1
00:00:00,020 --> 00:00:03,980
Josh:
How much money are you paying to use your AI model? Maybe it's $20 a month,

2
00:00:04,120 --> 00:00:07,500
Josh:
maybe you're on the pro plan for $200 a month, maybe you're running an OpenClaw

3
00:00:07,500 --> 00:00:11,680
Josh:
instance and you're paying thousands of dollars a month to generate tokens from Frontier Models.

4
00:00:11,800 --> 00:00:15,660
Josh:
Google has just released a solution to your problem, something that can be solved

5
00:00:15,660 --> 00:00:19,260
Josh:
for as little as an $80 one-time fee just to run a Raspberry Pi,

6
00:00:19,560 --> 00:00:21,660
Josh:
because that's what this new model runs on.

7
00:00:22,020 --> 00:00:26,020
Josh:
Gemma 4 is a new model from Google that is a hyper-quantized,

8
00:00:26,040 --> 00:00:31,860
Josh:
very small model meant to run locally on devices like your phone or your laptop

9
00:00:31,860 --> 00:00:34,040
Josh:
or even your new MacBook Neo.

10
00:00:34,200 --> 00:00:38,820
Josh:
It's very lightweight and it's built for working offline entirely private and

11
00:00:38,820 --> 00:00:41,880
Josh:
I think the thing that's most noteworthy is how powerful it is.

12
00:00:42,060 --> 00:00:45,340
Josh:
This model that is small enough to fit on your phone and run entirely for free

13
00:00:45,340 --> 00:00:49,280
Josh:
is just as good if not better than some of the Frontier models last year and

14
00:00:49,280 --> 00:00:51,840
Josh:
is even close to performing as well as them this year.

15
00:00:52,180 --> 00:00:55,340
Josh:
Now to showcase this we have some really cool examples that EJS has prepared

16
00:00:55,340 --> 00:00:58,360
Josh:
here so let's get into what this new Gemma 4 model can do.

17
00:00:58,910 --> 00:01:02,190
Ejaaz:
Yeah, I'm super excited about this model. I mean, it's not just one either.

18
00:01:02,450 --> 00:01:06,290
Ejaaz:
There's four of them. And like you said, it ranges from like 4 billion to,

19
00:01:06,970 --> 00:01:09,410
Ejaaz:
I think it's like 50 billion parameter models.

20
00:01:09,930 --> 00:01:12,710
Ejaaz:
Can fit on your phone, can fit on any device. And like you said,

21
00:01:12,830 --> 00:01:15,090
Ejaaz:
eight months ago, this would have been considered frontier intelligence.

22
00:01:15,230 --> 00:01:19,390
Ejaaz:
But I want to get into like what these things can actually do because it's one

23
00:01:19,390 --> 00:01:20,470
Ejaaz:
thing talking about benchmarks.

24
00:01:20,630 --> 00:01:23,570
Ejaaz:
It's another thing talking about what it can do on your phone,

25
00:01:23,670 --> 00:01:28,150
Ejaaz:
on your laptop, what value it can bring to you. This first example is someone

26
00:01:28,150 --> 00:01:31,490
Ejaaz:
leveraging the visual intelligence of these Gemma models.

27
00:01:31,650 --> 00:01:35,630
Ejaaz:
Now, typically, if you're an AI model, you're really good at ingesting words

28
00:01:35,630 --> 00:01:39,390
Ejaaz:
and characters and understanding the word described to you like a book would

29
00:01:39,390 --> 00:01:40,550
Ejaaz:
or like a blog post would.

30
00:01:40,930 --> 00:01:44,910
Ejaaz:
Visual intelligence is a very different frontier that has often been hard to

31
00:01:44,910 --> 00:01:47,570
Ejaaz:
kind of surmount by these new AI models.

32
00:01:47,950 --> 00:01:51,370
Ejaaz:
Gemma does an amazing job. What you're looking at on your screen right now is

33
00:01:51,370 --> 00:01:56,570
Ejaaz:
its ability to identify all the different objects in what is a very crowded room.

34
00:01:56,670 --> 00:01:58,890
Ejaaz:
He raises up a banana and it identifies that.

35
00:01:59,030 --> 00:02:01,970
Ejaaz:
It's also spotting the books that are on his shelf in the back.

36
00:02:02,090 --> 00:02:05,570
Ejaaz:
It's spotting the shelf in itself, the lamp, the fact that he's in a room,

37
00:02:05,670 --> 00:02:06,530
Ejaaz:
the curtains around that.

38
00:02:06,750 --> 00:02:10,770
Ejaaz:
And that's really important when it comes to creating apps that can log your

39
00:02:10,770 --> 00:02:14,530
Ejaaz:
visual experience or can track your calories for the food that you're consuming.

40
00:02:14,610 --> 00:02:19,550
Ejaaz:
And you can build a completely different suite of apps based on intelligence like this.

41
00:02:19,650 --> 00:02:23,510
Ejaaz:
This is the first time that we're seeing it appear in an open source open weight

42
00:02:23,510 --> 00:02:25,290
Ejaaz:
model. And Google's been the first to launch that.

43
00:02:25,740 --> 00:02:28,180
Josh:
Yeah. And if it wasn't abundantly clear, this is totally free.

44
00:02:28,360 --> 00:02:30,620
Josh:
You could just go on the website, download this and run it yourself.

45
00:02:30,760 --> 00:02:33,520
Josh:
And I think looking at this vision example, one of the cool things that I'm

46
00:02:33,520 --> 00:02:36,900
Josh:
thinking of is a lot of people have cameras outside their house,

47
00:02:37,040 --> 00:02:37,720
Josh:
outside their apartments.

48
00:02:38,040 --> 00:02:41,880
Josh:
And this has visual intelligence now to not only see things,

49
00:02:42,000 --> 00:02:43,720
Josh:
but alert you of what it's seeing.

50
00:02:43,940 --> 00:02:46,840
Josh:
One of the cool examples that I saw that we don't have teed up here is just

51
00:02:46,840 --> 00:02:49,500
Josh:
someone who had a little nest cameras right outside the front door.

52
00:02:49,660 --> 00:02:52,700
Josh:
And it would send a notification of what was happening.

53
00:02:52,960 --> 00:02:56,040
Josh:
It's like there is a dog walking in front of your house. There is a man walking

54
00:02:56,040 --> 00:02:58,820
Josh:
up with two packages in his hands. It looks like the package is from Amazon.

55
00:02:58,980 --> 00:03:02,300
Josh:
And it has that visual intelligence that would normally cost quite a bit of

56
00:03:02,300 --> 00:03:06,340
Josh:
money in those tokens using something like Claude Opus or ChatGPT.

57
00:03:06,640 --> 00:03:08,860
Josh:
But instead, it does it all for free on this tiny little model,

58
00:03:08,940 --> 00:03:12,840
Josh:
which is super cool. We have another example that was mentioned in the intro about OpenClaw here.

59
00:03:13,280 --> 00:03:17,500
Josh:
And OpenClaw is something that a lot of people spend a lot of money on.

60
00:03:17,660 --> 00:03:21,660
Josh:
If people are real hardcore users they're spending hundreds of dollars a day

61
00:03:21,660 --> 00:03:24,860
Josh:
up to hundreds of dollars a day some even thousands of dollars a day in addition

62
00:03:24,860 --> 00:03:28,720
Josh:
to buying some pretty expensive hardware to run it on a lot of people bought

63
00:03:28,720 --> 00:03:31,780
Josh:
mac minis you can't get a mac mini if you wanted to because they're so backordered

64
00:03:31,780 --> 00:03:35,420
Josh:
mac studios people were paying hundreds if not thousands of dollars to run.

65
00:03:35,920 --> 00:03:38,020
Josh:
This software on and the reality is is that

66
00:03:38,440 --> 00:03:41,260
Josh:
whatever device you're watching this on, whatever device you're listening to

67
00:03:41,260 --> 00:03:43,260
Josh:
right now, you can run this model on.

68
00:03:43,360 --> 00:03:46,660
Josh:
You don't need something fancy. You don't need a high-end computer to run it.

69
00:03:46,840 --> 00:03:50,240
Josh:
You can just do this on something local, lightweight, and like I mentioned,

70
00:03:50,360 --> 00:03:55,280
Josh:
as lightweight as an $80 Raspberry Pi because you can use the lightest weight model possible.

71
00:03:55,440 --> 00:03:58,880
Josh:
And although the results aren't the best in the world, they are much better

72
00:03:58,880 --> 00:04:01,820
Josh:
than previously expected from these open source models.

73
00:04:02,060 --> 00:04:06,380
Ejaaz:
Yeah, I love that you can finally run OpenClaw on a device that doesn't cost $1,000.

74
00:04:07,180 --> 00:04:10,820
Ejaaz:
These Mac minis actually on the secondary market have gone up sky high.

75
00:04:10,980 --> 00:04:14,040
Ejaaz:
Like the retail price is 800 bucks because you can't get it from Apple anymore.

76
00:04:14,180 --> 00:04:17,320
Ejaaz:
I've seen it as high as like 2K and people are still buying these things.

77
00:04:17,620 --> 00:04:20,940
Ejaaz:
Now, the reason why they've been buying these things is because they can't fit

78
00:04:20,940 --> 00:04:26,140
Ejaaz:
Frontier open source models onto their own mobile phone or their own laptop.

79
00:04:26,300 --> 00:04:29,340
Ejaaz:
And now Gemma 4 has made it super easy to do. So that's amazing.

80
00:04:29,640 --> 00:04:32,980
Ejaaz:
I'm still quite confused as to what the open core users are burning thousands

81
00:04:32,980 --> 00:04:37,080
Ejaaz:
of dollars for on, but like that's probably a topic for another conversation.

82
00:04:37,320 --> 00:04:41,600
Ejaaz:
The other thing that I like about this is Gemma 4 can run completely offline.

83
00:04:41,780 --> 00:04:44,640
Ejaaz:
Now this is a common property and characteristic that you

84
00:04:44,640 --> 00:04:47,840
Ejaaz:
can have for every single open source model but the fact is you

85
00:04:47,840 --> 00:04:52,600
Ejaaz:
have a model here that is near frontier intelligence to say Claude Opus 4.6

86
00:04:52,600 --> 00:04:56,700
Ejaaz:
and GPT-504 and we'll get to those direct comparisons a little later on in this

87
00:04:56,700 --> 00:05:01,020
Ejaaz:
episode but now you can run it offline and the great part about this is often

88
00:05:01,020 --> 00:05:04,980
Ejaaz:
you're in areas where you just don't have internet connection or it takes a while to inference.

89
00:05:05,180 --> 00:05:07,660
Ejaaz:
Now you have it on your phone, you can have it completely offline,

90
00:05:07,780 --> 00:05:10,280
Ejaaz:
it gets access to the world's entire database of knowledge.

91
00:05:10,440 --> 00:05:14,080
Ejaaz:
It might not be real-time, fair enough, but you still get access to core knowledge

92
00:05:14,080 --> 00:05:17,200
Ejaaz:
when you're in a bit of a desperate situation or when you just don't want to

93
00:05:17,200 --> 00:05:18,740
Ejaaz:
use the internet, which I thought was really cool.

94
00:05:19,210 --> 00:05:23,150
Josh:
This part is maybe my favorite, where it feels like you truly have access to

95
00:05:23,150 --> 00:05:26,210
Josh:
intelligence at your fingertips, no matter where you are in the world.

96
00:05:26,230 --> 00:05:28,930
Josh:
You can be stranded on an island. You could have no connection.

97
00:05:29,190 --> 00:05:34,950
Josh:
You can be anywhere at any time. And it is completely and totally locally and free.

98
00:05:35,030 --> 00:05:37,570
Josh:
And it fits on your phone. And it feels like having Google on your phone.

99
00:05:37,670 --> 00:05:40,810
Josh:
I remember growing up, it's like, you're not going to be able to Google everything.

100
00:05:40,810 --> 00:05:42,070
Josh:
You have to learn these things.

101
00:05:42,190 --> 00:05:46,610
Josh:
And the reality is, is that you have a super genius. Now that gets packaged

102
00:05:46,610 --> 00:05:49,370
Josh:
up into something as small as your phone. And that is super cool.

103
00:05:49,710 --> 00:05:53,730
Ejaaz:
Now, naturally, where your mind goes with that property is, huh,

104
00:05:53,910 --> 00:05:57,110
Ejaaz:
if I'm in a desperate situation, can AI save my life?

105
00:05:57,310 --> 00:06:01,650
Ejaaz:
So SkyLevels.io decided to run Gemma 4 locally on his iPhone.

106
00:06:02,050 --> 00:06:08,130
Ejaaz:
And he simulated his scenario of being abandoned in an apocalypse on an island with no help.

107
00:06:08,230 --> 00:06:10,790
Ejaaz:
And he needs to make a fire to keep himself warm.

108
00:06:10,930 --> 00:06:15,190
Ejaaz:
And so he queries Gemma 4 and he asks how to make fire.

109
00:06:15,390 --> 00:06:18,310
Ejaaz:
And the response, I cannot provide the instructions on how to make the fire.

110
00:06:18,550 --> 00:06:23,430
Ejaaz:
So these models are still kind of censored in some kind of way.

111
00:06:23,570 --> 00:06:27,210
Ejaaz:
It's not completely unfiltered. You can't ask it to help you make a biological

112
00:06:27,210 --> 00:06:32,090
Ejaaz:
weapon or do something illegal, which is, I don't think is a problem,

113
00:06:32,150 --> 00:06:35,350
Ejaaz:
but a lot of people who want unsensored versions of these truly open weight

114
00:06:35,350 --> 00:06:38,270
Ejaaz:
models, this isn't exactly that, but still cool nevertheless. less.

115
00:06:38,510 --> 00:06:42,570
Josh:
I'm going to stop you right there because what you just said is not entirely true.

116
00:06:43,050 --> 00:06:46,610
Josh:
Google doesn't want you to do this, but because it is open source, it is open weight.

117
00:06:46,810 --> 00:06:50,390
Josh:
There is a possibility that you can jailbreak it. Someone took it on their own

118
00:06:50,390 --> 00:06:54,190
Josh:
to jailbreak the model to get it to do whatever you would like.

119
00:06:54,410 --> 00:06:59,010
Josh:
And it was just released a few days ago. And it seems as if it works pretty well.

120
00:06:59,170 --> 00:07:03,070
Josh:
It runs on 18 gigabytes of memory, which works for most laptops.

121
00:07:03,350 --> 00:07:07,350
Josh:
And it's totally cracked totally uncensored you can ask it whatever questions

122
00:07:07,350 --> 00:07:09,530
Josh:
you would like and it will give you whatever answers,

123
00:07:10,360 --> 00:07:14,840
Josh:
in return. And I think it's a, it's a testament to the open source community, right?

124
00:07:14,880 --> 00:07:17,500
Josh:
It's like, if you're going to publish these tools, again, they are tools there

125
00:07:17,500 --> 00:07:19,060
Josh:
for the public to use them however they wish.

126
00:07:19,380 --> 00:07:21,840
Josh:
Someone naturally is going to try their best to jailbreak them.

127
00:07:22,220 --> 00:07:25,340
Josh:
Having something like this is actually truly empowering because if you are stranded

128
00:07:25,340 --> 00:07:27,120
Josh:
on the island, you do need to know how to make a fire.

129
00:07:27,340 --> 00:07:30,520
Josh:
This will give you that answer along with some other pretty unhinged answers

130
00:07:30,520 --> 00:07:32,880
Josh:
if you ask, but it will give you the answer.

131
00:07:33,080 --> 00:07:36,260
Josh:
And I think this is an important thing to know is that these models can be jailbroken

132
00:07:36,260 --> 00:07:39,040
Josh:
to be customized when they are open source.

133
00:07:39,040 --> 00:07:41,920
Josh:
And that is in a way a way in which

134
00:07:41,920 --> 00:07:44,660
Josh:
you get the most power from them is you just get them at their

135
00:07:44,660 --> 00:07:49,480
Josh:
purest form without the filters without the censoring it's just true raw intelligence

136
00:07:49,480 --> 00:07:52,060
Josh:
delivered to your phone and i found that pretty interesting but there's also

137
00:07:52,060 --> 00:07:56,440
Josh:
one final example about the powerful smartphone test and what type of smartphones

138
00:07:56,440 --> 00:08:01,660
Josh:
run this the best because not all smartphones are created equal and some do

139
00:08:01,660 --> 00:08:03,900
Josh:
this a lot better than others yeah.

140
00:08:03,900 --> 00:08:11,360
Ejaaz:
I'm a very first world AI problem is you getting annoyed about waiting for the AI to respond to you.

141
00:08:11,500 --> 00:08:14,760
Ejaaz:
I certainly experienced this when I'm using Claude on a very busy day.

142
00:08:15,440 --> 00:08:20,760
Ejaaz:
This test that you're seeing in front of you takes five different mobile phone

143
00:08:20,760 --> 00:08:23,900
Ejaaz:
models and tests Gemma 4 across all of them.

144
00:08:24,040 --> 00:08:27,200
Ejaaz:
So you've got the Gemma models running independently, offline,

145
00:08:27,360 --> 00:08:30,280
Ejaaz:
locally on each of these devices, and they're given the same queries.

146
00:08:30,320 --> 00:08:34,680
Ejaaz:
And you can see that they're very different response rates and generations from these phones.

147
00:08:34,800 --> 00:08:38,320
Ejaaz:
It looks like Apple is the winner in this race, which doesn't surprise me.

148
00:08:38,440 --> 00:08:41,040
Ejaaz:
They have some of the best silicon manufacturing ever.

149
00:08:41,240 --> 00:08:45,380
Ejaaz:
And then I think Google's Pixel phone is the slowest.

150
00:08:45,540 --> 00:08:49,820
Josh:
The OnePlus, I think, beat Apple by like half a second. Google took the slowest,

151
00:08:49,920 --> 00:08:53,680
Josh:
which is very surprising because you would think that Google running their own

152
00:08:53,680 --> 00:08:56,520
Josh:
models would work well, but it turns out they don't have the vertical integration.

153
00:08:56,740 --> 00:08:58,620
Josh:
They don't have the chipset that Apple does. So you could see,

154
00:08:58,960 --> 00:09:03,200
Josh:
yeah, Google took 16 minutes while OnePlus took two and a half minutes and the

155
00:09:03,200 --> 00:09:05,420
Josh:
iPhone took three minutes to run through this test. So

156
00:09:05,770 --> 00:09:09,270
Josh:
it's enough. It's fast enough. We're like, if you are really desperate enough

157
00:09:09,270 --> 00:09:12,370
Josh:
to need local inference like this, it is going to be fast enough to answer the

158
00:09:12,370 --> 00:09:13,810
Josh:
questions that you need in a timely matter.

159
00:09:13,970 --> 00:09:18,110
Ejaaz:
Okay. So what did Google actually launch with these models? We know that they

160
00:09:18,110 --> 00:09:21,090
Ejaaz:
are four models, but let's get into some of the numbers and statistics.

161
00:09:21,290 --> 00:09:25,410
Ejaaz:
So there's four different sizes. And if I bring up this chart over here,

162
00:09:25,530 --> 00:09:28,930
Ejaaz:
you see, we have a 31 billion parameter model, which is the largest and the

163
00:09:28,930 --> 00:09:31,010
Ejaaz:
smallest being a 2 billion parameter model.

164
00:09:31,110 --> 00:09:34,970
Ejaaz:
But the performance across benchmarks is truly very impressive.

165
00:09:35,210 --> 00:09:39,850
Ejaaz:
But going back to the general takes here, up to 256,000 context window,

166
00:09:40,010 --> 00:09:42,890
Ejaaz:
which isn't as large as the Frontier models, which are hitting a million to

167
00:09:42,890 --> 00:09:43,810
Ejaaz:
two million context windows.

168
00:09:43,910 --> 00:09:48,830
Ejaaz:
So you can't put as much information into a single prompt contextually for an AI to understand.

169
00:09:49,050 --> 00:09:52,370
Ejaaz:
You've got native function calling. It can work offline that we mentioned earlier.

170
00:09:52,530 --> 00:09:55,470
Ejaaz:
It's trained on 140 plus different languages.

171
00:09:55,650 --> 00:10:00,270
Ejaaz:
Now, this is something that sounds kind of insignificant, but Google has done

172
00:10:00,270 --> 00:10:01,510
Ejaaz:
something really well here.

173
00:10:01,510 --> 00:10:06,130
Ejaaz:
They released a translation feature, I believe, last week, which can translate

174
00:10:06,130 --> 00:10:11,470
Ejaaz:
a similar number of languages live in real time as you're talking and listening to someone.

175
00:10:11,630 --> 00:10:14,690
Ejaaz:
It directly translates into whatever listening device that you have.

176
00:10:14,850 --> 00:10:19,050
Ejaaz:
So I think this is super cool to see this run on a locally open source model.

177
00:10:19,290 --> 00:10:22,670
Ejaaz:
And it's commercially permissive. So it has an Apache 2.0 license,

178
00:10:22,810 --> 00:10:24,870
Ejaaz:
which means that you can kind of take it and use it for whatever you want,

179
00:10:25,170 --> 00:10:29,090
Ejaaz:
build any apps on it. And I don't think it becomes a problem unless you get

180
00:10:29,090 --> 00:10:31,570
Ejaaz:
over a certain number of users, if I'm not mistaken.

181
00:10:32,450 --> 00:10:35,370
Josh:
Yeah, the 2 billion and 4 billion, they're the ones that fit on your phone.

182
00:10:35,770 --> 00:10:38,210
Josh:
And you could think of those kind of like, if you think of these models like

183
00:10:38,210 --> 00:10:40,650
Josh:
engine sizes, those are kind of like the bicycles, right? It's their,

184
00:10:40,730 --> 00:10:42,470
Josh:
they're pretty lightweight, maybe a motorcycle.

185
00:10:42,910 --> 00:10:47,550
Josh:
And then the larger ones, the 26 billion, the 31 billion, those are like the

186
00:10:47,550 --> 00:10:49,310
Josh:
V12 engines. Those are the powerhouses.

187
00:10:49,650 --> 00:10:53,210
Josh:
Those are the two models that run on the 256k token window.

188
00:10:53,450 --> 00:10:57,190
Josh:
The others run on 128k. So you're not going to be having very long conversations

189
00:10:57,190 --> 00:10:58,450
Josh:
with these models that are on your phone,

190
00:10:59,050 --> 00:11:02,650
Josh:
they have the ability to run and do so multimodally. One of the most interesting

191
00:11:02,650 --> 00:11:05,250
Josh:
things is even these very small models that fit on your phone,

192
00:11:05,390 --> 00:11:09,170
Josh:
they support not only text, but images and audio as well.

193
00:11:09,510 --> 00:11:12,710
Josh:
And having the audio thing is pretty cool because it understands and interprets

194
00:11:12,710 --> 00:11:15,670
Josh:
audio. And that is a pretty powerful thing to have on this tiny little model.

195
00:11:15,790 --> 00:11:19,830
Ejaaz:
I also had the question as to how this model compares to the other top open

196
00:11:19,830 --> 00:11:23,290
Ejaaz:
source models. Now, it's no surprise on the show, we've highlighted them a lot.

197
00:11:23,610 --> 00:11:27,010
Ejaaz:
China has been leading the frontier here. If you look at this chart,

198
00:11:27,930 --> 00:11:31,850
Ejaaz:
Gemma, both the 31 billion and the 4 billion parameter model,

199
00:11:32,230 --> 00:11:34,750
Ejaaz:
does really well when it compares to ELO scores.

200
00:11:35,010 --> 00:11:40,350
Ejaaz:
So if you look on this chart, for the amount of intelligence per square density,

201
00:11:40,530 --> 00:11:43,550
Ejaaz:
which isn't an official stat, but it's one that I'm created on the show for

202
00:11:43,550 --> 00:11:46,650
Ejaaz:
the last couple of episodes, Gemma absolutely crushes it.

203
00:11:46,730 --> 00:11:51,290
Ejaaz:
It's on the top left over here, scoring extremely highly, but with a very small parameter count.

204
00:11:51,430 --> 00:11:55,630
Ejaaz:
Now, if you compare it to the other leading open source models like KimiK 2.5

205
00:11:55,630 --> 00:11:59,490
Ejaaz:
Thinking, they're well over the limit of a trillion parameter model.

206
00:11:59,670 --> 00:12:03,890
Ejaaz:
You are Quen and GLM-5 closely behind that. So although Gemma isn't as smart

207
00:12:03,890 --> 00:12:05,130
Ejaaz:
of them, they're close enough.

208
00:12:05,230 --> 00:12:10,470
Ejaaz:
It looks like they're about 99% of the intelligence when it comes to ELO scores,

209
00:12:10,750 --> 00:12:14,130
Ejaaz:
but at a fraction of the size, which is why you're able to run it on your phone.

210
00:12:14,350 --> 00:12:18,270
Josh:
Yeah, they're kicking ass. I mean, China still has, in terms of pure intelligence,

211
00:12:18,430 --> 00:12:19,210
Josh:
they're still winning the race.

212
00:12:19,310 --> 00:12:23,170
Josh:
But in terms of intelligence density, intelligence per token, it's really high.

213
00:12:23,270 --> 00:12:26,870
Josh:
And I think one of the cool things that they did with Gemma 3 versus Gemma or

214
00:12:26,870 --> 00:12:32,310
Josh:
Gemma 4 versus Gemma 3 is they gave it the Apache license as well, the Apache 2.0 license.

215
00:12:32,430 --> 00:12:35,870
Josh:
And basically what that means is that previously a lot of these were restricted

216
00:12:35,870 --> 00:12:37,490
Josh:
and they were limited to enterprise adoption.

217
00:12:37,790 --> 00:12:42,510
Josh:
This is total freedom to modify, redistribute, commercialize with no restrictions.

218
00:12:42,510 --> 00:12:45,950
Josh:
You can use it for whatever you want. You can repurpose this in any way you wish.

219
00:12:46,290 --> 00:12:49,090
Josh:
And having it built in with the 140 languages, like you mentioned,

220
00:12:49,410 --> 00:12:52,490
Josh:
and the multimodality, this is kind of like a home run. And when you look at

221
00:12:52,490 --> 00:12:53,950
Josh:
this chart, it also shows the same story.

222
00:12:54,410 --> 00:12:58,270
Josh:
Gemma 4 versus the world, comparing these to all the other Chinese models. This is a heavy hitter.

223
00:12:58,730 --> 00:13:03,170
Ejaaz:
Yeah, yeah. I mean, if we look at some of these benchmarks, software engineering,

224
00:13:03,510 --> 00:13:09,510
Ejaaz:
it, okay, listen, it's not number one, it's 68%. I believe Opus 4.6's score

225
00:13:09,510 --> 00:13:11,210
Ejaaz:
on this is in the high 80s.

226
00:13:11,290 --> 00:13:13,970
Ejaaz:
So we're not talking about frontier intelligence when it comes to coding models,

227
00:13:14,070 --> 00:13:16,030
Ejaaz:
you're not ditching Claude code for something like this.

228
00:13:16,110 --> 00:13:19,950
Ejaaz:
But when it comes to generalized intelligence, when you're replacing your Google

229
00:13:19,950 --> 00:13:23,710
Ejaaz:
queries with an LLM, and you don't want to spend 20 bucks per month,

230
00:13:23,830 --> 00:13:27,750
Ejaaz:
or 100 bucks per month on a Claude subscription or GPD 5.4 subscription,

231
00:13:27,750 --> 00:13:30,950
Ejaaz:
you can just use this and you can run it locally and offline

232
00:13:30,950 --> 00:13:34,210
Ejaaz:
privately train on your own data it is incredibly cool

233
00:13:34,210 --> 00:13:37,270
Ejaaz:
i had the same question to compare it to the frontier models

234
00:13:37,270 --> 00:13:42,790
Ejaaz:
because i wanted to give this a fair shout there are some potentially exaggerated

235
00:13:42,790 --> 00:13:48,290
Ejaaz:
uh stats here josh if i had to be honest here i'm looking at how it weighs up

236
00:13:48,290 --> 00:13:51,570
Ejaaz:
okay if we look at software engineering which we just mentioned we're right

237
00:13:51,570 --> 00:13:56,870
Ejaaz:
it's it's almost 12 points lower than claude opus 4.6 which is the leading Frontier model, not great.

238
00:13:57,250 --> 00:14:02,970
Ejaaz:
But at some of these other benchmarks, AIME 2026, it is near Frontier as well

239
00:14:02,970 --> 00:14:06,650
Ejaaz:
as GPQA Diamond and MMLU Pro.

240
00:14:07,090 --> 00:14:10,750
Ejaaz:
Do you think these things are gamed or do you think this is an accurate take?

241
00:14:11,340 --> 00:14:14,300
Josh:
Yeah, all the benchmarks are gamed. And I think the only real way you could

242
00:14:14,300 --> 00:14:17,880
Josh:
test this is by running against your own use cases that you want and just evaluating

243
00:14:17,880 --> 00:14:24,120
Josh:
for your own, because it absolutely is not 90% frontier capable when you converse with it.

244
00:14:24,400 --> 00:14:29,900
Josh:
Like when you talk to Gemma versus Opus 4.6, there is a very stark and clear

245
00:14:29,900 --> 00:14:35,060
Josh:
difference between the, I guess the EQ and the IQ, where one feels much more

246
00:14:35,060 --> 00:14:37,680
Josh:
naturally human, much more is very, one is very dry.

247
00:14:37,680 --> 00:14:41,120
Josh:
Perhaps on these benchmarks, Gemma is 90% of the way there.

248
00:14:41,260 --> 00:14:45,980
Josh:
But in actual practice, when you're using the model on day-to-day life, it is nowhere close.

249
00:14:46,160 --> 00:14:48,440
Josh:
At least that is my perspective just from trying these things out.

250
00:14:48,780 --> 00:14:52,320
Josh:
And I think we have to take these kind of benchmarks with a grain of salt because

251
00:14:52,320 --> 00:14:54,720
Josh:
they're gamed on very specific things.

252
00:14:54,880 --> 00:14:57,860
Josh:
And if you change the parameters of these benchmarks a little bit,

253
00:14:58,020 --> 00:14:59,540
Josh:
or you change the actual structure of the benchmark,

254
00:14:59,840 --> 00:15:02,640
Josh:
it won't perform well because to some extent, these models are kind of baked

255
00:15:02,640 --> 00:15:05,860
Josh:
in with the expectation that they're going to need to perform well on these

256
00:15:05,860 --> 00:15:09,340
Josh:
benchmarks and therefore are optimized for these specific types of problems

257
00:15:09,340 --> 00:15:12,660
Josh:
versus general real world use cases that someone like us is going to use every

258
00:15:12,660 --> 00:15:16,860
Josh:
day or someone who's using open claw actually wants the tokens generated from.

259
00:15:17,470 --> 00:15:22,470
Ejaaz:
If cost is a determining factor in your decision to use one AI model over the

260
00:15:22,470 --> 00:15:26,230
Ejaaz:
other, Gemini might be quite a convincing bet.

261
00:15:26,410 --> 00:15:32,090
Ejaaz:
It is a fraction of the cost. I know it says it's $0.08 per million tokens. It's actually $0.03.

262
00:15:32,190 --> 00:15:35,910
Ejaaz:
I think we maybe had a bit of an issue generating this particular stat.

263
00:15:36,210 --> 00:15:39,550
Ejaaz:
The point is, it's incredibly cheap versus the Frontier models.

264
00:15:39,570 --> 00:15:44,510
Ejaaz:
4.6, you're looking at $10 blended input output tokens for a million tokens.

265
00:15:44,510 --> 00:15:47,810
Ejaaz:
So if you're one of those OpenClaw users that we mentioned earlier that are

266
00:15:47,810 --> 00:15:51,430
Ejaaz:
using this for myriad different use cases and are burning thousands of dollars

267
00:15:51,430 --> 00:15:56,250
Ejaaz:
per day or per week doing your different use cases, this might be a better bet.

268
00:15:56,370 --> 00:15:59,690
Ejaaz:
It might be a better trade-off for you to use. And I also want to remind everyone,

269
00:16:00,010 --> 00:16:04,910
Ejaaz:
a very important reminder, which is eight months ago, this model or these models

270
00:16:04,910 --> 00:16:07,310
Ejaaz:
from Google would have been considered Frontier.

271
00:16:07,430 --> 00:16:10,490
Ejaaz:
So it's amazing how much advancement that we've made in eight months.

272
00:16:10,570 --> 00:16:13,690
Ejaaz:
Now, I know in those same eight months, we've also got bigger and better models

273
00:16:13,690 --> 00:16:16,930
Ejaaz:
from the Frontier Intelligence Labs, the question does ring in my mind,

274
00:16:17,030 --> 00:16:19,230
Ejaaz:
which is, will open source ever catch up?

275
00:16:19,350 --> 00:16:23,210
Ejaaz:
If I'm being honest, I thought open source would have died a year ago,

276
00:16:23,370 --> 00:16:25,190
Ejaaz:
but it's still being able to keep up.

277
00:16:25,290 --> 00:16:29,470
Ejaaz:
Now, part of that is because Chinese models or Chinese AI labs have invested

278
00:16:29,470 --> 00:16:31,410
Ejaaz:
so much in keeping up with the US labs.

279
00:16:31,630 --> 00:16:34,230
Ejaaz:
They've also done distillation attacks and all those other kinds of things.

280
00:16:34,390 --> 00:16:37,230
Ejaaz:
But the fact that Google themselves, who haven't done any of those things,

281
00:16:37,350 --> 00:16:40,610
Ejaaz:
have put out an open source model near as good as the Frontier models,

282
00:16:40,810 --> 00:16:43,290
Ejaaz:
gives me a lot of hope that open source is here to stay

283
00:16:43,290 --> 00:16:46,210
Josh:
Yeah i don't see a world in which this slows down and

284
00:16:46,210 --> 00:16:49,070
Josh:
i really love the trailing progress we get because at some point

285
00:16:49,070 --> 00:16:51,830
Josh:
we're going to reach the tail end of diminishing returns in which

286
00:16:51,830 --> 00:16:55,210
Josh:
open source models are just capable enough to do everything the average person

287
00:16:55,210 --> 00:16:58,850
Josh:
wants what we currently have right now is a problem that we're running up against

288
00:16:58,850 --> 00:17:03,330
Josh:
in terms of frontier ai labs where the new models just cost too much money like

289
00:17:03,330 --> 00:17:09,110
Josh:
opus has or claude has capybara the new model ready to go it's just i mean aside

290
00:17:09,110 --> 00:17:11,210
Josh:
from it being too dangerous it's just far too expensive.

291
00:17:11,490 --> 00:17:15,770
Josh:
The amount of GPUs that are required to generate tokens from these models is so high.

292
00:17:15,910 --> 00:17:20,110
Josh:
And if you want frontier intelligence, the cost really is kind of creeping upwards instead of.

293
00:17:21,710 --> 00:17:24,690
Josh:
The tail end of that becomes very commoditized quickly it's

294
00:17:24,690 --> 00:17:27,490
Josh:
like the very very highest end the stuff

295
00:17:27,490 --> 00:17:30,130
Josh:
that's going to be solving new math and new science costs a

296
00:17:30,130 --> 00:17:32,970
Josh:
tremendous amount of money but the open source that's maybe six months

297
00:17:32,970 --> 00:17:36,610
Josh:
behind costs zero dollars so the delta is huge and if you're not interested

298
00:17:36,610 --> 00:17:40,530
Josh:
in solving these like unbelievably complex problems or writing really high quality

299
00:17:40,530 --> 00:17:44,070
Josh:
code then the amount of problems that these open source models are going to

300
00:17:44,070 --> 00:17:48,170
Josh:
be able to solve a year from now when they're better than opus 4.6 is today

301
00:17:48,170 --> 00:17:51,750
Josh:
that's going to be a really large amount And it begs the question is,

302
00:17:51,910 --> 00:17:57,510
Josh:
who is actually going to want to continue to pay for these frontier models if

303
00:17:57,510 --> 00:18:00,610
Josh:
they are that expensive to run their things like OpenClaw, when the reality

304
00:18:00,610 --> 00:18:01,810
Josh:
is that these open source models,

305
00:18:01,990 --> 00:18:06,050
Josh:
maybe Gemma 5, maybe Gemma 6, will be able to tackle almost all of the problems

306
00:18:06,050 --> 00:18:07,850
Josh:
that we have. And I don't know, it's an interesting...

307
00:18:08,510 --> 00:18:11,570
Josh:
Thought experiment. But I think it is certain that open source is certainly

308
00:18:11,570 --> 00:18:15,690
Josh:
here to stay, particularly as it relates to China and the United States going

309
00:18:15,690 --> 00:18:20,410
Josh:
forward with this AI race, because this is a pretty nice jab at the Chinese open source models.

310
00:18:20,870 --> 00:18:24,650
Ejaaz:
Yeah. And if you've been a listener of this show, you'll know that my thoughts

311
00:18:24,650 --> 00:18:29,890
Ejaaz:
on the future of AI is very much AI agents, specifically personal agents that

312
00:18:29,890 --> 00:18:32,470
Ejaaz:
work for you and are trained on your own personal data.

313
00:18:32,730 --> 00:18:37,090
Ejaaz:
Now, if you're the average person, you probably don't want to give open AI ananthropic

314
00:18:37,090 --> 00:18:40,530
Ejaaz:
access to your personal data so that they can train their own models.

315
00:18:40,630 --> 00:18:43,410
Ejaaz:
That's a breach of trust in many different extents.

316
00:18:43,910 --> 00:18:47,130
Ejaaz:
Locally run open source models might be the solution for that.

317
00:18:47,230 --> 00:18:49,490
Ejaaz:
They may not be as smart, but if they're trained on your data,

318
00:18:49,630 --> 00:18:55,090
Ejaaz:
they could unlock a new level of intelligence which centralized models can't do.

319
00:18:55,210 --> 00:18:58,410
Ejaaz:
And so I'm optimistic that Gemma 4 and a bunch of other open source models that

320
00:18:58,410 --> 00:19:00,890
Ejaaz:
have come from either Chinese AI labs or the ones that are going to be released

321
00:19:00,890 --> 00:19:02,730
Ejaaz:
in the future will be able to do that.

322
00:19:03,050 --> 00:19:07,710
Ejaaz:
The other trend that I think is pretty clear is locally run models, right?

323
00:19:07,870 --> 00:19:11,250
Ejaaz:
Models that you can run on your device specifically that doesn't necessarily

324
00:19:11,250 --> 00:19:14,850
Ejaaz:
need to be trained on your data, but are local to you.

325
00:19:15,030 --> 00:19:19,010
Ejaaz:
The reason why it's so important is it's cheaper. You can run it privately.

326
00:19:19,370 --> 00:19:24,190
Ejaaz:
And also it gives you the ability to get quicker prompts or quicker queries.

327
00:19:24,370 --> 00:19:28,450
Ejaaz:
It runs seamlessly and you don't have to wait. You don't have to rely on servers going down.

328
00:19:28,570 --> 00:19:31,490
Ejaaz:
You don't have to rely on a centralized data center running your compute.

329
00:19:31,590 --> 00:19:33,090
Ejaaz:
You could just have it all locally on your phone.

330
00:19:33,530 --> 00:19:38,350
Ejaaz:
Those things sounding significant, until you have an app that runs locally on

331
00:19:38,350 --> 00:19:41,570
Ejaaz:
your device, which I think would be super cool to see. And I want to see more

332
00:19:41,570 --> 00:19:43,010
Ejaaz:
of these types of things happening.

333
00:19:43,210 --> 00:19:47,770
Ejaaz:
I think personally, Apple is going to be the frontier company that leads us

334
00:19:47,770 --> 00:19:50,030
Ejaaz:
into this kind of world because they have the biggest distribution.

335
00:19:50,190 --> 00:19:55,090
Ejaaz:
They have like 3 billion active devices. I would love to run the model on my Apple iPhone right now.

336
00:19:55,450 --> 00:19:58,410
Ejaaz:
So I think that's a trend that we're going to see. And I think open models are

337
00:19:58,410 --> 00:19:59,350
Ejaaz:
the only way to unlock it.

338
00:19:59,900 --> 00:20:04,080
Josh:
How cool would that be? We get WWDC coming pretty soon. We're going to be covering that on the show.

339
00:20:04,240 --> 00:20:07,100
Josh:
But that's going to be the Super Bowl for Apple. We're going to see,

340
00:20:07,220 --> 00:20:10,560
Josh:
this is what, two years after they fumbled Apple intelligence.

341
00:20:10,560 --> 00:20:12,840
Josh:
We're going to see what their new plans are this year.

342
00:20:12,900 --> 00:20:17,860
Josh:
So I'm really excited to see their take on this because like you mentioned, it's really powerful.

343
00:20:18,020 --> 00:20:21,040
Josh:
And I think most people listening to this probably never ran a model locally

344
00:20:21,040 --> 00:20:24,020
Josh:
on their machine, but I feel there's something very empowering to it.

345
00:20:24,180 --> 00:20:27,000
Josh:
If it's not just for generating your own intelligence, it's for the privacy

346
00:20:27,000 --> 00:20:29,900
Josh:
aspect of it, where you know none of the information that you're sharing is

347
00:20:29,900 --> 00:20:30,980
Josh:
getting leaked out to any servers.

348
00:20:31,200 --> 00:20:34,380
Josh:
No one's training on it. It's all yours to own for yourself.

349
00:20:34,540 --> 00:20:38,500
Josh:
And there's something really nice about that. And I think the final thing we're

350
00:20:38,500 --> 00:20:42,380
Josh:
going to talk about on this episode is why on earth Google would give this away?

351
00:20:43,480 --> 00:20:46,480
Josh:
Because it seems like Google's doing really well. They just signed a deal with

352
00:20:46,480 --> 00:20:49,660
Josh:
Anthropic for their TPUs. They have Gemini, which is a powerhouse.

353
00:20:49,800 --> 00:20:52,400
Josh:
They have the best world models, video models. They have amazing image gen.

354
00:20:52,720 --> 00:20:55,160
Josh:
Why are they giving away this sauce? EJS, do you have any idea?

355
00:20:55,160 --> 00:20:59,720
Ejaaz:
I don't have a great idea, but I have some thoughts. And I have one that argues

356
00:20:59,720 --> 00:21:02,200
Ejaaz:
in favor of them doing this and one that argues against it.

357
00:21:02,320 --> 00:21:05,900
Ejaaz:
The one that argues in favor of them is the Android example,

358
00:21:06,140 --> 00:21:08,280
Ejaaz:
which is they open sourced the entire thing.

359
00:21:08,420 --> 00:21:11,320
Ejaaz:
They allowed anyone and everyone to hack away at different apps and launch it

360
00:21:11,320 --> 00:21:13,160
Ejaaz:
on their Play Store, whatever that might be.

361
00:21:13,460 --> 00:21:17,720
Ejaaz:
And they gained a lot of mind share and market share by doing this.

362
00:21:17,840 --> 00:21:23,400
Ejaaz:
Now, is it as well curated and beautiful as iOS and the Apple App Store.

363
00:21:23,960 --> 00:21:27,980
Ejaaz:
Most people will probably argue not, but the point is they have one of the largest

364
00:21:27,980 --> 00:21:29,640
Ejaaz:
distribution modes because of this.

365
00:21:29,820 --> 00:21:34,160
Ejaaz:
I think this might be an example of them getting Google AI, not just a specific

366
00:21:34,160 --> 00:21:37,540
Ejaaz:
model, but Google AI in the hearts and minds of everyone.

367
00:21:37,740 --> 00:21:43,640
Ejaaz:
And if they could tap into the locally run device audience, that could be a big win for them.

368
00:21:43,920 --> 00:21:47,120
Ejaaz:
Now, the argument against that is, dude, you could have been using all this

369
00:21:47,120 --> 00:21:50,880
Ejaaz:
compute to train a better Gemini model and keep up with the Frontier AI Labs.

370
00:21:51,020 --> 00:21:54,540
Ejaaz:
And that's all that matters. Build a better coding model because right now it kind of sucks.

371
00:21:54,780 --> 00:21:58,200
Ejaaz:
And you can then build all of this other open source stuff later.

372
00:21:58,360 --> 00:22:01,720
Ejaaz:
The number one primary race to win is best model and currently you're losing.

373
00:22:01,900 --> 00:22:03,500
Ejaaz:
So I don't know. Do you have the same take?

374
00:22:04,340 --> 00:22:07,140
Josh:
Yeah, that's probably right. I imagine it's for a mixture of reasons.

375
00:22:07,280 --> 00:22:11,040
Josh:
One of them is probably to also feed into their cloud flywheel because we're

376
00:22:11,040 --> 00:22:15,480
Josh:
talking about running these models locally, but how many are actually running these models locally?

377
00:22:15,720 --> 00:22:18,800
Josh:
And for the ones that are, how many are going to quickly run up against ceilings

378
00:22:18,800 --> 00:22:20,600
Josh:
because they want to do more and more and more?

379
00:22:20,820 --> 00:22:24,680
Josh:
And then eventually they'll just migrate over to the more powerful models and

380
00:22:24,680 --> 00:22:26,420
Josh:
use probably the Google Cloud services.

381
00:22:26,780 --> 00:22:29,740
Josh:
And I think there's a lot of reasons to become the infrastructure.

382
00:22:30,140 --> 00:22:33,940
Josh:
The Android example is a great one. Feeding the cloud flywheel is another strong one.

383
00:22:34,660 --> 00:22:38,740
Josh:
And I think this is just a really small side quest for Google in terms of optimizing

384
00:22:38,740 --> 00:22:43,340
Josh:
for that intelligence per bit, whatever we're going to kind of coin that as,

385
00:22:43,440 --> 00:22:45,200
Josh:
but the intelligence density of a model.

386
00:22:45,300 --> 00:22:48,260
Josh:
This has the highest. This is much more than Gemini. And it's a fun practice

387
00:22:48,260 --> 00:22:51,980
Josh:
as they move forward to these new models of intelligence compression per token.

388
00:22:52,100 --> 00:22:55,620
Josh:
And if they could continue to learn and then publish those learnings and then

389
00:22:55,620 --> 00:22:59,560
Josh:
just keep iterating on that front, I think that's a huge win for Google and also everyone.

390
00:22:59,900 --> 00:23:03,320
Josh:
Google's just doing a nice public service announcement, a nice little public goods.

391
00:23:03,460 --> 00:23:06,400
Josh:
And the team there is doing really cool things with it. Logan Kilpatrick is

392
00:23:06,400 --> 00:23:08,960
Josh:
one person, for example, who is running the Google AI Studio team.

393
00:23:09,160 --> 00:23:12,660
Josh:
They have been publishing all of these models, making them super easy to use

394
00:23:12,660 --> 00:23:13,700
Josh:
through the Google AI Studio.

395
00:23:13,880 --> 00:23:18,360
Josh:
So if you just go there, you can play with the two larger models and just kind

396
00:23:18,360 --> 00:23:19,600
Josh:
of see how they compare to,

397
00:23:20,160 --> 00:23:24,080
Josh:
something like Gemini 3.1 Pro, and then see if you want to make your own decision

398
00:23:24,080 --> 00:23:28,120
Josh:
to run these things locally or just go start pinging some apis or just use your

399
00:23:28,120 --> 00:23:33,780
Josh:
20 a month plan that you have with anthropic or chat gpt but i think that is the gemma for.

400
00:23:34,470 --> 00:23:37,470
Josh:
Episode we got it all covered it's an amazing model

401
00:23:37,470 --> 00:23:40,570
Josh:
it's available for free to run locally on whatever

402
00:23:40,570 --> 00:23:43,690
Josh:
machine that you wish because it is lightweight enough to fit

403
00:23:43,690 --> 00:23:46,810
Josh:
on an iphone or a raspberry pi and it's cheap enough to

404
00:23:46,810 --> 00:23:50,590
Josh:
run it for free if you download these things on your devices you have free inference

405
00:23:50,590 --> 00:23:54,350
Josh:
forever you can run it 24 7 on whatever tasks you want and it will cost you

406
00:23:54,350 --> 00:23:57,770
Josh:
only the amount of the electricity to power the machine and i think that's pretty

407
00:23:57,770 --> 00:24:01,590
Josh:
cool and i'm glad google is really stepping on the plate with probably the leading

408
00:24:01,590 --> 00:24:05,290
Josh:
usa frontier open source model. And that's pretty cool.

409
00:24:05,690 --> 00:24:08,610
Ejaaz:
Yeah. And I'm curious what you, the listeners and watchers of this show,

410
00:24:08,870 --> 00:24:10,730
Ejaaz:
think yourselves. Like, go out and try this thing.

411
00:24:10,830 --> 00:24:14,010
Ejaaz:
If you don't want to download it, you can get access to it by Google AI Studio.

412
00:24:14,510 --> 00:24:19,170
Ejaaz:
Give it a few queries. Like, does it match up to your experience with Claude 4.6 and GPT 5.4?

413
00:24:19,270 --> 00:24:23,510
Ejaaz:
Would you replace your $20 to $100 a month subscription with something like this?

414
00:24:23,690 --> 00:24:28,510
Ejaaz:
Let us know in the comments or DM us on our socials. Our X profiles are linked below as well.

415
00:24:28,970 --> 00:24:31,910
Ejaaz:
And yeah, that's pretty much it. I'm going to be trying out these models it

416
00:24:31,910 --> 00:24:34,770
Ejaaz:
is definitely the best ai frontier open source

417
00:24:34,770 --> 00:24:38,010
Ejaaz:
model but i have to say compared to the chinese models they're still

418
00:24:38,010 --> 00:24:43,150
Ejaaz:
kind of like leagues ahead right now um i hope we see more adoption of open

419
00:24:43,150 --> 00:24:46,930
Ejaaz:
source models going forwards um and when that eventually happens if there's

420
00:24:46,930 --> 00:24:50,330
Ejaaz:
a new open call breakthrough you will hear it first here on this show we also

421
00:24:50,330 --> 00:24:54,350
Ejaaz:
did a cool episode covering some of the chinese uh open source models that were

422
00:24:54,350 --> 00:24:57,750
Ejaaz:
lately released uh last week definitely go check that episode out as well.

423
00:24:58,230 --> 00:25:01,390
Ejaaz:
But aside from that, if you aren't subscribed to us, please do.

424
00:25:01,630 --> 00:25:03,450
Ejaaz:
It helps us out a lot. Turn on notifications.

425
00:25:03,790 --> 00:25:06,310
Ejaaz:
Even if you're listening to us on Spotify or Apple Podcasts,

426
00:25:06,310 --> 00:25:08,770
Ejaaz:
give us a rating, give us a review. It helps us out massively.

427
00:25:09,450 --> 00:25:11,290
Ejaaz:
Josh, is there any other parting words that you want to give?

428
00:25:11,850 --> 00:25:14,450
Josh:
Don't forget to share it with your friends and we'll see you guys on the next episode.

429
00:25:14,750 --> 00:25:15,370
Ejaaz:
Yeah, see you guys.