's going on with AI and 90% of the people don't have a [ __ ] clue
619. 1:48:30and they look at what we're doing and they're like, "Uh, how did they do that?" Wait, Chef Kelly, I thought she
620. 1:48:36was a Is she like super loaded? Is she Did she hire an animator? She made that herself. How's she find time to cook?
621. 1:48:46You know, right? Like, we're in this weird time. So, anyway, congrats on that. All right, the creator is here. Well done. I know.
622. 1:48:56Exactly, right? Chef Kelly is here. So, yeah, she just made that thing. Really beautiful. All right, I'm gonna get out of here. Um
623. 1:49:04trying to think what's going on next week. Not a ton. Um the week of July 28th. So, so July going into August, I'm doing a fiveday
624. 1:49:19um AI crash course. It's going to be five, it's going to be like AI overview, AI in business, AI and creativity
625. 1:49:28something something and then like a Q&A day, a fiveday crash course. Not next week, but the following week. I have
626. 1:49:35some work to do. Um but I want you to to spread the word about it and I'm I'm going to start promoting it next week.
627. 1:49:42Um, but anyway, I'll be back here Monday. What I want you to work on over the weekend is if you can do something
628. 1:49:48like Chef Kelly did, whip that up over the weekend. That would be great. Um, but I want you to use GenSpark and Manis
629. 1:49:55if you haven't used them before, especially, but even if you have, go back and and start thinking about
630. 1:50:00creative ways that you can use a tool like that that that just does something in a completely unexpected way. That
631. 1:50:09that to me feels important. and and uh it'll it'll set you up well for when agent shows up in chat GPT. All right.
632. 1:50:18All right. That's it, people. I'm leaving. Have a fantastic weekend. Hope you had fun tonight and didn't burn burn
633. 1:50:25your tongue on your hot pocket. All right. Peace out.