1
00:00:00,080 --> 00:00:04,480
Welcome back to In the Loop and 
after what feels like years of 

2
00:00:04,480 --> 00:00:10,160
waiting, GPT 5 is finally here. 
So about a week on opinion is 

3
00:00:10,160 --> 00:00:12,360
quite divided. 
You've got many people saying 

4
00:00:12,360 --> 00:00:15,680
this is revolutionary, but just 
as many people saying it's 

5
00:00:15,680 --> 00:00:18,160
disappointing. 
You had on the launch event open

6
00:00:18,200 --> 00:00:22,040
AI showing bar charts that were 
just straight out lies. 

7
00:00:22,200 --> 00:00:25,240
And you have prediction markets 
betting against open AI just 

8
00:00:25,240 --> 00:00:28,160
minutes after their launch. 
So if you're sat here thinking, 

9
00:00:28,720 --> 00:00:32,280
I have no idea what's going on, 
well, that's probably the only 

10
00:00:32,280 --> 00:00:33,840
reasonable response you could 
have. 

11
00:00:34,040 --> 00:00:36,160
In today's episode, you're going
to learn everything that is 

12
00:00:36,160 --> 00:00:39,520
worth knowing about this new 
model release, what's new, 

13
00:00:39,520 --> 00:00:42,920
what's improved and what was 
just a lot of good marketing. 

14
00:00:43,080 --> 00:00:45,360
We've spent many hours 
collecting all the research and 

15
00:00:45,360 --> 00:00:49,400
reviews so you don't have to. 
This is in the leap of Jack 

16
00:00:49,400 --> 00:00:51,320
Horton. 
I hope you enjoyed the show. 

17
00:01:03,040 --> 00:01:05,840
So before I get into the 
episode, I think it's important 

18
00:01:05,840 --> 00:01:09,920
to set the scene and add just a 
little bit of context to this 

19
00:01:09,920 --> 00:01:13,560
model release. 
For the past year, we've been 

20
00:01:13,560 --> 00:01:17,720
living in basically model chaos.
It feels like Open AI has been 

21
00:01:17,720 --> 00:01:20,440
throwing spaghetti at the wall 
when it comes to language 

22
00:01:20,440 --> 00:01:24,840
models. 
We've had GPT 44.5, the O series

23
00:01:24,840 --> 00:01:28,440
with different specialised 
models, and much of it felt 

24
00:01:28,440 --> 00:01:31,520
sometimes like they were trying 
to justify a $200 per month 

25
00:01:31,720 --> 00:01:33,440
price tag for their pro 
membership. 

26
00:01:33,640 --> 00:01:36,560
And people were often so 
confused about which model they 

27
00:01:36,560 --> 00:01:39,080
were going to use for this 
specific task of theirs that 

28
00:01:39,640 --> 00:01:41,400
they had to even release white 
papers on it. 

29
00:01:41,560 --> 00:01:45,360
And meanwhile, Anthropic has 
been completely eating away at 

30
00:01:45,360 --> 00:01:49,600
their enterprise market. 
So particularly in coding, we've

31
00:01:49,600 --> 00:01:53,200
covered this in a number of 
episodes, but called code and 

32
00:01:53,560 --> 00:01:56,920
Anthropic Clod language model 
has become the go to choice for 

33
00:01:56,920 --> 00:01:59,280
developers across the world. 
And I guess the general take 

34
00:01:59,280 --> 00:02:02,120
away has been quite clear, which
is most people would go to 

35
00:02:02,120 --> 00:02:05,600
ChatGPT for general use. 
But for most people, if you're 

36
00:02:05,600 --> 00:02:08,720
going to go do a coding task, 
you're going to go and use clod.

37
00:02:08,880 --> 00:02:13,240
And in this context frames the 
release of GPT 5 really well. 

38
00:02:13,320 --> 00:02:17,760
This release wasn't just a 
better model, it was all about 

39
00:02:17,760 --> 00:02:21,480
cleaning up this chaotic mess of
different language models and 

40
00:02:21,480 --> 00:02:24,440
trying to take back market share
that they've lost over the last,

41
00:02:24,840 --> 00:02:26,920
I'd say 69 and 12 months to 
clock. 

42
00:02:27,080 --> 00:02:28,720
Now with that being said, let's 
get into it. 

43
00:02:28,840 --> 00:02:30,760
And I think it's going to be 
really helpful to start with a 

44
00:02:31,040 --> 00:02:35,080
high level overview of some of 
Chachi BT5's new features and 

45
00:02:35,080 --> 00:02:36,680
abilities. 
So one of the first things 

46
00:02:36,680 --> 00:02:39,280
you're going to notice since 
using the new Chachi BT5 model 

47
00:02:39,280 --> 00:02:41,240
is that it's called a hybrid 
model. 

48
00:02:41,320 --> 00:02:43,080
And the reason this is 
noticeable is because it's got 

49
00:02:43,080 --> 00:02:46,880
both thinking and non thinking 
versions built into the exact 

50
00:02:46,880 --> 00:02:48,600
same model. 
So previously if you asked a 

51
00:02:48,600 --> 00:02:52,000
question, unless you specified 
the O series or the thinking 

52
00:02:52,000 --> 00:02:55,760
model, you wouldn't have any 
reasoning and and thinking in 

53
00:02:55,760 --> 00:02:57,200
any of the responses that you 
get from it. 

54
00:02:57,240 --> 00:02:59,480
You'd have to specifically 
choose a reasoning model. 

55
00:02:59,720 --> 00:03:02,520
And for those on a free tier, 
you've only got a limited number

56
00:03:02,520 --> 00:03:04,760
of queries on that. 
And in fact, initially, they 

57
00:03:04,760 --> 00:03:07,840
actually called to deprecate 
many of their former models, so 

58
00:03:08,160 --> 00:03:12,360
they completely took away 
Chachi, BT 44.1, the entire O 

59
00:03:12,360 --> 00:03:15,160
series of models. 
There's been such a big uprising

60
00:03:15,160 --> 00:03:18,360
about this that Sam Altman's 
ended up releasing what is 

61
00:03:18,360 --> 00:03:22,200
basically a small essay of 
tweets and open AI, actually 

62
00:03:22,200 --> 00:03:26,000
bringing back many of the models
that people loved into the 

63
00:03:26,480 --> 00:03:28,760
subscription package. 
For those paying $200 a month, 

64
00:03:28,880 --> 00:03:32,280
this could be an episode on its 
own just due to the amount of, I

65
00:03:32,280 --> 00:03:35,760
guess, emotional attachment that
people have clearly had to some 

66
00:03:35,760 --> 00:03:38,000
of these models and the fact 
that it was then taken away from

67
00:03:38,000 --> 00:03:41,920
them caused real upset online. 
Another new addition is that 

68
00:03:42,160 --> 00:03:45,080
you're going to now be able to 
short circuit it's thinking when

69
00:03:45,080 --> 00:03:47,400
it goes into a thinking loop for
a question that to be honest, 

70
00:03:47,400 --> 00:03:49,320
was quite simple. 
It's also going to be available 

71
00:03:49,320 --> 00:03:51,680
to everybody, so including all 
free users. 

72
00:03:51,920 --> 00:03:55,480
I think they really learned from
the Deep Sea Saga last year that

73
00:03:55,480 --> 00:03:58,560
everyone must experience their 
best models. 

74
00:03:58,720 --> 00:04:01,480
And there are also three 
versions of Chat Chi BT5. 

75
00:04:01,760 --> 00:04:04,160
We have the standard one, the 
Mini 1 and the Nano. 

76
00:04:05,080 --> 00:04:07,760
So if you then hit usage limits 
on one model, it will 

77
00:04:07,760 --> 00:04:09,200
automatically switch to another 
model. 

78
00:04:09,400 --> 00:04:12,800
Also, it's a massive context 
window, which is for me a huge 

79
00:04:13,000 --> 00:04:14,880
improvement. 
It's context window is now 

80
00:04:14,880 --> 00:04:18,680
400,000 tokens, which is larger 
than all previous models. 

81
00:04:18,680 --> 00:04:22,560
And if I put that into context, 
that's about 300,000 words. 

82
00:04:23,200 --> 00:04:25,800
And again, if you listen to the 
episode on context engineering a

83
00:04:25,800 --> 00:04:30,440
few weeks ago, context is the 
ability for the agent to have a 

84
00:04:30,440 --> 00:04:32,560
set of information when you ask 
it a question. 

85
00:04:33,320 --> 00:04:37,160
So it could be a long part of a 
copy and paste of information. 

86
00:04:37,200 --> 00:04:40,600
In document terms, that's about 
600 to 800 pages of text it can 

87
00:04:40,600 --> 00:04:44,040
keep in its short term memory 
when you're having a 

88
00:04:44,040 --> 00:04:46,200
conversation with it. 
And previous models were about 

89
00:04:46,200 --> 00:04:49,520
128,000 tokens, so that's 96,000
words. 

90
00:04:49,640 --> 00:04:53,240
So that jumped from 96,000 words
of information it can store in 

91
00:04:53,240 --> 00:04:55,960
its memory when you're having a 
conversation with it to now. 

92
00:04:55,960 --> 00:05:00,040
300,000 is a huge, huge leap. 
There's nothing more than more 

93
00:05:00,040 --> 00:05:03,480
annoying than putting 
information into a model and 

94
00:05:03,720 --> 00:05:05,880
hitting a wall because of a 
context of it. 

95
00:05:06,080 --> 00:05:09,880
Another huge addition has been 
coding and front end creation 

96
00:05:10,520 --> 00:05:13,560
just using ChatGPT. 
They've clearly gone all out to 

97
00:05:13,560 --> 00:05:15,640
also try and take Liverpool's 
market share. 

98
00:05:16,400 --> 00:05:18,920
It now has a really good 
understanding of spacing and 

99
00:05:18,920 --> 00:05:21,560
typography and even white space 
and design. 

100
00:05:21,800 --> 00:05:26,480
You can vibe code entire user 
interface really, really simply.

101
00:05:26,480 --> 00:05:29,520
And in fact, in the actual press
release, which I really highly 

102
00:05:29,520 --> 00:05:32,400
recommend you go and check out, 
they provide a load of prompts 

103
00:05:32,400 --> 00:05:35,120
that can generate really cool 
applications. 

104
00:05:35,360 --> 00:05:38,240
And this is the exact reason 
that Sam Altman recently came 

105
00:05:38,240 --> 00:05:41,560
out and said that this is the 
fast fashion era of technology. 

106
00:05:41,640 --> 00:05:43,680
We'll get into this a little bit
more throughout the episode, but

107
00:05:44,040 --> 00:05:47,120
you want to keep it top of mind.
They've also really lowered the 

108
00:05:47,120 --> 00:05:49,080
barrier of price. 
It's now much, much cheaper. 

109
00:05:49,080 --> 00:05:51,600
You're paying about $10 per 
1,000,000 tokens. 

110
00:05:52,320 --> 00:05:56,080
And output tokens, by the way, 
is I guess is the number of 

111
00:05:56,080 --> 00:05:57,800
words it uses when it responds 
to you. 

112
00:05:59,200 --> 00:06:02,200
For the same 1,000,000 tokens 
that you pay $10.00 for with 

113
00:06:02,200 --> 00:06:05,960
Open AI, you're actually having 
to pay $75.00 for Opus 1. 

114
00:06:05,960 --> 00:06:08,720
So that's Claude's good model. 
So that pricing is absolutely 

115
00:06:08,720 --> 00:06:10,240
crazy. 
It's 10 times cheaper than 

116
00:06:10,240 --> 00:06:14,080
Claude, which is, to be honest, 
really difficult to say no to. 

117
00:06:14,280 --> 00:06:18,040
And it's actually been reported 
that API usage for companies 

118
00:06:18,040 --> 00:06:21,200
using ChatGPT inside their 
application or existing 

119
00:06:21,200 --> 00:06:25,680
workflows and systems has 
doubled in the last few days of 

120
00:06:25,720 --> 00:06:30,040
the ChatGPT launch and kept on 
growing subsequently after that 

121
00:06:30,040 --> 00:06:32,080
launch. 
It's been reported that the 

122
00:06:32,080 --> 00:06:36,240
usage of Open AISAPI, so that's 
using the language models inside

123
00:06:36,240 --> 00:06:39,840
existing software actually 
doubled within just the first 48

124
00:06:39,840 --> 00:06:43,360
hours of ChatGPT 5 coming out. 
This is clearly an aggressive 

125
00:06:43,360 --> 00:06:46,800
pricing strategy designed to win
back market share, especially in

126
00:06:46,800 --> 00:06:50,240
that coding space where 
Anthropic has grown rapidly. 

127
00:06:50,360 --> 00:06:53,400
And finally, they've launched a 
what they described as a unified

128
00:06:53,400 --> 00:06:54,800
system. 
But that basically means that 

129
00:06:54,800 --> 00:06:58,720
they've added a router, which 
quickly decides which model is 

130
00:06:58,720 --> 00:07:01,080
going to be used to perform that
task. 

131
00:07:01,240 --> 00:07:04,240
And that's going to be based on 
the complexity of the tool 

132
00:07:04,240 --> 00:07:05,480
types. 
So if it needs to do a web 

133
00:07:05,480 --> 00:07:09,760
search tool or maybe the type of
question that's been asked, all 

134
00:07:09,760 --> 00:07:12,680
of that sounds really cool. 
And to be honest, the main area 

135
00:07:12,680 --> 00:07:15,200
they've invested in clearly is 
the coding space. 

136
00:07:15,200 --> 00:07:18,160
And to be honest, if you just 
watched their entire launch 

137
00:07:18,160 --> 00:07:22,920
video, if you had no context on 
what Open AI did, you would 

138
00:07:22,920 --> 00:07:25,840
probably assume that they are 
predominantly A coding 

139
00:07:25,840 --> 00:07:38,840
application. 
So as we go into what the 

140
00:07:38,840 --> 00:07:41,000
independent benchmarks say, I 
think that's a really important 

141
00:07:41,000 --> 00:07:44,760
thing to keep top of mind. 
Now, as you will know, I'm not a

142
00:07:44,760 --> 00:07:47,920
massive fan of these benchmarks,
to be honest. 

143
00:07:48,040 --> 00:07:50,560
I think it's a game that all of 
them play and they're all 

144
00:07:50,560 --> 00:07:53,200
obviously going to try and make 
each other look as best they 

145
00:07:53,200 --> 00:07:56,440
possibly can on launch day. 
But I think it helps gives us, I

146
00:07:56,440 --> 00:07:59,000
guess, an initial gauge of 
whether there's been a big leap 

147
00:07:59,000 --> 00:08:00,480
or not. 
Artificial analysis, which is 

148
00:08:00,480 --> 00:08:03,440
the gold standard for 
independent AI benchmarks, found

149
00:08:03,440 --> 00:08:09,120
that when ChatGPT was tested and
it used its reasoning, it was 

150
00:08:09,440 --> 00:08:12,840
the best model out there. 
However, when GPT 5 didn't have 

151
00:08:12,840 --> 00:08:16,960
reasoning, it performs closer to
GPT 4.1's level. 

152
00:08:17,720 --> 00:08:21,000
This part is because actually 
what it shows us is again, the 

153
00:08:21,000 --> 00:08:23,120
open air isn't selling US1 
model. 

154
00:08:23,480 --> 00:08:26,320
They're selling an entire new 
packaging and pricing system. 

155
00:08:27,080 --> 00:08:31,080
This minimal version that most 
people, to be honest, will often

156
00:08:31,080 --> 00:08:36,159
get is basically just their old 
models, and their high reasoning

157
00:08:36,159 --> 00:08:39,880
version that gets the incredible
flashy headlines about coming 

158
00:08:39,880 --> 00:08:43,240
first place in the most 
important benchmarks is going to

159
00:08:43,240 --> 00:08:46,040
be rarely used. 
So through very clever pricing 

160
00:08:46,040 --> 00:08:49,080
and packaging, they're trying to
increase the perception of value

161
00:08:49,240 --> 00:08:51,320
across all their tiers. 
And if you think about what that

162
00:08:51,320 --> 00:08:54,680
means for say, competition, 
Anthropic might charge premium 

163
00:08:54,680 --> 00:08:58,920
prices for Clawed because they 
position it as consistently high

164
00:08:58,920 --> 00:09:00,960
quality. 
But Open AI has built a new 

165
00:09:00,960 --> 00:09:04,920
packaging and pricing system 
that enables them to undercut 

166
00:09:04,960 --> 00:09:08,040
Clawed and other model 
providers, all by using that 

167
00:09:08,040 --> 00:09:11,160
router system, which is 
automatically routing people to 

168
00:09:11,520 --> 00:09:13,960
different models. 
And subsequently, Open AI has 

169
00:09:13,960 --> 00:09:16,200
announced that they're going to 
bring out the ability for you to

170
00:09:16,200 --> 00:09:18,400
understand which models being 
used more often. 

171
00:09:18,680 --> 00:09:22,160
Which for me is a really big win
because I personally haven't 

172
00:09:22,160 --> 00:09:25,680
been trusting the quality of 
responses. 

173
00:09:25,680 --> 00:09:28,880
And I can't help but think, am I
getting the rubbish models and 

174
00:09:28,880 --> 00:09:30,880
still paying money for it? 
When it comes to coding 

175
00:09:30,880 --> 00:09:34,680
benchmarks, Vellum's analysis 
found that Grok 4, GPT 5, and 

176
00:09:34,680 --> 00:09:38,600
Cloud Opus are all performing a 
very similar level on the SWE 

177
00:09:38,600 --> 00:09:43,480
bench, though ChatGPT did come 
first place in many of the other

178
00:09:43,480 --> 00:09:45,280
coding benchmarks like Design 
Arena. 

179
00:09:45,440 --> 00:09:48,880
And so even though ChatGPT has 
led to the number one best 

180
00:09:48,880 --> 00:09:53,720
performing model in many coding 
benchmarks, Group 4, So Elon 

181
00:09:53,720 --> 00:09:58,120
Musk's language model still 
actually gets the best marks on 

182
00:09:58,120 --> 00:10:01,560
some of the hardest types of 
assessments on the ARC AGI two 

183
00:10:01,560 --> 00:10:04,680
test, which is designed 
basically to measure abstract 

184
00:10:04,680 --> 00:10:06,920
reasoning. 
That's more like human problem 

185
00:10:06,920 --> 00:10:10,680
solving, the kind of figure it 
out yourself thinking that 

186
00:10:10,680 --> 00:10:13,920
humans have to do when they've 
never actually seen a specific 

187
00:10:13,920 --> 00:10:16,600
problem before. 
Well, Grokforce scores highest 

188
00:10:16,600 --> 00:10:21,080
on that assessment at 15.9%, 
while ChatGPT 5 actually lags 

189
00:10:21,080 --> 00:10:25,280
behind it, less than 10%, so 
9.9%, which is a huge gap. 

190
00:10:26,400 --> 00:10:30,880
So in short, ChatGPT 5 is 
faster, it's better at coding, 

191
00:10:30,880 --> 00:10:35,600
it's cheaper, but across many of
the reasoning benchmarks, it's 

192
00:10:35,600 --> 00:10:47,650
not as good. 
That's just the benchmarks. 

193
00:10:47,650 --> 00:10:50,370
And as we all know with language
models, it's just as much about 

194
00:10:50,370 --> 00:10:52,490
the vibes. 
And certainly there's been a lot

195
00:10:52,490 --> 00:10:55,970
of noise and conversation about 
those vibes with ChatGPT 5, and 

196
00:10:55,970 --> 00:10:57,610
that's included very divided 
opinions. 

197
00:10:57,610 --> 00:11:00,530
So on Polymarket, which is a 
decentralised prediction market.

198
00:11:00,530 --> 00:11:02,970
So you can bet on whether 
Apple's going to bring out a 

199
00:11:02,970 --> 00:11:08,000
foldable phone in 2026. 
They actually placed 80% odds on

200
00:11:08,000 --> 00:11:11,600
Open AI having the best leading 
AI model by the end of August 

201
00:11:11,960 --> 00:11:14,840
before the demo. 
And then Google was lagging 

202
00:11:14,840 --> 00:11:18,920
behind it, just 20%. 
But by the time Sam Altman and 

203
00:11:18,920 --> 00:11:22,200
the Open AI team ward off that 
stage, the title completely 

204
00:11:22,200 --> 00:11:27,200
turned and Google surged to 85% 
and Open AI to 14%. 

205
00:11:27,520 --> 00:11:31,640
So that only 14% of betters 
believe that Open AI will have 

206
00:11:31,640 --> 00:11:37,640
the best model in 2026 from 80%,
14% and this speaks volume on 

207
00:11:37,640 --> 00:11:40,360
the divided opinion that people 
have got around this new model 

208
00:11:40,360 --> 00:11:44,120
release During the ChatGPT 5 
release, there was also let's 

209
00:11:44,120 --> 00:11:48,240
call it chart crime Open AI 
showed off several charts on the

210
00:11:48,240 --> 00:11:52,040
screen as they spoke to 
demonstrate how good ChatGPT 5 

211
00:11:52,040 --> 00:11:54,440
really is and it honestly left 
me baffled. 

212
00:11:54,440 --> 00:11:57,320
So it's hard to explain, but if 
you want a podcast platform that

213
00:11:57,320 --> 00:11:59,640
has video will actually show the
bar chart because it's it's 

214
00:11:59,640 --> 00:12:03,240
hilarious. 
They had bars that were next to 

215
00:12:03,240 --> 00:12:06,840
each other but drastically 
different percentage outcomes. 

216
00:12:07,240 --> 00:12:10,840
So we had one bar that got a 69%
score on coding deception, 

217
00:12:11,120 --> 00:12:14,720
another one next to it that got 
30.8, but they were displayed as

218
00:12:14,720 --> 00:12:20,040
identical bar heights, while the
Chachi BT5 results were 52.8%. 

219
00:12:20,640 --> 00:12:25,280
Yeah, was visually showed as 
being much, much bigger than the

220
00:12:25,320 --> 00:12:28,600
other bars that were at 69% and 
30%. 

221
00:12:28,760 --> 00:12:31,160
It's criminal. 
Someone addressed it afterwards 

222
00:12:31,160 --> 00:12:34,960
and said it was a mega chart 
screw up from them, but I guess 

223
00:12:34,960 --> 00:12:37,240
many people were claiming that 
they vibe coded it using 

224
00:12:37,240 --> 00:12:39,400
ChatGPT. 
You have Professor Ethan Mollick

225
00:12:39,400 --> 00:12:42,960
predicting that you're going to 
see a lot of varied results 

226
00:12:42,960 --> 00:12:45,720
posted online about chat GB 25 
because it's using multiple 

227
00:12:45,720 --> 00:12:49,320
models in its responses, some of
which are very good and some of 

228
00:12:49,320 --> 00:12:52,040
which are, to be honest, meh. 
And since the underlying model 

229
00:12:52,040 --> 00:12:54,720
selection isn't very 
transparent, it becomes very 

230
00:12:54,720 --> 00:12:56,680
confusing. 
And this is exactly my 

231
00:12:56,680 --> 00:12:59,080
experience so far. 
They also had to do a massive 

232
00:12:59,080 --> 00:13:02,320
patch just after their release 
because usage went through the 

233
00:13:02,320 --> 00:13:04,960
roof and everything started to 
fail, including that auto 

234
00:13:04,960 --> 00:13:06,800
Switcher. 
So switching to different models

235
00:13:07,080 --> 00:13:09,560
and as a result, it just started
putting every single person on 

236
00:13:09,560 --> 00:13:12,640
the worst possible model. 
This blog by Matt Schumer, I 

237
00:13:12,640 --> 00:13:15,160
think, captures the broader 
market reaction really, really 

238
00:13:15,160 --> 00:13:17,280
well. 
And he said, when I started 

239
00:13:17,280 --> 00:13:19,840
using it, I wasn't blown away. 
And in fact, I felt very let 

240
00:13:19,840 --> 00:13:23,920
down given all of this hype and 
a massive expectation around it.

241
00:13:23,920 --> 00:13:27,240
But what Schumer said was that 
if you're using Chachi BT5 for 

242
00:13:27,240 --> 00:13:31,920
tasks that the previous O3 
models or even the Cloud Sonic 

243
00:13:31,920 --> 00:13:35,160
can do really well, then you're 
not really going to get a gauge 

244
00:13:35,160 --> 00:13:38,120
of how good it is to see how 
good this truly is. 

245
00:13:38,120 --> 00:13:42,000
You have to ask it to do things 
that AI simply can't do very 

246
00:13:42,000 --> 00:13:44,120
easily. 
And in short, that is coding. 

247
00:13:45,000 --> 00:13:47,960
And I think this gets the meat 
of what open AI I've clearly 

248
00:13:47,960 --> 00:13:50,960
obsessed over getting right, 
because the vibe coding 

249
00:13:50,960 --> 00:13:55,320
revolution is here, and that 
means it's the era of creating 

250
00:13:55,320 --> 00:13:57,800
things all of the time. 
Before I go into some of the 

251
00:13:57,800 --> 00:14:01,680
examples of what's been built on
ChatGPT 5, I just want to 

252
00:14:01,680 --> 00:14:04,360
quickly stop and acknowledge 
that we're growing incredibly 

253
00:14:04,360 --> 00:14:06,920
fast. 
And according to the stats, only

254
00:14:06,920 --> 00:14:10,400
45% of you listening right now 
are following us or are 

255
00:14:10,400 --> 00:14:12,040
subscribed. 
So if you could press that 

256
00:14:12,040 --> 00:14:15,760
button on your podcast app, that
would help so, so much. 

257
00:14:16,440 --> 00:14:20,080
A shed load of effort goes into 
every episode and that number is

258
00:14:20,080 --> 00:14:22,760
part of what makes it worth it. 
So the Vive coding revolution 

259
00:14:22,760 --> 00:14:24,840
really is here. 
And Matthew Burnham, who has 

260
00:14:24,840 --> 00:14:30,160
480,000 subscribers on YouTube, 
actually focuses all on AI 

261
00:14:30,160 --> 00:14:33,520
coding and he got access to the 
model quite early and he built a

262
00:14:33,520 --> 00:14:35,120
bunch of really, really cool 
things. 

263
00:14:35,400 --> 00:14:38,840
So you had ChatGPT 5 generate 
several very specific 

264
00:14:38,840 --> 00:14:41,360
application to test its vibe 
coding abilities. 

265
00:14:41,520 --> 00:14:43,680
Again, if you're watching the 
video, we'll put them on the 

266
00:14:43,680 --> 00:14:45,960
screen. 
But he made a jumping ball 

267
00:14:45,960 --> 00:14:48,520
runner game. 
So a simple game that although 

268
00:14:48,520 --> 00:14:52,520
is very basic, was functional. 
He created a pixel art tool, so 

269
00:14:52,520 --> 00:14:55,600
an entire application that you 
can draw things on and it can 

270
00:14:55,600 --> 00:14:58,120
change colours. 
He said it was a bit laggy, but 

271
00:14:58,560 --> 00:15:00,200
not too bad. 
There was also a game he 

272
00:15:00,200 --> 00:15:03,200
generated which measures your 
typing speed, so which is a 

273
00:15:03,200 --> 00:15:06,480
functional typing test that 
essentially tracked the accuracy

274
00:15:06,480 --> 00:15:08,040
and speed of you typing out 
words. 

275
00:15:08,200 --> 00:15:10,960
The game even included proper 
feedback and scoring systems. 

276
00:15:12,080 --> 00:15:14,680
He also created a drum 
simulator, which is an 

277
00:15:14,680 --> 00:15:17,480
interactive music application 
that you could literally make 

278
00:15:17,480 --> 00:15:21,280
music on and finally a tool that
visualises music and lo fi, 

279
00:15:21,520 --> 00:15:23,000
which is also just really, 
really cool. 

280
00:15:23,160 --> 00:15:25,640
So the fact that you don't need 
to go to another application 

281
00:15:25,640 --> 00:15:29,560
now, ChatGPT literally does this
for you from a single prompt. 

282
00:15:30,200 --> 00:15:31,560
And this is the world of vibe 
coding. 

283
00:15:31,560 --> 00:15:34,120
Every application is now a vibe 
coding application. 

284
00:15:34,280 --> 00:15:36,880
I think another fantastic review
and one that I wanted to know 

285
00:15:36,880 --> 00:15:39,880
because I really, really enjoyed
the reading was won by the 

286
00:15:39,880 --> 00:15:43,200
Leighton Space and it was one of
the more nuanced and 

287
00:15:43,200 --> 00:15:47,360
philosophical pieces on ChatGPT 
5 and they called it the Stone 

288
00:15:47,360 --> 00:15:48,880
Age. 
This wasn't a criticism. 

289
00:15:49,160 --> 00:15:52,200
This is a reflection of where 
they think we are in human 

290
00:15:52,200 --> 00:15:55,760
civilisation right now, because 
this is one of the first model 

291
00:15:55,760 --> 00:16:00,120
that excels at using tools 
during conversations, just as 

292
00:16:00,120 --> 00:16:01,720
humans did a long, long time 
ago. 

293
00:16:02,680 --> 00:16:05,720
So as they put it, the Stone Age
marked the dawn of human 

294
00:16:05,720 --> 00:16:09,440
intelligence, but what exactly 
made it so significant is the 

295
00:16:09,440 --> 00:16:12,200
fact that we shaped tools and 
our tools shaped us. 

296
00:16:21,780 --> 00:16:24,020
So I guess what does all this 
mean for Anthropic? 

297
00:16:24,180 --> 00:16:26,300
So the threat to Anthropic is 
really interesting because 

298
00:16:26,300 --> 00:16:31,280
apparently according to reports 
both Cursor and GitHub copilot 

299
00:16:31,280 --> 00:16:35,880
drive approximately $1.2 billion
of Anthropics 4 billion revenue.

300
00:16:36,360 --> 00:16:40,440
So that is just two applications
built and using clawed every 

301
00:16:40,440 --> 00:16:42,880
single day as part of their 
application that is driving most

302
00:16:42,880 --> 00:16:44,920
of their revenue while 30% of it
anyway. 

303
00:16:45,040 --> 00:16:48,280
And with Chat GPT's new model 
coming out and it's incredible 

304
00:16:48,280 --> 00:16:53,040
abilities with coding and 
costing 7 times less than Claws 

305
00:16:53,040 --> 00:16:56,680
models, well, it's going to be 
hard for an enterprise customer 

306
00:16:56,680 --> 00:16:58,920
to say no. 
And lots of coding applications 

307
00:16:58,920 --> 00:17:02,600
like Livable Cursor and others 
have made ChatGPT 5 their 

308
00:17:02,600 --> 00:17:06,760
default model again because 
Claude has brought out Claude 

309
00:17:06,760 --> 00:17:09,440
Code, which is actually a 
competitor to many of the 

310
00:17:09,440 --> 00:17:12,160
companies that are using their 
language model. 

311
00:17:12,760 --> 00:17:15,920
Many of those companies are now 
jumped straight to Open AI to 

312
00:17:15,920 --> 00:17:18,160
make it their default model. 
This doesn't mean that 

313
00:17:18,160 --> 00:17:21,920
developers will not choose 
Claude still, but it's certainly

314
00:17:21,920 --> 00:17:35,720
a massive risk to Anthropic. 
So I guess to conclude, ChatGPT 

315
00:17:35,720 --> 00:17:40,320
represents both impressive 
technical achievements and 

316
00:17:40,960 --> 00:17:43,120
really importantly, very 
aggressive competitive 

317
00:17:43,120 --> 00:17:46,640
strategies against Anthropic. 
For coding, it appears to have 

318
00:17:47,040 --> 00:17:50,840
made a huge leap forward, but 
for other applications the 

319
00:17:50,840 --> 00:17:53,720
improvements feel incredibly 
incremental and to be honest, 

320
00:17:53,720 --> 00:17:56,520
overall disappointing. 
This was just as much of A 

321
00:17:56,680 --> 00:18:01,720
business model change disguised 
as technical leaps because Open 

322
00:18:01,720 --> 00:18:04,000
AI have just built a system 
where they can offer cheaper 

323
00:18:04,000 --> 00:18:07,400
language models to everybody 
with minimal reasoning, while 

324
00:18:07,400 --> 00:18:10,360
still managed to claim that they
have the very best model on the 

325
00:18:10,360 --> 00:18:12,320
market. 
And obviously it's working. 

326
00:18:12,400 --> 00:18:15,280
Cursor switched overnight, 
prediction markets lost 

327
00:18:15,280 --> 00:18:18,880
confidence in Open AI and 
Anthropic is watching 30% of 

328
00:18:18,880 --> 00:18:22,880
their revenue become potentially
vulnerable to their pricing. 

329
00:18:22,960 --> 00:18:26,120
And I guess anecdotally, the 
chart crimes, the very mixed 

330
00:18:26,120 --> 00:18:29,840
reviews, the expectation gaps 
are probably a symptom of now a 

331
00:18:29,840 --> 00:18:34,320
maturing market where these 
revolutionary AI narratives just

332
00:18:35,000 --> 00:18:37,240
don't quite work in the same way
as they used to. 

333
00:18:37,840 --> 00:18:41,320
The real story here is where the
ChatGPT can write better code, 

334
00:18:41,320 --> 00:18:45,520
write better content, and solve 
real problems for everyday 

335
00:18:45,520 --> 00:18:47,440
people. 
Anyway, that's it for today. 

336
00:18:47,920 --> 00:18:50,480
I hope you enjoyed that deep 
dive into ChatGPT 5. 

337
00:18:51,040 --> 00:18:53,640
Thanks for listening and I will 
see you next week.