1
00:00:00,020 --> 00:00:04,980
Josh:
Over the last few weeks, one of the hottest new topics that exists in the world of AI has been agents.

2
00:00:05,200 --> 00:00:08,080
Josh:
Agents that can actually get into your computer and do things for you.

3
00:00:08,160 --> 00:00:11,620
Josh:
We've seen this with Claude Cowork, which can actually access files and make

4
00:00:11,620 --> 00:00:12,520
Josh:
changes on your computer.

5
00:00:12,720 --> 00:00:16,400
Josh:
And then to the fullest extent, we just recorded an episode on Claude Bot,

6
00:00:16,580 --> 00:00:20,380
Josh:
which allows an actual computer to fully take over your life,

7
00:00:20,500 --> 00:00:23,640
Josh:
send messages on your behalf, take care of emails, book reservations.

8
00:00:24,180 --> 00:00:28,600
Josh:
The problem with that is that even an afternoon of use can cost hundreds of dollars.

9
00:00:28,780 --> 00:00:31,460
Josh:
So what we've done today is we've actually figured out a

10
00:00:31,460 --> 00:00:34,460
Josh:
way to replace that totally for free where

11
00:00:34,460 --> 00:00:37,680
Josh:
you get the same quality outputs but with none of the cost and

12
00:00:37,680 --> 00:00:41,020
Josh:
to do that i'm going to start on anthropics website because the

13
00:00:41,020 --> 00:00:44,000
Josh:
reality is that the screen that you're seeing right now isn't actually anthropics

14
00:00:44,000 --> 00:00:47,900
Josh:
website in fact it was built using this new tool completely for free in about

15
00:00:47,900 --> 00:00:51,280
Josh:
25 minutes which i thought was such an amazing demo it was built using a tool

16
00:00:51,280 --> 00:00:55,320
Josh:
called kimmy k 2.5 which is the newest model coming out of china that is fully

17
00:00:55,320 --> 00:01:00,200
Josh:
open source fully open weight and in order to build this all i had to do was

18
00:01:00,200 --> 00:01:02,540
Josh:
feed it a video so you'll see on the screen here i

19
00:01:03,010 --> 00:01:08,190
Josh:
generated a video on my desktop using a screen recorder that copied the Anthropic

20
00:01:08,190 --> 00:01:11,230
Josh:
website, I said, hey, just take the screen recording of the website and create

21
00:01:11,230 --> 00:01:12,890
Josh:
an exact replica of the website for me.

22
00:01:13,270 --> 00:01:16,950
Josh:
25 minutes later, without any additional prompts, it listed all the things it did.

23
00:01:17,090 --> 00:01:20,350
Josh:
It went through all the design and then actually published a full preview of

24
00:01:20,350 --> 00:01:22,570
Josh:
the website that we can see here on this read record.

25
00:01:22,710 --> 00:01:25,530
Josh:
So this model is incredible. I don't know if you've had a chance to play around

26
00:01:25,530 --> 00:01:29,810
Josh:
with it or check it out, but this is a huge change in the world of agents because

27
00:01:29,810 --> 00:01:32,150
Josh:
of how capable it is for such a low cost.

28
00:01:32,270 --> 00:01:36,470
Ejaaz:
This model is trending pretty heavily online right now.

29
00:01:36,850 --> 00:01:41,130
Ejaaz:
I actually saw someone describe Moonshot Labs, the creator of this model,

30
00:01:41,550 --> 00:01:43,290
Ejaaz:
as the anthropic of China.

31
00:01:43,610 --> 00:01:49,490
Ejaaz:
This was a quiet release, Josh. So the creators of this model kind of just updated

32
00:01:49,490 --> 00:01:54,850
Ejaaz:
their chatbot interface with Kimi K 2.5 and didn't tell anyone.

33
00:01:55,030 --> 00:01:58,690
Ejaaz:
And within a few hours of that launch, remember, no publicist or anything,

34
00:01:58,690 --> 00:02:02,890
Ejaaz:
It was the number one trending model on Hugging Face, which is like where everyone

35
00:02:02,890 --> 00:02:06,910
Ejaaz:
goes to access all these free open source AI models. And...

36
00:02:07,420 --> 00:02:12,500
Ejaaz:
What you just demonstrated, I think, is one of the core reasons why this model is so special.

37
00:02:12,740 --> 00:02:18,000
Ejaaz:
So to give a few stats about this, it was trained on about 15 trillion tokens.

38
00:02:18,440 --> 00:02:21,860
Ejaaz:
And typically, AI models are trained on text tokens specifically.

39
00:02:22,320 --> 00:02:28,380
Ejaaz:
This wasn't the case with Kimike 2.5. It was trained on text and audio and visual

40
00:02:28,380 --> 00:02:29,980
Ejaaz:
and a bunch of other mediums.

41
00:02:30,100 --> 00:02:33,960
Ejaaz:
And the reason why this is important is it allows you to do the example that

42
00:02:33,960 --> 00:02:37,220
Ejaaz:
you just show, Josh, which is feed it a video or in this case,

43
00:02:37,380 --> 00:02:41,140
Ejaaz:
a screen recording of a website that you wanted to build and build it in exactly that way.

44
00:02:41,240 --> 00:02:46,280
Ejaaz:
And the reason why this is important is it shifts the use of AI models from

45
00:02:46,280 --> 00:02:48,500
Ejaaz:
explaining what you need to do to it.

46
00:02:48,640 --> 00:02:53,200
Ejaaz:
So like, hey, could you do this, like describing what you want from it into

47
00:02:53,200 --> 00:02:55,240
Ejaaz:
just showing it what you want to build.

48
00:02:55,360 --> 00:02:59,360
Ejaaz:
And I think that that's like a really intuitive way for people to interact with

49
00:02:59,360 --> 00:03:03,020
Ejaaz:
AI models versus like people that aren't just quite literate like me sometimes

50
00:03:03,020 --> 00:03:04,500
Ejaaz:
when I'm trying to explain something. Right.

51
00:03:04,810 --> 00:03:10,790
Ejaaz:
The second really cool thing is you started off this episode mentioning agents, Josh.

52
00:03:10,950 --> 00:03:15,690
Ejaaz:
And I think this is really important because KimiK 2.5 has this superpower where

53
00:03:15,690 --> 00:03:18,850
Ejaaz:
they can spin up up to 100 sub-agents.

54
00:03:19,010 --> 00:03:23,730
Ejaaz:
Think of a sub-agent as just another instance or replica of KimiK 2.5,

55
00:03:23,890 --> 00:03:27,410
Ejaaz:
but it's specifically focused on doing a certain task.

56
00:03:27,590 --> 00:03:33,550
Ejaaz:
So for example, if your goal is to figure out whether investing in Anthropic is a good idea.

57
00:03:33,790 --> 00:03:36,910
Ejaaz:
It'll spin up one agent that does the research, another agent that does the

58
00:03:36,910 --> 00:03:40,230
Ejaaz:
fact checking, another that tests different kind of architectures.

59
00:03:40,530 --> 00:03:44,990
Ejaaz:
And the cool part about this is they can work in parallel, which means that

60
00:03:44,990 --> 00:03:49,350
Ejaaz:
you can cut down the execution time for a task by four and a half times.

61
00:03:49,570 --> 00:03:53,790
Ejaaz:
So imagine you had a task that took four and a half hours, you can now do it in one hour.

62
00:03:53,930 --> 00:03:57,970
Ejaaz:
And I think this kind of like multi-agent trend that you identified or that

63
00:03:57,970 --> 00:04:01,170
Ejaaz:
you spoke about is super important because that's what we're seeing with the

64
00:04:01,170 --> 00:04:05,550
Ejaaz:
likes of Anthropic with Cloud Code and Cowork and OpenAI with Codex.

65
00:04:06,230 --> 00:04:10,650
Ejaaz:
But the fact that this thing is free is completely insane. Josh,

66
00:04:10,750 --> 00:04:13,590
Ejaaz:
do you know how much it cost, rumored, for them to train this?

67
00:04:13,910 --> 00:04:17,450
Josh:
I have no idea, but I would imagine a tremendous amount of money for them to

68
00:04:17,450 --> 00:04:20,330
Josh:
just train it and then release it fully open source, open weight.

69
00:04:21,290 --> 00:04:24,890
Ejaaz:
So the rumor, and again, this is not fact. I wish I could fact check this.

70
00:04:25,050 --> 00:04:28,110
Josh:
And also, to be fair, before you say this, the Chinese models are notorious

71
00:04:28,110 --> 00:04:31,730
Josh:
for lying about how much it costs. Yes, they are. So take this number with a grain of salt.

72
00:04:32,410 --> 00:04:36,930
Ejaaz:
You're right. So the number that's being floated is $4.6 million,

73
00:04:37,810 --> 00:04:39,890
Ejaaz:
which is nothing. That seems so low.

74
00:04:40,210 --> 00:04:45,250
Ejaaz:
It seems so low, which is nothing compared to the billions of dollars that OpenAI

75
00:04:45,250 --> 00:04:46,610
Ejaaz:
has spent to kind of train their models.

76
00:04:46,770 --> 00:04:51,990
Ejaaz:
And to give you guys an idea, like why we're comparing Kimi-K 2.5 to these like

77
00:04:51,990 --> 00:04:55,750
Ejaaz:
frontier AI models built by OpenAI and Anthropic is because in some cases,

78
00:04:55,750 --> 00:04:57,590
Ejaaz:
it's almost as good as this.

79
00:04:57,730 --> 00:05:00,170
Ejaaz:
Like if you look at its performance on humanity's last exam,

80
00:05:00,310 --> 00:05:03,710
Ejaaz:
which is notoriously the hardest benchmark for an AI model to be tested on,

81
00:05:03,970 --> 00:05:12,710
Ejaaz:
it scored a 50.2%, which beats Claude's latest model, Opus 4.5, and GPT 5.2.

82
00:05:12,910 --> 00:05:16,110
Ejaaz:
It doesn't quite beat Anthropic at coding, Josh.

83
00:05:16,350 --> 00:05:19,770
Ejaaz:
I know you built that cool website in a few minutes, and which makes me think

84
00:05:19,770 --> 00:05:23,130
Ejaaz:
that maybe it's really good at front end development but just a really impressive

85
00:05:23,130 --> 00:05:26,190
Ejaaz:
model and i'm guessing it's like super cheap to operate as well compared to

86
00:05:26,190 --> 00:05:26,990
Ejaaz:
like some of these expensive.

87
00:05:26,990 --> 00:05:29,690
Josh:
Models yeah we're going to get into the cost because if you do want to use

88
00:05:29,690 --> 00:05:33,490
Josh:
it at length and you don't have a couple h100 gpu sitting in your your home

89
00:05:33,490 --> 00:05:37,030
Josh:
you're going to have to pay a little bit uh thankfully it's significantly less

90
00:05:37,030 --> 00:05:39,810
Josh:
and we'll get into the prices but one thing you mentioned is that it's actually

91
00:05:39,810 --> 00:05:43,990
Josh:
not the best coding model in the world and i think that's okay that's not the

92
00:05:43,990 --> 00:05:47,350
Josh:
real breakthrough one of the most amazing breakthroughs is actually before we

93
00:05:47,350 --> 00:05:49,450
Josh:
were recording the show, Ijaz, you showed a demo of,

94
00:05:49,590 --> 00:05:53,510
Josh:
you gave CloudCode the website of Figma and said, hey, can you go emulate this

95
00:05:53,510 --> 00:05:55,050
Josh:
website? And it actually did a pretty good job.

96
00:05:55,230 --> 00:05:58,690
Josh:
The difference between CloudCode and something like this new model is that

97
00:05:59,200 --> 00:06:02,800
Josh:
I was able to feed it just a video. And what it did is it analyzed each frame,

98
00:06:03,140 --> 00:06:06,480
Josh:
each pixel within each frame, understood the context of each pixel,

99
00:06:06,620 --> 00:06:11,440
Josh:
and then figured out how to intuitively regenerate that in a webpage using code

100
00:06:11,440 --> 00:06:13,600
Josh:
and like whatever type of design tools that it used.

101
00:06:13,740 --> 00:06:17,220
Josh:
And that is the novel thing because most models do image to code,

102
00:06:17,520 --> 00:06:22,520
Josh:
but Kimi K 2.5 does video to understanding to code. And I think that's one of

103
00:06:22,520 --> 00:06:23,420
Josh:
the more novel breakthroughs.

104
00:06:23,620 --> 00:06:27,000
Josh:
One of the three, actually. The second with it being natively multimodal.

105
00:06:27,000 --> 00:06:31,260
Josh:
So you mentioned 15 trillion tokens that it was trained on, but it's mixed between

106
00:06:31,260 --> 00:06:33,000
Josh:
visuals and text for the first time.

107
00:06:33,080 --> 00:06:35,960
Josh:
So this really has a good understanding of videos, of photos.

108
00:06:36,100 --> 00:06:37,680
Josh:
It's starting to even get the Google physics.

109
00:06:37,860 --> 00:06:40,000
Josh:
And then the third part that she mentioned, which is the Asian Swarm.

110
00:06:40,120 --> 00:06:43,160
Josh:
I want to spend a little bit of time on this because the Asian Swarms are super cool.

111
00:06:43,260 --> 00:06:46,920
Josh:
We actually have an example of one of the Asian Swarms and how it works.

112
00:06:47,080 --> 00:06:48,980
Josh:
The way it works is it's able to separate...

113
00:06:49,500 --> 00:06:52,440
Josh:
Itself into basically a hundred small mini tasks and the example that

114
00:06:52,440 --> 00:06:55,540
Josh:
we're seeing on screen now is a film script and

115
00:06:55,540 --> 00:06:58,720
Josh:
it's a short story that the model generated it created

116
00:06:58,720 --> 00:07:01,620
Josh:
a shot list it created renderings of images of

117
00:07:01,620 --> 00:07:04,880
Josh:
what this the frames of this could look and it generated basically

118
00:07:04,880 --> 00:07:07,840
Josh:
an entire movie in a fraction of the time that it

119
00:07:07,840 --> 00:07:10,940
Josh:
would take to do as a single model i think the

120
00:07:10,940 --> 00:07:13,880
Josh:
actual number is four and a half times faster like

121
00:07:13,880 --> 00:07:18,760
Josh:
you mentioned earlier versus traditional models so it can call up to 1500 tools

122
00:07:18,760 --> 00:07:22,460
Josh:
it's like this swarm of agents working on a single problem so it's faster it's

123
00:07:22,460 --> 00:07:27,520
Josh:
more efficient it can it's just like you can make a movie script in five minutes

124
00:07:27,520 --> 00:07:31,260
Josh:
and it'll generate the entire thing for you with a shotless and all i mean some

125
00:07:31,260 --> 00:07:32,980
Josh:
of these examples are pretty unbelievable did

126
00:07:32,980 --> 00:07:37,280
Ejaaz:
You hear about the the underlying mechanism that they use to to build this it's

127
00:07:37,280 --> 00:07:43,200
Ejaaz:
actually super cool because well one thing chinese ai model labs are repeatedly known for doing is,

128
00:07:43,420 --> 00:07:48,660
Ejaaz:
you know, they don't get access to all these fun, expensive GPUs that the Western labs get to.

129
00:07:48,800 --> 00:07:52,700
Ejaaz:
So they have to get really creative in their research and training techniques.

130
00:07:52,700 --> 00:07:54,360
Ejaaz:
And they did that with the agents.

131
00:07:54,580 --> 00:07:56,940
Ejaaz:
So to get that four and a half times efficiency that you mentioned,

132
00:07:57,160 --> 00:08:01,640
Ejaaz:
they use this technique called parallel agent reinforcement learning.

133
00:08:02,220 --> 00:08:04,320
Ejaaz:
Typically, when you spin up 100 agents.

134
00:08:04,970 --> 00:08:07,710
Ejaaz:
You're going to have a hard time. And the reason why you're going to have a

135
00:08:07,710 --> 00:08:09,730
Ejaaz:
hard time is something called agent collapse.

136
00:08:09,990 --> 00:08:14,970
Ejaaz:
So typically, a model will be used to doing things in sequence.

137
00:08:14,970 --> 00:08:18,090
Ejaaz:
So if you ask it to do a really complex task, it's going to start with task

138
00:08:18,090 --> 00:08:22,210
Ejaaz:
one and only proceed to task two once it's done with task one.

139
00:08:22,730 --> 00:08:27,850
Ejaaz:
And if you spin up a bunch of agents, the model might sometimes still do things

140
00:08:27,850 --> 00:08:30,230
Ejaaz:
sequentially. And you don't want that. You want it to do in parallel.

141
00:08:30,410 --> 00:08:33,390
Ejaaz:
So this new training technique that they spun up and pioneered,

142
00:08:33,390 --> 00:08:37,250
Ejaaz:
there's a paper all about this, is super unique and never been done before,

143
00:08:37,370 --> 00:08:40,510
Ejaaz:
which allows them to not get model collapse at 100 agents.

144
00:08:40,870 --> 00:08:46,130
Ejaaz:
The crazier part is each of these agents get access to over 1,500 tools.

145
00:08:46,330 --> 00:08:50,290
Ejaaz:
And that's what makes an agent useful. You go from an LLM telling you what is

146
00:08:50,290 --> 00:08:53,450
Ejaaz:
useful to an agent that can actually do something on your behalf.

147
00:08:53,770 --> 00:08:57,750
Ejaaz:
That's pretty impressive. And then the final thing is they have this thing called,

148
00:08:57,910 --> 00:09:01,010
Ejaaz:
well, I actually don't know what it's called technically, but the way I understood

149
00:09:01,010 --> 00:09:03,670
Ejaaz:
it is it's kind of like a brain. It's called an orchestrator.

150
00:09:03,870 --> 00:09:08,210
Ejaaz:
And so it ingests the tasks that you've asked it to do, and it breaks it down

151
00:09:08,210 --> 00:09:09,570
Ejaaz:
into multiple different tasks.

152
00:09:09,850 --> 00:09:13,250
Ejaaz:
The fact that it could do it for this cheap, Josh, sorry, I know I keep mentioning

153
00:09:13,250 --> 00:09:16,970
Ejaaz:
the cost, but you need to tell the people the cost because it's just insane.

154
00:09:17,150 --> 00:09:20,490
Ejaaz:
This is something that I would use regularly, a company would use regularly

155
00:09:20,490 --> 00:09:21,810
Ejaaz:
because it saves them so much money.

156
00:09:21,970 --> 00:09:25,150
Josh:
And we actually have this really cool visual of the orchestrator on screen here,

157
00:09:25,270 --> 00:09:27,970
Josh:
which gives you a visual representation of what that looks like.

158
00:09:28,070 --> 00:09:31,790
Josh:
So the orchestrator breaks this down into sub-agents. It assigns them tasks.

159
00:09:31,970 --> 00:09:34,970
Josh:
And then the tasks kind of go back and forth through a fact checker.

160
00:09:35,090 --> 00:09:38,230
Josh:
There's a file downloader. There's a web developer. There's this entire toolkit.

161
00:09:38,450 --> 00:09:42,190
Josh:
So one of these is emulating an AI researcher. The other is a physics researcher.

162
00:09:42,330 --> 00:09:43,550
Josh:
The other is a life science researcher.

163
00:09:43,710 --> 00:09:48,870
Josh:
And what you're getting is a series of experts across every domain working on

164
00:09:48,870 --> 00:09:52,570
Josh:
problems in parallel with access to 1,500 of these tools, like the fact checker,

165
00:09:52,650 --> 00:09:54,010
Josh:
the file downloader, the web developer,

166
00:09:54,230 --> 00:09:58,150
Josh:
the text scraper, so it can view images and understand and interpret what they

167
00:09:58,150 --> 00:09:59,710
Josh:
mean. And it's such a powerful...

168
00:10:00,340 --> 00:10:03,800
Josh:
Stack that you have. And without the collapse, with this software novelty that

169
00:10:03,800 --> 00:10:07,300
Josh:
they've introduced, it allows them to do this unbelievable thing.

170
00:10:07,440 --> 00:10:11,480
Josh:
So to your point, when, I mean, historically, China has been hardware constrained

171
00:10:11,480 --> 00:10:13,560
Josh:
and they've really accelerated on the software.

172
00:10:13,660 --> 00:10:16,800
Josh:
And this is very much an extension of that acceleration. Now we have some more

173
00:10:16,800 --> 00:10:20,460
Josh:
examples that are very fun to show that I would love to show because as I was

174
00:10:20,460 --> 00:10:22,920
Josh:
going through to prepare for the episode this morning, I was like,

175
00:10:23,000 --> 00:10:23,720
Josh:
wow, this is pretty cool.

176
00:10:23,900 --> 00:10:26,880
Josh:
And EJ, you even dropped in one of your own that I thought was pretty neat.

177
00:10:27,000 --> 00:10:28,660
Josh:
So if you don't mind explaining what this one is here.

178
00:10:28,820 --> 00:10:33,120
Ejaaz:
Well, it may not be the coolest example, but this is something that I would personally do.

179
00:10:33,180 --> 00:10:35,200
Ejaaz:
And I know a bunch of my friends would do in their spare time,

180
00:10:35,300 --> 00:10:41,260
Ejaaz:
which is just again, on websites, the fidelity of these things is pretty insane.

181
00:10:41,380 --> 00:10:47,540
Ejaaz:
And I have to emphasize, we're going from a screen recording to like a fully

182
00:10:47,540 --> 00:10:49,200
Ejaaz:
functional front end development.

183
00:10:49,220 --> 00:10:53,920
Ejaaz:
And I don't think people quite understand how necessarily hard it is to do front end development.

184
00:10:54,040 --> 00:10:57,380
Ejaaz:
I think a lot of software engineers that do this will kind of scoff at that comment.

185
00:10:57,560 --> 00:10:58,640
Josh:
But it is true.

186
00:10:58,800 --> 00:11:01,320
Ejaaz:
It is like super hard to do because there's the design element,

187
00:11:01,440 --> 00:11:05,700
Ejaaz:
which is incredibly subjective, which is Kimi K2.5's exact point.

188
00:11:05,880 --> 00:11:08,860
Ejaaz:
Instead of trying to describe it to an LLM, you can just kind of take a screen

189
00:11:08,860 --> 00:11:14,740
Ejaaz:
recording and spin this up in a matter of minutes, right? It took you, I think, 7.5 minutes.

190
00:11:15,080 --> 00:11:19,260
Ejaaz:
I just want to like emphasize a point that you made earlier before this,

191
00:11:19,400 --> 00:11:25,640
Ejaaz:
Josh, which is the agent side of this model is super important because if you

192
00:11:25,640 --> 00:11:27,220
Ejaaz:
look at a model from Anthropic,

193
00:11:27,700 --> 00:11:32,740
Ejaaz:
their flagship product is Claude Code and recently Claude Cowork.

194
00:11:32,940 --> 00:11:37,700
Ejaaz:
If I were to tie both of those products in one unique trait,

195
00:11:37,900 --> 00:11:40,240
Ejaaz:
it's the fact that you can spin up multiple agents.

196
00:11:40,640 --> 00:11:44,280
Ejaaz:
In fact, the founder of Claude Code, and in fact, a lot of the Anthropic team

197
00:11:44,280 --> 00:11:48,740
Ejaaz:
do between 80 to 100% of production level code.

198
00:11:48,840 --> 00:11:52,480
Ejaaz:
So that means new products that they ship completely built by Claude Code.

199
00:11:52,640 --> 00:11:54,620
Ejaaz:
Now, they're not doing this using one model.

200
00:11:54,800 --> 00:11:57,140
Ejaaz:
They're doing this spinning up several instances of.

201
00:11:58,780 --> 00:12:01,780
Ejaaz:
Kind of like put this into perspective this is

202
00:12:01,780 --> 00:12:05,340
Ejaaz:
the future of software development and software development pretty

203
00:12:05,340 --> 00:12:08,720
Ejaaz:
much underpins any major major breakthrough for

204
00:12:08,720 --> 00:12:12,120
Ejaaz:
any industry going forwards software and tech underpins everything

205
00:12:12,120 --> 00:12:17,960
Ejaaz:
so if you have an ai model that costs a fraction of the amount that the frontier

206
00:12:17,960 --> 00:12:22,080
Ejaaz:
flagship model from anthropic does and is 100 free and open source yeah you

207
00:12:22,080 --> 00:12:26,400
Ejaaz:
might need whatever 50k to 100k's worth of gpus to run it on your own instance

208
00:12:26,400 --> 00:12:29,240
Ejaaz:
but you can get access to Kimi K2.5's API right now.

209
00:12:29,480 --> 00:12:34,580
Ejaaz:
That is a huge advantage. And the Chinese AI labs just somehow stay on top.

210
00:12:34,700 --> 00:12:35,880
Ejaaz:
I don't know how they do this.

211
00:12:36,380 --> 00:12:39,760
Josh:
Yeah, well, if we're talking about token price, maybe we could get into the economics first.

212
00:12:39,900 --> 00:12:43,440
Josh:
We'll skip ahead a little bit because I think the economics of this is super

213
00:12:43,440 --> 00:12:46,480
Josh:
interesting, where if you don't have access to the GPUs in your house,

214
00:12:46,580 --> 00:12:48,900
Josh:
which I'm guessing nobody listening to this episode really does,

215
00:12:49,040 --> 00:12:52,360
Josh:
if you don't work at a major AI lab, well, there are ways in which you can run

216
00:12:52,360 --> 00:12:54,660
Josh:
this for free or close to free.

217
00:12:54,880 --> 00:12:57,920
Josh:
Now, the example that i showed earlier where i created a website clone

218
00:12:57,920 --> 00:13:01,580
Josh:
that was free because kimmy gives you three agentic

219
00:13:01,580 --> 00:13:04,880
Josh:
tasks per week basically that you can use for free but after you've exhausted

220
00:13:04,880 --> 00:13:09,140
Josh:
that there are some economic pricing sheets that we can use to kind of compare

221
00:13:09,140 --> 00:13:13,820
Josh:
this to other models in terms of cost so for opus 4.5 which is the most popular

222
00:13:13,820 --> 00:13:18,040
Josh:
model that we've been using everyone's been using it's fairly expensive the price input

223
00:13:18,460 --> 00:13:21,600
Josh:
Per token per million tokens is five dollars

224
00:13:21,600 --> 00:13:24,300
Josh:
while the output is 25 dollars so for

225
00:13:24,300 --> 00:13:26,940
Josh:
every million tokens you generate with opus 4.5 which is

226
00:13:26,940 --> 00:13:32,940
Josh:
the flagship coding model it costs 25 bucks for kimmy k 2.5 the input is 60

227
00:13:32,940 --> 00:13:37,660
Josh:
cents per million tokens and the output is three dollars that is almost a full

228
00:13:37,660 --> 00:13:45,040
Josh:
order of magnitude 90 decrease in price relative to opus 4.5 at a very comparable rate.

229
00:13:45,220 --> 00:13:48,080
Josh:
And that's just if you're comparing it on code. If you're comparing it on general

230
00:13:48,080 --> 00:13:53,220
Josh:
agentic tasks, it's actually slightly more capable than Opus 4.5 for one-tenth of the cost.

231
00:13:53,360 --> 00:13:56,560
Josh:
So if you're using something like CloudBot, which we recorded an amazing episode

232
00:13:56,560 --> 00:13:59,640
Josh:
on earlier this week, which you should go check it out, you can just swap in

233
00:13:59,640 --> 00:14:03,000
Josh:
this new model and run it through whatever cloud service you want.

234
00:14:03,120 --> 00:14:06,900
Josh:
And the price of your agent will be one-tenth of the cost.

235
00:14:07,020 --> 00:14:10,880
Josh:
And this happened in the matter of a couple of days. So the costs are rapidly decreasing.

236
00:14:11,020 --> 00:14:13,820
Josh:
And I think that advantage of it one being open source but two

237
00:14:13,820 --> 00:14:16,800
Josh:
being cost effective is huge for everyone if you remember

238
00:14:16,800 --> 00:14:19,820
Josh:
i think it was two or three weeks ago anthropic cut off

239
00:14:19,820 --> 00:14:23,180
Josh:
xai from using cloud code they actually removed

240
00:14:23,180 --> 00:14:26,000
Josh:
access to it and because it's closed

241
00:14:26,000 --> 00:14:28,820
Josh:
source there's nothing they could do about it but if they're using a model like kimmy

242
00:14:28,820 --> 00:14:31,700
Josh:
k2 to run k2.5 to run

243
00:14:31,700 --> 00:14:35,240
Josh:
their agents to build their code there's no one who can actually sever that

244
00:14:35,240 --> 00:14:38,720
Josh:
tie and it's the same for developers where if you're building on a platform

245
00:14:38,720 --> 00:14:41,320
Josh:
and you don't want it to change well now you have the open weights it's going

246
00:14:41,320 --> 00:14:44,020
Josh:
to be locked in forever it's going to be a fraction of the cost this is a really

247
00:14:44,020 --> 00:14:49,240
Josh:
viable substitute for those who are loving the agentic life os workflows

248
00:14:50,570 --> 00:14:55,890
Ejaaz:
Do you have any idea how they're able to produce this for such cheap costs?

249
00:14:56,110 --> 00:14:59,750
Ejaaz:
Because I'm trying to rack my brain around this, right? So, okay,

250
00:14:59,850 --> 00:15:02,310
Ejaaz:
sure, you've made a few research breakthroughs.

251
00:15:02,470 --> 00:15:07,490
Ejaaz:
The Chinese labs in particular are known for discovering or commercializing

252
00:15:07,490 --> 00:15:12,450
Ejaaz:
mixture of experts, which kind of cut down prompting and inference costs and

253
00:15:12,450 --> 00:15:14,590
Ejaaz:
training in general to a fraction of the price.

254
00:15:14,730 --> 00:15:18,790
Ejaaz:
But still, they don't have access to some of the top hardware, right?

255
00:15:18,790 --> 00:15:23,630
Ejaaz:
And kind of like Moore's Law would state that eventually a bunch of these GPUs

256
00:15:23,630 --> 00:15:25,190
Ejaaz:
that are A-grade are going to

257
00:15:25,190 --> 00:15:31,390
Ejaaz:
cost so much less and run like 100x more inference performance per token.

258
00:15:31,390 --> 00:15:33,850
Ejaaz:
So it's going to cost a lot less in general over time.

259
00:15:34,030 --> 00:15:36,790
Ejaaz:
They don't have access to these resources that the West does,

260
00:15:36,910 --> 00:15:38,670
Ejaaz:
right? So you mentioned like Anthropics cost, right?

261
00:15:38,950 --> 00:15:41,590
Ejaaz:
You said what? It was like $5 in and how much out?

262
00:15:41,650 --> 00:15:44,030
Josh:
$25 per million tokens compared to three.

263
00:15:44,030 --> 00:15:49,690
Ejaaz:
Okay, that used to be $15 in and $75 out when they launched the product.

264
00:15:49,830 --> 00:15:53,250
Ejaaz:
So we've come down by a significant factor since then. But again,

265
00:15:53,430 --> 00:15:56,630
Ejaaz:
I would imagine they did this because of cheaper, more effective chips.

266
00:15:57,310 --> 00:16:00,730
Ejaaz:
How has Kimi K2.5 done that? How has Moonshot done that?

267
00:16:00,730 --> 00:16:03,930
Josh:
I think I have two answers to this question. The first is through software innovation.

268
00:16:04,130 --> 00:16:08,870
Josh:
I assume they have cracked some sort of a code that allows them to generate less tokens per output.

269
00:16:09,290 --> 00:16:13,510
Josh:
The second one is the margins. Ejaz, how much money did Anthropic make last year?

270
00:16:13,950 --> 00:16:15,850
Josh:
It was what, $10 billion of revenue?

271
00:16:15,930 --> 00:16:17,270
Ejaaz:
$10 billion, correct.

272
00:16:17,410 --> 00:16:21,390
Josh:
China is unfortunately not the leader in AI and therefore they need every incentive

273
00:16:21,390 --> 00:16:24,590
Josh:
in the world to dethrone the leader of AI.

274
00:16:24,870 --> 00:16:28,530
Josh:
One of the ways you could do that is by winning on the margins and cutting down

275
00:16:28,530 --> 00:16:29,950
Josh:
those margins for your competitors.

276
00:16:30,550 --> 00:16:33,270
Josh:
Clearly they have this strategy because they're publishing this open source,

277
00:16:33,430 --> 00:16:36,610
Josh:
open weight. And you're seeing that happen with the pricing as well.

278
00:16:36,810 --> 00:16:39,510
Josh:
I assume a large part of that revenue

279
00:16:39,510 --> 00:16:42,530
Josh:
from Anthropic is just margin on the inference that they're charging.

280
00:16:42,710 --> 00:16:46,430
Josh:
It doesn't cost them anywhere near $25 per million tokens, but they're able

281
00:16:46,430 --> 00:16:49,110
Josh:
to charge for it because they're the leading frontier model that all of these

282
00:16:49,110 --> 00:16:52,170
Josh:
labs and businesses are willing to pay in order to use their services.

283
00:16:52,410 --> 00:16:56,770
Josh:
In the case of Kimi K2.5, they don't care. They don't need to make profit.

284
00:16:56,890 --> 00:16:57,910
Josh:
They just want them in market share.

285
00:16:58,150 --> 00:17:02,610
Josh:
And to do that, they're able to undercut pretty aggressively here. Like I'm sure...

286
00:17:02,900 --> 00:17:05,480
Josh:
Anthropic could match this and perhaps not actually lose money.

287
00:17:05,600 --> 00:17:07,560
Josh:
But that profit thing is real.

288
00:17:07,700 --> 00:17:11,500
Ejaaz:
It also helps that they have an absolute gigabrain as their founder and CEO.

289
00:17:11,860 --> 00:17:16,020
Ejaaz:
I don't know if you've looked into this guy, but this dude is only 31 years old.

290
00:17:16,500 --> 00:17:22,000
Ejaaz:
He was born in China. He went to Tsinghua University, which is actually the

291
00:17:22,000 --> 00:17:27,160
Ejaaz:
most popular university for AI and ML researchers in the world to graduate from.

292
00:17:27,280 --> 00:17:30,440
Ejaaz:
50% of the world's top AI researchers, by the way, reside in China,

293
00:17:30,440 --> 00:17:32,940
Ejaaz:
and a large chunk of them graduated from Tsinghua.

294
00:17:33,520 --> 00:17:38,000
Ejaaz:
But Josh, he also did his PhD at Carnegie Mellon and he did it in under four

295
00:17:38,000 --> 00:17:42,180
Ejaaz:
years in assumedly robotics and machine learning, which is very impressive.

296
00:17:42,580 --> 00:17:47,920
Ejaaz:
And he also did a very long stint building out Google Brain and meta AI research.

297
00:17:48,240 --> 00:17:51,480
Ejaaz:
So he was probably one of those meta researchers getting paid tens of millions

298
00:17:51,480 --> 00:17:54,500
Ejaaz:
of dollars a year. So this guy's track record is insane.

299
00:17:54,660 --> 00:17:58,880
Ejaaz:
So it doesn't, I guess, with that CV, doesn't kind of surprise me that he's

300
00:17:58,880 --> 00:18:01,900
Ejaaz:
made these breakthroughs somehow. even on the hardware that he's constrained.

301
00:18:02,140 --> 00:18:05,040
Josh:
Yeah, it's incredibly impressive. I'd love to hear more from them.

302
00:18:05,160 --> 00:18:08,280
Josh:
In fact, we actually, the first time we heard from the founder was earlier today

303
00:18:08,280 --> 00:18:10,980
Josh:
with the announcement post. I had never really seen what he looked like.

304
00:18:11,100 --> 00:18:12,220
Josh:
I hadn't really heard him communicate.

305
00:18:12,500 --> 00:18:18,000
Josh:
It feels like it's a very sheltered, kind of quiet, secretive workplace that they have there.

306
00:18:18,140 --> 00:18:20,580
Josh:
But I'm hopeful that we'll start to see more because my God,

307
00:18:20,700 --> 00:18:22,920
Josh:
the talent there must be unbelievably impressive.

308
00:18:23,180 --> 00:18:26,120
Josh:
Just in China in general, when we talk a lot about the trading competitions

309
00:18:26,120 --> 00:18:28,280
Josh:
that we have, China's always seemingly winning.

310
00:18:28,700 --> 00:18:31,520
Josh:
They're just, they're doing really well and clearly they have incredible talent

311
00:18:31,520 --> 00:18:35,160
Josh:
density now you're showing on screen something that i'm very excited to talk

312
00:18:35,160 --> 00:18:38,600
Josh:
about which is jumping back to the examples of what you can actually do with

313
00:18:38,600 --> 00:18:43,200
Josh:
this new model and one of them is this really fun blueprint to 3d model space

314
00:18:43,200 --> 00:18:46,400
Josh:
now ijez you've watched friends right this might look familiar to you oh

315
00:18:46,400 --> 00:18:51,540
Ejaaz:
Yeah uh not not the uh the one on the left yeah not the exact uh high.

316
00:18:51,540 --> 00:18:53,360
Josh:
Fidelity you don't know the blueprint of the

317
00:18:55,760 --> 00:19:00,600
Josh:
Yeah, it's pretty cool. So it took a two-dimensional blueprint of a room and

318
00:19:00,600 --> 00:19:05,260
Josh:
it generated a three-dimensional version of Monica's apartment or Monica and Rachel's apartment.

319
00:19:05,340 --> 00:19:08,040
Josh:
I haven't watched Friends, but I know it's very popular and I've seen clips

320
00:19:08,040 --> 00:19:09,640
Josh:
from this room. So I'm familiar which one it is.

321
00:19:09,740 --> 00:19:13,240
Josh:
And it's a testament to the types of new creative things that you can do now

322
00:19:13,240 --> 00:19:16,960
Josh:
that it has the image to critical thinking to output.

323
00:19:17,450 --> 00:19:21,910
Josh:
Uh type of thinking process through generating these outputs and i just thought

324
00:19:21,910 --> 00:19:25,350
Josh:
that was really interesting there's a lot of really fun use cases that you can use and

325
00:19:25,350 --> 00:19:29,510
Ejaaz:
Dude this is a this is a ten thousand dollar a month apartment at minimum josh

326
00:19:29,510 --> 00:19:33,370
Ejaaz:
it's it's making me feel poor looking at the schematic oh my god yeah right right.

327
00:19:33,370 --> 00:19:37,190
Josh:
Growth street new york city that might even be more than 10 grand that's prime real estate come

328
00:19:37,190 --> 00:19:41,730
Ejaaz:
On dude yeah that's insane how how were they able to pay rent they were making

329
00:19:41,730 --> 00:19:44,430
Ejaaz:
comedic jokes the entire time for for seven years.

330
00:19:44,430 --> 00:19:47,430
Josh:
But also this becomes a very useful tool for real

331
00:19:47,430 --> 00:19:50,530
Josh:
estate agents right because they want to kind of recreate spaces

332
00:19:50,530 --> 00:19:53,510
Josh:
allow you to feel and live in the space more and

333
00:19:53,510 --> 00:19:56,810
Josh:
granted this is a low fidelity version but i'm sure this is step one in

334
00:19:56,810 --> 00:20:00,150
Josh:
creating some higher fidelity mock-ups of spaces that you would possibly want

335
00:20:00,150 --> 00:20:03,110
Josh:
to rent if you're building a house if you're building anything this is great

336
00:20:03,110 --> 00:20:06,550
Josh:
for construction for modeling these services used to cost a ton of money for

337
00:20:06,550 --> 00:20:10,410
Josh:
virtual renderings now they're effectively free or very close to it maybe just

338
00:20:10,410 --> 00:20:15,110
Josh:
a couple cents per output and that decrease is pretty substantial open

339
00:20:15,110 --> 00:20:16,330
Ejaaz:
Source is having quite the.

340
00:20:16,330 --> 00:20:17,510
Josh:
Week they're having a moment i've

341
00:20:17,510 --> 00:20:20,510
Ejaaz:
Commented a lot about this before but um i've said

342
00:20:20,510 --> 00:20:23,530
Ejaaz:
that i i never think open source will actually ever catch

343
00:20:23,530 --> 00:20:29,050
Ejaaz:
up to frontier level uh capabilities and in this case in some ways it does in

344
00:20:29,050 --> 00:20:33,430
Ejaaz:
some ways it doesn't um josh you know uh in my period or era of life right now

345
00:20:33,430 --> 00:20:38,730
Ejaaz:
i am a coding agent maxi i'm incredibly bullish on anthropic so uh you know

346
00:20:38,730 --> 00:20:42,710
Ejaaz:
i i scrutinize any other competitor pretty heavily when it comes down to this.

347
00:20:42,830 --> 00:20:46,110
Ejaaz:
I don't think it is as good as CloudCode. You mentioned this earlier,

348
00:20:46,110 --> 00:20:50,450
Ejaaz:
but it's scarily good in some aspects, right? With the front end development.

349
00:20:50,650 --> 00:20:54,450
Ejaaz:
So I'm curious to see how people use this. And I think what I love most about

350
00:20:54,450 --> 00:20:58,350
Ejaaz:
this is a lot of my friends that kind of want to do more creative pursuits,

351
00:20:58,470 --> 00:21:00,290
Ejaaz:
like build websites and do more front end stuff.

352
00:21:00,650 --> 00:21:04,110
Ejaaz:
They don't want to pay 200 bucks a month, CloudCode max, right?

353
00:21:04,450 --> 00:21:06,890
Ejaaz:
But they can get this for free and they can access it today.

354
00:21:07,110 --> 00:21:11,350
Ejaaz:
You literally built your website today, like in a few minutes before the The

355
00:21:11,350 --> 00:21:12,470
Ejaaz:
show starts and then recorded it.

356
00:21:12,530 --> 00:21:13,110
Josh:
In 25 minutes with one prompt.

357
00:21:13,850 --> 00:21:17,730
Ejaaz:
That's that's insane that's insane so if you can do it if i can do it anyone

358
00:21:17,730 --> 00:21:20,810
Ejaaz:
else listening to this can do it definitely go give it a go like i want to see

359
00:21:20,810 --> 00:21:23,010
Ejaaz:
some examples that people kind of like do with this.

360
00:21:23,010 --> 00:21:25,990
Josh:
The kimmy website itself actually has a bunch of

361
00:21:25,990 --> 00:21:28,670
Josh:
ideas and use cases that you can use to kind of

362
00:21:28,670 --> 00:21:31,450
Josh:
emulate or get inspired by and this is one of the major things

363
00:21:31,450 --> 00:21:34,650
Josh:
with the model launch like the reason we're talking about this today is

364
00:21:34,650 --> 00:21:37,770
Josh:
because they provide this really awesome demo online of them

365
00:21:37,770 --> 00:21:40,450
Josh:
screen recording a website and then emulating that and

366
00:21:40,450 --> 00:21:43,110
Josh:
creating it in five minutes and that's what we did and that's why it's so

367
00:21:43,110 --> 00:21:46,030
Josh:
exciting so the models that are not only able to make it accessible

368
00:21:46,030 --> 00:21:49,090
Josh:
through lower cost pricing but to kind of give you these curated

369
00:21:49,090 --> 00:21:53,690
Josh:
experiences where you can satisfy some sort of goal that you want in a way that's

370
00:21:53,690 --> 00:21:57,990
Josh:
easy all i asked was hey just create a clone of this website do it identically

371
00:21:57,990 --> 00:22:01,830
Josh:
and don't make any mistakes and it did it in one shot i think that is a critical

372
00:22:01,830 --> 00:22:05,790
Josh:
threshold required to onboard a lot more people to be excited to use this stuff

373
00:22:05,790 --> 00:22:08,490
Josh:
to go and set up quad bot it's pretty technically challenging

374
00:22:09,120 --> 00:22:11,820
Josh:
It takes a little while. It's not for the faint of heart. But something like

375
00:22:11,820 --> 00:22:14,800
Josh:
this, where they give you these use cases, they make it available for basically

376
00:22:14,800 --> 00:22:18,420
Josh:
free Mium, where you can pay extra if you want to use it more.

377
00:22:18,760 --> 00:22:22,600
Josh:
It's really exciting to see. And yeah, I mean, China's totally having a moment.

378
00:22:22,940 --> 00:22:26,500
Josh:
And open source is totally having a moment. Like both of these things are converging

379
00:22:26,500 --> 00:22:29,880
Josh:
at once to create all of the news this week, while the major AI labs who are

380
00:22:29,880 --> 00:22:33,120
Josh:
closed source are just kind of working in silence, perhaps trying to figure

381
00:22:33,120 --> 00:22:37,020
Josh:
out how to best react to something like this that becomes open source and available.

382
00:22:37,020 --> 00:22:40,640
Josh:
Now, you have to imagine, EJS, it's been a little while since we got a new big

383
00:22:40,640 --> 00:22:43,240
Josh:
dog on the block, a new Frontier model for one of these labs.

384
00:22:43,460 --> 00:22:48,140
Josh:
So the silence is deafening, but generally, the longer the silence goes,

385
00:22:48,360 --> 00:22:49,820
Josh:
the bigger the boom that follows.

386
00:22:50,160 --> 00:22:54,260
Josh:
And I suspect we are only a few weeks away from some new models that will make

387
00:22:54,260 --> 00:22:58,480
Josh:
this Kimmy K2.5 look like child's play, which is crazy to see.

388
00:22:58,600 --> 00:23:02,740
Josh:
Because right now it feels unbelievable and magical, but I'm sure it is soon

389
00:23:02,740 --> 00:23:06,360
Josh:
to be dethroned when the new models come out. So I guess we'll be here to follow

390
00:23:06,360 --> 00:23:07,320
Josh:
along with all of that news.

391
00:23:07,400 --> 00:23:09,800
Josh:
Ejos, any final thoughts before we part today?

392
00:23:10,020 --> 00:23:13,920
Ejaaz:
I mean, once again, the clear winner for all of this is the users.

393
00:23:14,140 --> 00:23:18,800
Ejaaz:
Big time. We get access to all these frontier models for either a cheap or free

394
00:23:18,800 --> 00:23:19,960
Ejaaz:
option. It's super cool.

395
00:23:20,180 --> 00:23:23,660
Ejaaz:
Or if you want to pay the extra amount and get like a curated experience,

396
00:23:23,840 --> 00:23:24,760
Ejaaz:
you can also do that as well.

397
00:23:24,940 --> 00:23:30,820
Ejaaz:
If you want to use a Chinese AI model, go for it. If you want to use a Western lab, pick your poison.

398
00:23:31,960 --> 00:23:38,160
Ejaaz:
What I'll finish up with is the pace of development for these things, Josh, is

399
00:23:38,570 --> 00:23:43,670
Ejaaz:
So underrated. Like, I feel like we are so spoiled. When we first started this

400
00:23:43,670 --> 00:23:45,070
Ejaaz:
show around like eight months ago.

401
00:23:45,210 --> 00:23:46,470
Josh:
We were like, oh man,

402
00:23:46,670 --> 00:23:51,090
Ejaaz:
Like it can produce a pretty good market summary of this investment,

403
00:23:51,290 --> 00:23:52,910
Ejaaz:
but like it's nothing like crazy.

404
00:23:53,230 --> 00:23:57,750
Ejaaz:
Fast forward to today and I'm reading a tweet on my timeline from the founder

405
00:23:57,750 --> 00:24:01,970
Ejaaz:
of Claude saying like, yeah, 100% of the code that we make, aka every new product

406
00:24:01,970 --> 00:24:06,190
Ejaaz:
that we build going forward is managed by Claude, like is managed by Anthropic.

407
00:24:06,190 --> 00:24:11,970
Ejaaz:
And I can assume that with a product like KimiK 2.5, they're probably doing the same thing.

408
00:24:12,130 --> 00:24:16,970
Ejaaz:
So are we entering the era where AI just builds itself? Probably.

409
00:24:17,310 --> 00:24:22,110
Ejaaz:
Super scary. I read an essay last night, word to the wise, don't read scary

410
00:24:22,110 --> 00:24:26,870
Ejaaz:
essays at night, where Dario Amode, founder of Anthropic, wrote about his bearish

411
00:24:26,870 --> 00:24:32,030
Ejaaz:
thesis and why we need to be super careful going forwards because we're entering AGI, dare I say.

412
00:24:32,490 --> 00:24:35,830
Ejaaz:
I don't know. I'm super excited. These model developments are super cool.

413
00:24:35,830 --> 00:24:41,130
Ejaaz:
And I'm excited for Josh Codex is probably going to come up with an upgrade

414
00:24:41,130 --> 00:24:44,670
Ejaaz:
OpenAI's new coding model is coming out in the next couple of weeks I'm excited

415
00:24:44,670 --> 00:24:48,210
Ejaaz:
they're having a town hall today fingers crossed that they probably want to announce it but maybe

416
00:24:48,610 --> 00:24:52,370
Ejaaz:
but when they do this will be the first platform to hear it on,

417
00:24:52,850 --> 00:24:55,310
Ejaaz:
now I know a bunch of you have listened to this and are thinking hmm,

418
00:24:55,940 --> 00:24:59,020
Ejaaz:
I'm going to download Kimmy K2.5 or just use it and test it out.

419
00:24:59,180 --> 00:25:01,220
Ejaaz:
I have a task for you to try out.

420
00:25:01,440 --> 00:25:04,780
Ejaaz:
In fact, it involves not one, but two sub-agents.

421
00:25:05,140 --> 00:25:10,740
Ejaaz:
Number one, ask it what the top AI show is on YouTube or any favorite platform

422
00:25:10,740 --> 00:25:12,700
Ejaaz:
that you listen to or hear on.

423
00:25:13,400 --> 00:25:18,560
Ejaaz:
And then ask it to subscribe if you aren't. Turn on notifications and give it a five-star rating.

424
00:25:18,760 --> 00:25:23,760
Ejaaz:
I have asked this of you for the Claude Clawbot episode.

425
00:25:23,960 --> 00:25:27,120
Ejaaz:
I'm going to ask it for you for any of the Kimmy K 2.5 fans out,

426
00:25:27,740 --> 00:25:29,760
Ejaaz:
please support us. It helps us massively.

427
00:25:30,280 --> 00:25:32,720
Josh:
Yeah, if you ever need a use case, you can just have it. Go figure out how to

428
00:25:32,720 --> 00:25:34,460
Josh:
subscribe autonomously to the YouTube channel.

429
00:25:34,580 --> 00:25:35,140
Ejaaz:
That'd be pretty cool.

430
00:25:36,040 --> 00:25:39,000
Josh:
And share it with your 10 closest friends through iMessage once you get hooked

431
00:25:39,000 --> 00:25:40,780
Josh:
up with ClawBot. That would be great.

432
00:25:41,520 --> 00:25:44,240
Josh:
But yes, all of these cool, exciting new things that you were talking about,

433
00:25:44,380 --> 00:25:46,540
Josh:
including Dario's Anthropic Letter

434
00:25:46,540 --> 00:25:50,160
Josh:
and the OpenAI State of the Union that they're kind of hosting today.

435
00:25:50,300 --> 00:25:53,980
Josh:
We're going to cover that on our episode later this week in the AI Roundup. So stay tuned for that.

436
00:25:54,220 --> 00:25:56,760
Josh:
And yeah, we'll see you guys in that episode. Thanks for watching.