النص 🔗
WEBVTT
00:00:05.128 --> 00:00:06.310
Thanks for watching!
00:00:21.263 --> 00:00:22.843
welcome again to another episode of cloud
00:00:22.882 --> 00:00:28.243
unplugged we have two big topics uh well
00:00:28.364 --> 00:00:30.885
one big topic one smaller one um we
00:00:30.925 --> 00:00:35.305
have anthropic uh accusing alibaba of
00:00:35.765 --> 00:00:37.386
basically a distillation attack against
00:00:37.405 --> 00:00:40.866
their model so reverse engineering um what
00:00:40.887 --> 00:00:42.307
they can work out how the model was
00:00:42.366 --> 00:00:44.088
built um what type of features and
00:00:44.107 --> 00:00:45.148
capabilities it's providing
00:00:46.109 --> 00:00:48.750
to accelerate their own AI growth.
00:00:49.051 --> 00:00:51.093
And then we also have the fact that
00:00:51.133 --> 00:00:52.674
Mapbox just got twenty percent more
00:00:52.734 --> 00:00:54.877
expensive because of AI.
00:00:55.156 --> 00:00:56.698
So if you're out to buy a MacBook,
00:00:56.737 --> 00:00:58.119
you better go and get a bank loan.
00:00:59.640 --> 00:01:03.283
So how are things going with the Anthropic
00:01:03.384 --> 00:01:05.626
story, Salman?
00:01:05.927 --> 00:01:08.248
What was going on there?
00:01:09.234 --> 00:01:11.216
Well, John, here's the question.
00:01:11.677 --> 00:01:13.519
Why would you want to spend hundreds of
00:01:13.558 --> 00:01:16.180
millions of dollars to train a model when
00:01:16.200 --> 00:01:17.662
you could just learn from one that already
00:01:17.703 --> 00:01:18.143
exists?
00:01:18.283 --> 00:01:21.046
And that is the question Alibaba asked.
00:01:21.605 --> 00:01:22.447
And what they did,
00:01:23.087 --> 00:01:25.709
what Anthropic realized that between April
00:01:25.750 --> 00:01:26.170
and June,
00:01:26.230 --> 00:01:28.632
so I think the timelines are between the
00:01:28.653 --> 00:01:30.293
twenty second of April and the fifth of
00:01:30.313 --> 00:01:30.534
June.
00:01:31.275 --> 00:01:33.376
Anthropic identified that there's a
00:01:33.436 --> 00:01:36.215
coordinated campaign by Alibaba's Qwenn AI
00:01:36.236 --> 00:01:36.536
models.
00:01:36.555 --> 00:01:36.756
You know,
00:01:36.915 --> 00:01:39.096
the Qwenn AI models that perhaps you've
00:01:39.216 --> 00:01:40.897
used or perhaps you like them or not.
00:01:42.016 --> 00:01:44.218
They used almost twenty five thousand
00:01:44.237 --> 00:01:47.019
accounts and they conducted around twenty
00:01:47.058 --> 00:01:48.739
million interactions with Claude and we'll
00:01:48.759 --> 00:01:50.198
come to what an interaction is in a
00:01:50.259 --> 00:01:50.618
second.
00:01:51.620 --> 00:01:55.040
The goal was to systematically extract
00:01:55.441 --> 00:01:56.801
Claude's capabilities.
00:01:57.908 --> 00:01:59.489
how it clothes reasons,
00:01:59.888 --> 00:02:03.371
how it uses the tools, how, you know,
00:02:03.512 --> 00:02:06.433
the coding abilities to train their own
00:02:06.513 --> 00:02:09.175
smaller coin model.
00:02:09.215 --> 00:02:11.276
So this is what Anthropic found out.
00:02:11.575 --> 00:02:13.617
And what they did was as soon as
00:02:13.638 --> 00:02:14.478
they identified,
00:02:14.959 --> 00:02:16.498
and they called it corporate espionage
00:02:16.780 --> 00:02:18.360
because it is corporate espionage,
00:02:19.020 --> 00:02:21.081
what they found out is then they went
00:02:21.122 --> 00:02:22.103
to the U.S.
00:02:22.143 --> 00:02:23.963
senators, a couple of U.S.
00:02:24.003 --> 00:02:26.025
senators, and told them about it, that,
00:02:26.064 --> 00:02:27.165
you know, this is what's happened.
00:02:28.746 --> 00:02:31.248
And they sent a letter to the Senate
00:02:31.269 --> 00:02:32.049
Banking Committee.
00:02:33.450 --> 00:02:34.572
And then two days later,
00:02:34.752 --> 00:02:36.454
once they identified it and sent the
00:02:36.473 --> 00:02:38.055
letter to the Banking Committee,
00:02:38.074 --> 00:02:40.037
you remember when Mythos and Fable got
00:02:40.557 --> 00:02:43.460
banned from non US use
00:02:44.121 --> 00:02:46.162
That ban got put in place.
00:02:46.521 --> 00:02:50.585
But the initial thing is that Alibaba was
00:02:50.724 --> 00:02:53.645
using these models, the Claude models,
00:02:53.686 --> 00:02:54.947
to train their own coin models.
00:02:55.487 --> 00:02:57.187
That's the headline story, John.
00:02:57.508 --> 00:02:57.889
And of course,
00:02:57.908 --> 00:03:00.210
there's a lot to it than just what's
00:03:00.250 --> 00:03:00.550
happened.
00:03:00.890 --> 00:03:02.411
But this is what they found out.
00:03:03.292 --> 00:03:03.931
Do you think,
00:03:03.991 --> 00:03:06.133
because I don't know if you saw,
00:03:06.174 --> 00:03:09.675
but Anthropic...
00:03:10.426 --> 00:03:12.888
have obviously beaten OpenAI on the
00:03:12.907 --> 00:03:13.307
market.
00:03:13.568 --> 00:03:17.248
I think OpenAI dropped to like twenty
00:03:17.288 --> 00:03:19.110
seven percent of market share for
00:03:19.370 --> 00:03:20.030
enterprises.
00:03:20.110 --> 00:03:23.290
But it sounds like maybe because how many
00:03:23.591 --> 00:03:25.532
subscriptions were they getting from
00:03:25.592 --> 00:03:25.972
China?
00:03:26.032 --> 00:03:28.872
So we're saying that basically that twenty
00:03:28.932 --> 00:03:29.193
percent,
00:03:29.212 --> 00:03:30.933
that additional twenty percent that made
00:03:30.954 --> 00:03:32.814
them forty seven percent market share was
00:03:32.854 --> 00:03:33.875
really coming from China.
00:03:33.914 --> 00:03:34.775
And maybe there is no,
00:03:35.254 --> 00:03:36.455
maybe they're on a par.
00:03:37.336 --> 00:03:39.217
I mean, were they making money from China?
00:03:39.236 --> 00:03:39.616
We don't know.
00:03:40.717 --> 00:03:41.780
I mean, that's right, right, John?
00:03:41.860 --> 00:03:44.643
Look, you know, we said that Airbnb,
00:03:44.943 --> 00:03:46.564
not Airbnb, Uber came around and
Load full transcript
said,
00:03:46.604 --> 00:03:46.824
well,
00:03:47.406 --> 00:03:48.667
we're not really seeing a return on
00:03:48.707 --> 00:03:49.247
investment.
00:03:49.467 --> 00:03:50.308
Well, Alibaba did.
00:03:50.610 --> 00:03:51.570
They're training their models.
00:03:52.151 --> 00:03:53.652
They've got really good ROI.
00:03:53.712 --> 00:03:54.373
They've got the best ROI.
00:03:54.394 --> 00:03:56.556
Yeah, the best ROI out of that.
00:03:57.662 --> 00:03:59.544
So, John, it's like,
00:04:00.104 --> 00:04:01.825
if you look at an interaction,
00:04:01.966 --> 00:04:03.887
it's anywhere between five hundred to a
00:04:03.926 --> 00:04:04.788
thousand tokens, right?
00:04:04.828 --> 00:04:05.188
Roughly.
00:04:05.609 --> 00:04:07.189
Just talking about roughly five hundred to
00:04:07.210 --> 00:04:07.830
a thousand tokens.
00:04:08.251 --> 00:04:09.932
Let's just say a thousand tokens.
00:04:10.372 --> 00:04:11.933
And there's about twenty nine million
00:04:12.455 --> 00:04:13.216
interactions.
00:04:13.876 --> 00:04:14.856
So it's about, you can say,
00:04:14.896 --> 00:04:18.220
between ten to twenty billion tokens that
00:04:18.240 --> 00:04:19.341
were used, right?
00:04:19.401 --> 00:04:20.081
Roughly, roughly.
00:04:20.101 --> 00:04:21.783
It's just rough, rough estimates.
00:04:23.714 --> 00:04:24.975
is a reasonable amount of money,
00:04:25.016 --> 00:04:28.197
like maybe in hundreds of thousands of
00:04:28.257 --> 00:04:28.557
dollars.
00:04:28.677 --> 00:04:31.841
So perhaps not a lot in the grand
00:04:31.880 --> 00:04:34.362
scheme of things.
00:04:34.723 --> 00:04:36.184
And we're like, turn it back on.
00:04:36.204 --> 00:04:37.625
We've got an IPO coming up.
00:04:38.305 --> 00:04:39.906
Come on, guys.
00:04:39.925 --> 00:04:40.807
Come on, China.
00:04:41.706 --> 00:04:43.208
Get some more subscriptions going.
00:04:43.247 --> 00:04:46.971
We're about to IPO.
00:04:47.011 --> 00:04:48.971
So the thing here is that, of course,
00:04:48.992 --> 00:04:50.333
this is not the first time that's
00:04:50.814 --> 00:04:51.153
happened.
00:04:52.536 --> 00:04:54.098
They also found out in the past that
00:04:54.218 --> 00:04:56.117
Deep Seek carried out about one hundred
00:04:56.137 --> 00:04:58.218
and fifty thousand of these interactions.
00:04:58.959 --> 00:05:00.399
Moonshot AI, again,
00:05:00.439 --> 00:05:02.040
another model from China,
00:05:02.461 --> 00:05:03.600
about three point four million
00:05:03.661 --> 00:05:05.401
extractions, right exchanges.
00:05:05.822 --> 00:05:07.302
And there was another one called Minimax,
00:05:07.422 --> 00:05:08.502
about thirteen million.
00:05:08.622 --> 00:05:10.202
I mean, there's still still a lot,
00:05:10.524 --> 00:05:12.783
but not at this level of almost thirty
00:05:12.824 --> 00:05:13.644
million interactions.
00:05:13.704 --> 00:05:13.884
Right.
00:05:13.904 --> 00:05:14.264
So this is.
00:05:14.605 --> 00:05:16.026
this has
happened in the past of course
00:05:16.045 --> 00:05:18.428
anthropic every time they find it they
00:05:18.449 --> 00:05:20.069
improve their anomaly detection they've
00:05:20.730 --> 00:05:22.052
been trying to figure out how they can
00:05:22.091 --> 00:05:24.413
make it more secure because this is one
00:05:24.454 --> 00:05:25.675
of the things that people have been
00:05:25.694 --> 00:05:27.697
talking about this model distillation
00:05:28.156 --> 00:05:31.040
where people can use your model to learn
00:05:31.560 --> 00:05:33.581
the secrets that you're using but then you
00:05:33.601 --> 00:05:34.622
know do you know how they do it
00:05:34.663 --> 00:05:36.504
though as in like i mean i i
00:05:36.605 --> 00:05:37.706
let's put you on the spot because you
00:05:37.725 --> 00:05:38.826
may or may not know but i'm not
00:05:38.867 --> 00:05:39.848
sure how
00:05:41.636 --> 00:05:44.716
how do they use that information to then
00:05:44.797 --> 00:05:47.218
improve their own model and accelerate the
00:05:47.278 --> 00:05:48.439
learnings through another model?
00:05:49.658 --> 00:05:50.199
Yeah, I think,
00:05:50.540 --> 00:05:52.040
I don't know the exact details, John,
00:05:52.079 --> 00:05:54.901
but it's feeding whatever output of those
00:05:54.942 --> 00:05:57.442
specific questions that they have and bake
00:05:57.723 --> 00:06:01.524
it into the model patterns, right?
00:06:01.543 --> 00:06:04.165
So they ask specific questions,
00:06:04.384 --> 00:06:05.386
specific patterns,
00:06:05.505 --> 00:06:07.326
and then just feed that into the model
00:06:07.346 --> 00:06:07.726
itself, perhaps.
00:06:07.747 --> 00:06:09.326
Right, they're basically prompting.
00:06:09.367 --> 00:06:10.288
So they've got a load of,
00:06:10.908 --> 00:06:13.891
specific tailored specific that are then
00:06:14.170 --> 00:06:16.312
used as a mechanism to then obviously
00:06:16.432 --> 00:06:19.274
extract data in a format that then helps
00:06:19.295 --> 00:06:20.776
them with the model that then they can
00:06:20.797 --> 00:06:22.898
then accelerate their own training using
00:06:22.918 --> 00:06:26.000
the app of another model correct and yeah
00:06:26.040 --> 00:06:27.843
that's that that's correct this feels like
00:06:27.862 --> 00:06:28.843
a sense of um
00:06:30.745 --> 00:06:33.146
karma isn't it isn't that called karma
00:06:33.346 --> 00:06:34.925
where you train a model based on every
00:06:34.946 --> 00:06:36.545
other people's work but you're not happy
00:06:36.586 --> 00:06:38.247
that somebody else trains their model and
00:06:38.266 --> 00:06:39.586
your model that was trained on other
00:06:39.627 --> 00:06:41.107
people's work just so i understand the
00:06:41.487 --> 00:06:44.408
slight hypocrisy here so they're not happy
00:06:45.427 --> 00:06:49.028
that somebody else is getting ahead based
00:06:49.088 --> 00:06:51.048
on the data that they train their model
00:06:51.129 --> 00:06:54.310
on which obviously wasn't theirs um but
00:06:54.329 --> 00:06:55.709
they're not happy that somebody else does
00:06:55.750 --> 00:06:57.370
it to them is that is that what
00:06:57.391 --> 00:06:58.050
i'm hearing here
00:06:58.802 --> 00:06:59.963
That's exactly what you're hearing.
00:07:00.002 --> 00:07:01.684
That's exactly what some people are
00:07:01.723 --> 00:07:02.043
saying.
00:07:02.163 --> 00:07:03.845
They are saying that, well,
00:07:03.944 --> 00:07:06.586
isn't what anthropic getting the taste of
00:07:06.625 --> 00:07:07.886
the bit of the taste of their own
00:07:07.925 --> 00:07:10.026
medicine because they trained it on data
00:07:10.047 --> 00:07:11.048
that already exists, right?
00:07:11.088 --> 00:07:11.987
Of course, it's open.
00:07:12.468 --> 00:07:14.968
But this comes under John, of course,
00:07:15.009 --> 00:07:16.850
with all of these commercial things,
00:07:16.970 --> 00:07:18.971
fair usage policy, right?
00:07:18.990 --> 00:07:22.512
So this is protecting their IP to some
00:07:22.552 --> 00:07:22.971
extent.
00:07:23.992 --> 00:07:26.112
Because if you use fake accounts,
00:07:26.434 --> 00:07:27.213
bots or whatever,
00:07:27.874 --> 00:07:29.175
about twenty five hundred twenty five
00:07:29.196 --> 00:07:32.319
thousand of those i think legally it does
00:07:32.699 --> 00:07:35.422
it you know it does count as corporate
00:07:35.461 --> 00:07:39.045
espionage but this is why this is why
00:07:39.105 --> 00:07:41.788
anthropic went to the senate banking
00:07:41.827 --> 00:07:44.029
committee and the us the the commerce
00:07:44.069 --> 00:07:46.692
department to restrict maybe to put in
00:07:47.213 --> 00:07:48.494
some of these uh some
00:07:49.314 --> 00:07:50.815
some of these guard rails around to
00:07:50.875 --> 00:07:51.596
protect themselves.
00:07:51.716 --> 00:07:54.817
I think it's a bit more geopolitical than
00:07:56.077 --> 00:07:58.139
technical, but that's what's happened.
00:07:58.478 --> 00:07:59.778
What do you think about all of this?
00:08:00.019 --> 00:08:01.060
I just find it ironic.
00:08:01.139 --> 00:08:02.661
I mean, I don't know if it's ironic,
00:08:02.680 --> 00:08:04.742
but just hypocritical.
00:08:04.922 --> 00:08:08.423
Just the hypocrisy levels of when it
00:08:08.442 --> 00:08:11.024
happens to others, we're okay with that.
00:08:11.064 --> 00:08:12.045
When it happens to me,
00:08:12.464 --> 00:08:13.725
I'm really not okay with that.
00:08:13.785 --> 00:08:16.425
Like the double standards as a business.
00:08:18.365 --> 00:08:19.404
You know,
00:08:19.586 --> 00:08:23.687
it does feel a little bit ridiculous that
00:08:23.706 --> 00:08:25.307
they have an issue with their own business
00:08:25.327 --> 00:08:27.769
potentially getting risked because they
00:08:27.788 --> 00:08:28.728
don't work for the state.
00:08:28.949 --> 00:08:30.389
The state hasn't got investment in
00:08:30.709 --> 00:08:31.269
Anthropic.
00:08:31.310 --> 00:08:33.931
It's not a government model, right?
00:08:33.971 --> 00:08:37.832
So it's a pure commercial risk.
00:08:38.273 --> 00:08:38.993
Let's be honest.
00:08:40.953 --> 00:08:41.933
Because there is no,
00:08:42.955 --> 00:08:44.676
you don't need to go to the US
00:08:44.735 --> 00:08:45.495
Senate for
00:08:46.508 --> 00:08:48.870
something that is a real commercial threat
00:08:48.909 --> 00:08:50.650
potentially it doesn't become a national
00:08:51.191 --> 00:08:52.711
threat because somebody can produce a
00:08:52.772 --> 00:08:56.095
model like yours because say the argument
00:08:56.134 --> 00:08:57.855
was that their quen model was on a
00:08:58.336 --> 00:09:01.337
on a par uh with fable or mythos
00:09:01.398 --> 00:09:03.919
then what would they have done like you
00:09:03.940 --> 00:09:05.600
wouldn't need to block because they've put
00:09:05.620 --> 00:09:06.802
the investment in they've trained the
00:09:06.822 --> 00:09:08.342
model to the same standard so it's just
00:09:09.224 --> 00:09:11.345
equal measure so there's nothing to do
00:09:11.424 --> 00:09:13.366
right to anthropic anthropic couldn't do
00:09:13.407 --> 00:09:13.787
anything
00:09:14.678 --> 00:09:16.580
in that situation where China's built a
00:09:16.639 --> 00:09:19.400
comparable model to them, right?
00:09:20.201 --> 00:09:20.422
Yeah.
00:09:20.442 --> 00:09:24.263
But I think, John,
00:09:25.583 --> 00:09:26.945
based on the back of this,
00:09:27.524 --> 00:09:30.466
the ban on this export control of Mythos
00:09:31.126 --> 00:09:34.827
and Fable shows that this is being treated
00:09:34.868 --> 00:09:36.509
the same way as other technology.
00:09:36.849 --> 00:09:39.250
That's an export ban that doesn't get sold
00:09:39.269 --> 00:09:40.230
to every country.
00:09:41.091 --> 00:09:43.312
So this is where this is going.
00:09:44.392 --> 00:09:46.131
Yes, the government is not involved,
00:09:46.692 --> 00:09:48.613
but the government sees the potential on
00:09:48.633 --> 00:09:49.972
what these models can be used.
00:09:50.153 --> 00:09:53.453
Don't forget that Anthropic are the ones
00:09:53.474 --> 00:09:56.274
who denied US government usage of their
00:09:56.313 --> 00:09:58.375
models with the removal of guardrails so
00:09:58.414 --> 00:09:59.794
they can use it for whatever they wanted
00:09:59.815 --> 00:10:00.375
to use it for.
00:10:00.394 --> 00:10:01.936
So it's the same company.
00:10:02.336 --> 00:10:04.076
And then they went back on it.
00:10:04.135 --> 00:10:05.755
But here's the thing, though, John,
00:10:06.115 --> 00:10:07.236
if you compare these models,
00:10:07.277 --> 00:10:08.297
I was just looking at the cost.
00:10:08.956 --> 00:10:10.356
these models right i don't know i've never
00:10:10.476 --> 00:10:13.717
used any of the alibaba's quen models but
00:10:14.298 --> 00:10:16.678
the entropic if you just take opus opus
00:10:16.739 --> 00:10:19.119
is five dollars per million input tokens
00:10:19.639 --> 00:10:21.600
and twenty five dollars per million output
00:10:21.620 --> 00:10:23.220
tokens right now the mythos and fable of
00:10:23.279 --> 00:10:25.160
course they were more expensive yeah do
00:10:25.181 --> 00:10:27.140
you want to guess how much quen is
00:10:28.041 --> 00:10:29.642
oh i mean is it done in dollars
00:10:29.662 --> 00:10:31.522
you've done it in dollars you haven't like
00:10:31.542 --> 00:10:32.701
i'm just doing a dollar now i'm not
00:10:32.721 --> 00:10:34.102
changing it to yen or anything like that
00:10:34.182 --> 00:10:36.163
yeah okay so that's good um
00:10:38.154 --> 00:10:38.735
I don't know.
00:10:38.794 --> 00:10:40.176
I guess what are we talking?
00:10:40.216 --> 00:10:43.357
Probably I'd be annoying if I'm right as
00:10:43.398 --> 00:10:44.717
well, which is kind of the risk,
00:10:44.739 --> 00:10:45.019
isn't it?
00:10:45.038 --> 00:10:46.119
That I fluke myself.
00:10:46.879 --> 00:10:48.080
I fluke myself into the number.
00:10:48.120 --> 00:10:49.020
I'm going to say ten.
00:10:51.241 --> 00:10:51.782
Ten dollars.
00:10:51.881 --> 00:10:52.482
Ten dollars.
00:10:52.623 --> 00:10:52.982
Okay.
00:10:53.403 --> 00:10:55.644
It's five cents per million input tokens
00:10:55.663 --> 00:10:58.505
and twenty cents per million output
00:10:58.546 --> 00:10:58.985
tokens.
00:10:59.645 --> 00:11:00.927
Why are we not using Quen then?
00:11:03.283 --> 00:11:08.071
Why are we doing nonthropic then Salman?
00:11:08.650 --> 00:11:10.192
So what I'm telling you is that there
00:11:10.232 --> 00:11:12.616
are some models out there which are way
00:11:12.677 --> 00:11:16.400
more because you could train based on
00:11:16.442 --> 00:11:16.761
other models.
00:11:16.802 --> 00:11:18.244
I don't know because
00:11:20.184 --> 00:11:20.924
The market,
00:11:21.105 --> 00:11:23.225
the way they present themselves is that
00:11:23.265 --> 00:11:25.826
this is more affordable model to use.
00:11:26.106 --> 00:11:26.466
Of course,
00:11:26.486 --> 00:11:27.707
I've never checked the capabilities.
00:11:27.807 --> 00:11:29.347
There's a bunch of benchmarks out there
00:11:29.688 --> 00:11:31.729
that you can see always put Claude Opus
00:11:32.309 --> 00:11:32.809
to the top.
00:11:32.909 --> 00:11:34.110
But okay, fair enough.
00:11:34.509 --> 00:11:35.671
Claude is the best Opus.
00:11:36.130 --> 00:11:37.350
The Opus model are the best at the
00:11:37.390 --> 00:11:37.672
moment.
00:11:38.091 --> 00:11:39.331
But do you need that?
00:11:39.511 --> 00:11:40.773
Maybe you can get away with Quinn.
00:11:40.793 --> 00:11:42.273
It's five cents.
00:11:42.373 --> 00:11:42.572
John,
00:11:42.613 --> 00:11:46.615
this is like hundred times cheaper than
00:11:46.654 --> 00:11:47.735
the models that we use every day.
00:11:50.518 --> 00:11:51.639
Outside of the cost,
00:11:51.679 --> 00:11:52.961
which is obviously ridiculous,
00:11:53.383 --> 00:11:55.005
that there's such a disparity between the
00:11:55.025 --> 00:11:55.385
costs.
00:11:56.648 --> 00:12:03.840
But you're in a global economic situation,
00:12:03.899 --> 00:12:04.081
right?
00:12:05.394 --> 00:12:06.674
We've got globalization.
00:12:07.115 --> 00:12:09.475
Companies now have a global footprint.
00:12:09.895 --> 00:12:11.476
You've got people in Europe using
00:12:11.596 --> 00:12:12.996
Anthropic, et cetera, et cetera.
00:12:13.017 --> 00:12:15.238
Obviously,
00:12:15.258 --> 00:12:16.599
there's a whole sovereignty thing going
00:12:16.658 --> 00:12:18.178
on, which we've mentioned several times.
00:12:19.039 --> 00:12:20.759
So I think everybody's looking at everyone
00:12:21.581 --> 00:12:23.581
on the risk of not owning something,
00:12:25.182 --> 00:12:26.903
especially a frontier model,
00:12:27.202 --> 00:12:38.807
where if for some reason the adoption –
00:12:33.136 --> 00:12:34.856
kind of goes through the roof and the
00:12:34.976 --> 00:12:36.557
ROI does start to get recognized.
00:12:36.618 --> 00:12:38.597
And then there is a disruption to
00:12:38.618 --> 00:12:39.138
employment.
00:12:39.158 --> 00:12:40.938
Let's just assume that path gets taken.
00:12:42.158 --> 00:12:44.119
You don't own anything.
00:12:44.158 --> 00:12:45.840
All of that value has gone to a
00:12:45.860 --> 00:12:47.240
specific company.
00:12:47.340 --> 00:12:49.039
And that company has nothing to do with
00:12:49.440 --> 00:12:50.541
your country, right?
00:12:50.660 --> 00:12:52.061
It's not European tax.
00:12:52.081 --> 00:12:55.022
It's not going back into Europe or any
00:12:55.081 --> 00:12:55.961
European country.
00:12:56.461 --> 00:12:58.923
It's going into the States and it's
00:12:58.962 --> 00:13:01.003
obviously grown the US economy.
00:13:02.798 --> 00:13:06.539
And so knowing that you're in a globalised
00:13:06.580 --> 00:13:10.160
position and that you've got a competitor,
00:13:10.780 --> 00:13:12.121
Quen,
00:13:12.182 --> 00:13:16.543
that's then sneakily trying to accelerate
00:13:16.623 --> 00:13:18.722
its AI adoption.
00:13:18.763 --> 00:13:21.423
Is it a little bit more underhand?
00:13:21.583 --> 00:13:25.105
Is it like someone maybe taking steroids
00:13:25.144 --> 00:13:25.884
in the Olympics?
00:13:25.945 --> 00:13:26.326
Maybe.
00:13:26.546 --> 00:13:27.725
Maybe it is a little bit like that.
00:13:27.745 --> 00:13:29.725
But they're all competing.
00:13:31.926 --> 00:13:34.106
They're all competing at the end of the
00:13:34.147 --> 00:13:34.447
day,
00:13:34.667 --> 00:13:39.568
and you can't suddenly start be okay on
00:13:39.668 --> 00:13:40.289
one hand.
00:13:41.149 --> 00:13:44.250
When we make the rules, it's okay.
00:13:44.331 --> 00:13:45.631
And we're controlling the globe.
00:13:45.692 --> 00:13:46.432
That's okay.
00:13:46.731 --> 00:13:48.033
And so long as you follow our pattern,
00:13:48.052 --> 00:13:48.592
that's fine.
00:13:48.633 --> 00:13:50.433
But if you try and do it as
00:13:50.494 --> 00:13:52.235
well, that's kind of not okay, guys.
00:13:52.254 --> 00:13:54.515
It's not okay that you improve your mind.
00:13:54.535 --> 00:13:56.697
It's not okay that you can monetize yours.
00:13:56.716 --> 00:13:57.876
Do you know what I mean?
00:13:57.917 --> 00:13:59.357
It just feels a little bit like that.
00:14:01.177 --> 00:14:02.399
And you're charging five cents.
00:14:02.519 --> 00:14:04.360
What are you doing?
00:14:04.600 --> 00:14:06.221
Yeah,
00:14:06.240 --> 00:14:08.182
charging five cents is a bit outrageous.
00:14:08.643 --> 00:14:10.124
They can't make any money out of that,
00:14:10.163 --> 00:14:10.644
surely.
00:14:11.464 --> 00:14:13.745
But, John, it's the scale, right?
00:14:13.806 --> 00:14:15.966
So it depends how they train,
00:14:16.047 --> 00:14:17.067
how they run these models.
00:14:17.207 --> 00:14:20.750
And, of course, here's the thing, though.
00:14:20.770 --> 00:14:22.312
Sorry, just a quick one, though, on that.
00:14:22.392 --> 00:14:22.532
Yeah.
00:14:24.635 --> 00:14:27.157
Does Alibaba Cloud have an open source
00:14:27.197 --> 00:14:27.837
model as well?
00:14:28.557 --> 00:14:30.139
Have they bothered with that or they
00:14:30.178 --> 00:14:30.359
don't?
00:14:30.379 --> 00:14:31.700
They're just all commercial?
00:14:32.379 --> 00:14:32.879
I have no idea.
00:14:32.899 --> 00:14:32.980
Yeah,
00:14:33.039 --> 00:14:36.581
I know you've got obviously some open
00:14:36.601 --> 00:14:37.783
source, but I don't think they do,
00:14:37.822 --> 00:14:40.283
do they?
00:14:40.403 --> 00:14:41.205
At the moment,
00:14:41.245 --> 00:14:42.524
the only country that's doing
00:14:45.219 --> 00:14:47.299
doing a lot of open source model is
00:14:47.340 --> 00:14:48.321
China, right?
00:14:48.360 --> 00:14:50.081
So China's like deep seeks coming out as
00:14:50.121 --> 00:14:50.701
open source.
00:14:50.721 --> 00:14:51.842
There's so many models that come out.
00:14:51.923 --> 00:14:52.884
That was it.
00:14:52.903 --> 00:14:53.663
That's where the thread was.
00:14:53.683 --> 00:14:53.903
Yeah,
00:14:53.923 --> 00:14:55.184
deep seek is coming out as open source.
00:14:55.225 --> 00:14:57.667
You've obviously got Gemma as well,
00:14:58.407 --> 00:14:59.368
which is Google, isn't it?
00:15:00.347 --> 00:15:01.788
And I have just fact checked, John,
00:15:02.389 --> 00:15:04.711
because we can do that with the use
00:15:04.750 --> 00:15:05.311
of internet.
00:15:05.851 --> 00:15:10.153
Quinn is most widely adopted open source
00:15:10.274 --> 00:15:10.794
AI families.
00:15:11.054 --> 00:15:11.654
That was it.
00:15:12.115 --> 00:15:13.255
I thought it was open source.
00:15:13.275 --> 00:15:13.777
You're right.
00:15:15.393 --> 00:15:17.534
So I guess the risk is then beyond
00:15:17.595 --> 00:15:19.135
commercial because it basically means,
00:15:19.176 --> 00:15:20.376
well, if this local model,
00:15:21.057 --> 00:15:25.682
open source model gets more widely
00:15:25.701 --> 00:15:26.802
adopted, you know,
00:15:26.863 --> 00:15:28.203
then how do you make money?
00:15:28.364 --> 00:15:29.345
If it's on a par and you can
00:15:29.365 --> 00:15:30.005
run it yourself,
00:15:30.025 --> 00:15:31.126
then we're going to go out of business.
00:15:31.147 --> 00:15:32.028
So we need to stop it.
00:15:32.988 --> 00:15:33.489
So, yeah.
00:15:35.133 --> 00:15:38.076
you know another thing is right so they
00:15:38.115 --> 00:15:40.677
caught this attack but they caught it
00:15:40.817 --> 00:15:43.539
after the fact right yeah so you mean
00:15:43.559 --> 00:15:44.980
has it already got better so what you're
00:15:45.000 --> 00:15:46.961
saying is john we need to get quen
00:15:47.403 --> 00:15:48.943
it's now we didn't get quen john i'm
00:15:48.964 --> 00:15:51.645
telling you it's basically you know went
00:15:51.686 --> 00:15:53.187
up right now let's go and get it
00:15:53.486 --> 00:15:54.047
Yeah.
00:15:54.126 --> 00:15:55.187
You know, in Scooby-Doo,
00:15:55.248 --> 00:15:56.208
at the end of it,
00:15:56.229 --> 00:15:58.049
when they catch the perpetrator and they
00:15:58.070 --> 00:16:00.551
take the mask off, they take, like,
00:16:00.571 --> 00:16:01.631
the Quinn's mask off.
00:16:02.032 --> 00:16:03.312
Actually, underneath it's just Opus.
00:16:03.832 --> 00:16:04.833
So I think that's what it is.
00:16:04.854 --> 00:16:05.614
Oh, yeah, underneath it.
00:16:05.634 --> 00:16:06.534
You know, it's one of them.
00:16:06.554 --> 00:16:08.275
So it's just Opus.
00:16:08.296 --> 00:16:09.176
Oh, no, it'll be Mythos.
00:16:09.535 --> 00:16:10.677
That's the illusion,
00:16:10.736 --> 00:16:11.998
because we don't know Mythos,
00:16:12.018 --> 00:16:13.119
because we've never heard of it.
00:16:13.178 --> 00:16:14.940
The mask illusion, that is Mythos.
00:16:15.080 --> 00:16:16.860
And you actually see, like, Kaiser Soze.
00:16:18.143 --> 00:16:18.903
Yeah, exactly.
00:16:19.123 --> 00:16:20.964
But you know, the thing here is, John,
00:16:21.183 --> 00:16:22.004
yes, over time,
00:16:22.024 --> 00:16:23.585
they've improved their anomaly detection,
00:16:23.625 --> 00:16:24.745
trying to figure out because it's kind of
00:16:24.764 --> 00:16:26.725
hard to figure out these attacks and the
00:16:26.785 --> 00:16:27.826
patterns that they exist.
00:16:27.865 --> 00:16:29.706
So they can see how they how they
00:16:29.765 --> 00:16:30.886
catch these people, right?
00:16:30.907 --> 00:16:31.647
People were doing this.
00:16:32.466 --> 00:16:34.187
But imagine all the other people who are
00:16:34.207 --> 00:16:35.248
getting away with distillation.
00:16:36.355 --> 00:16:37.835
All the other sports.
00:16:38.336 --> 00:16:39.296
I'm not saying it's happening.
00:16:39.615 --> 00:16:41.437
Perhaps some of the other competitors.
00:16:41.496 --> 00:16:44.898
That sounds so guilty right now.
00:16:44.957 --> 00:16:45.577
That was such a,
00:16:46.118 --> 00:16:47.538
the way you phrased that was like a
00:16:47.677 --> 00:16:49.038
real,
00:16:49.519 --> 00:16:51.119
it was like you really wanted to confess,
00:16:51.158 --> 00:16:53.080
but had to hold yourself back on the
00:16:53.100 --> 00:16:54.779
fact that you've got something going down.
00:16:55.081 --> 00:16:56.081
Is that what you've been up to?
00:16:56.976 --> 00:16:57.277
Listen,
00:16:57.376 --> 00:16:59.097
I'm not going to say any of the
00:16:59.558 --> 00:17:00.938
SAR models that are going to come out.
00:17:01.340 --> 00:17:02.620
I'm not talking about any of that.
00:17:02.980 --> 00:17:04.101
You'll find out later on.
00:17:04.882 --> 00:17:05.801
I'm just telling you, John.
00:17:06.603 --> 00:17:09.903
SAR AI.
00:17:09.923 --> 00:17:12.405
I'm just telling you that this kind of
00:17:12.425 --> 00:17:14.086
attack is... But again,
00:17:14.307 --> 00:17:15.748
the model is there for the usage.
00:17:16.288 --> 00:17:18.628
If you stick within fair usage policy,
00:17:19.950 --> 00:17:21.131
you're learning from the teacher, John.
00:17:22.251 --> 00:17:22.832
Is that legal?
00:17:24.912 --> 00:17:26.752
Yeah, they learned from the best.
00:17:26.772 --> 00:17:29.875
They copied their model and was like, oh,
00:17:30.255 --> 00:17:31.496
you can steal and train.
00:17:32.496 --> 00:17:34.396
Yeah, let's follow suit.
00:17:34.517 --> 00:17:35.698
I do think, though,
00:17:36.337 --> 00:17:37.479
somebody was using Quen.
00:17:37.499 --> 00:17:38.598
That's why in the back of my mind,
00:17:38.618 --> 00:17:40.019
because I remember having a conversation
00:17:40.039 --> 00:17:41.941
around somebody was using the Quen open
00:17:42.000 --> 00:17:43.481
source model for quite a lot of their
00:17:44.622 --> 00:17:47.144
coding and then there was some internal
00:17:47.163 --> 00:17:48.704
discussions not to go off on too much
00:17:48.744 --> 00:17:50.306
of a tangent but there's internal
00:17:50.326 --> 00:17:53.548
conversations around actually you could
00:17:53.627 --> 00:17:56.970
conserve your money by running something
00:17:57.009 --> 00:17:58.671
like quen to kind of get going to
00:17:58.691 --> 00:18:00.372
a certain extent say we're going to write
00:18:00.412 --> 00:18:01.732
some code or whatever you're going to do
00:18:01.752 --> 00:18:04.795
you could use your local quen model get
00:18:04.835 --> 00:18:07.356
it so far you know to iterate to
00:18:07.376 --> 00:18:10.077
a certain point and then use the polish
00:18:10.479 --> 00:18:11.859
and the rest of it maybe like the
00:18:13.151 --> 00:18:14.132
getting it reviewed,
00:18:14.211 --> 00:18:16.034
making those changes with something then
00:18:16.094 --> 00:18:18.234
like Anthropic so that actually you can
00:18:19.596 --> 00:18:22.398
kind of save your costs by trying to
00:18:22.439 --> 00:18:23.359
get to a
certain point.
00:18:23.380 --> 00:18:24.961
So you don't obviously hemorrhage tokens.
00:18:24.980 --> 00:18:26.102
And actually some people have already
00:18:26.122 --> 00:18:28.344
started to do that as a bit of
00:18:28.865 --> 00:18:29.625
a way of working,
00:18:30.046 --> 00:18:31.046
which is kind of interesting.
00:18:31.767 --> 00:18:33.228
you know what john there's probably a very
00:18:33.248 --> 00:18:36.328
good topic to discuss at a future uh
00:18:36.588 --> 00:18:38.108
cloud and plug right okay i get it
00:18:38.209 --> 00:18:39.329
i can i take the hint i take
00:18:39.369 --> 00:18:40.490
the hits on them and that's all right
00:18:40.529 --> 00:18:42.590
how do we how do we i'll drop
00:18:42.631 --> 00:18:45.711
that no sorry sorry um i'll drop that
00:18:45.832 --> 00:18:46.672
just for another time
00:18:47.833 --> 00:18:50.415
How do we efficiently use our tokens,
00:18:50.435 --> 00:18:50.596
right?
00:18:50.615 --> 00:18:51.477
Because you're absolutely right.
00:18:51.497 --> 00:18:52.659
It's all about the tokens, John.
00:18:53.240 --> 00:18:54.701
And if you can use a cheaper model,
00:18:55.342 --> 00:18:56.423
it's all about them tokens.
00:18:56.983 --> 00:18:58.246
I need to mark the token max.
00:18:59.106 --> 00:19:01.388
If we can use some of these cheaper
00:19:01.429 --> 00:19:04.073
models to do most of the groundwork, then,
00:19:04.093 --> 00:19:05.273
of course, why we can't polish it.
00:19:06.154 --> 00:19:07.115
I kind of do the same with...
00:19:07.696 --> 00:19:09.258
With Claude, I use Haiku,
00:19:09.298 --> 00:19:11.837
which is much cheaper than Opus to get
00:19:11.857 --> 00:19:13.419
going with a bunch of stuff.
00:19:13.459 --> 00:19:15.380
And then I switch over to Opus when
00:19:15.839 --> 00:19:17.480
Haiku starts to lag a little bit and
00:19:17.559 --> 00:19:18.740
it's not really doing what it's supposed
00:19:18.760 --> 00:19:20.541
to do.
00:19:20.642 --> 00:19:23.001
Because the Claude models,
00:19:23.021 --> 00:19:25.103
they're not five cents per million input
00:19:25.143 --> 00:19:25.883
tokens, right, John?
00:19:25.923 --> 00:19:28.463
So I have to do something about that.
00:19:29.085 --> 00:19:30.005
Yeah.
00:19:30.065 --> 00:19:32.346
But what about, I guess,
00:19:32.766 --> 00:19:35.826
moving on to MacBooks?
00:19:37.731 --> 00:19:39.153
And the fact that they're going to cost
00:19:39.173 --> 00:19:40.255
twenty percent more.
00:19:41.036 --> 00:19:42.837
I mean, they weren't cheap.
00:19:44.159 --> 00:19:49.144
MacBooks are more on the expensive end of
00:19:49.184 --> 00:19:49.925
the price spectrum,
00:19:50.026 --> 00:19:52.848
but they're going to go up by roughly
00:19:52.909 --> 00:19:54.069
fifteen to twenty percent.
00:19:54.411 --> 00:19:55.971
Fifteen to twenty five percent this week,
00:19:55.991 --> 00:19:56.292
they said.
00:19:57.422 --> 00:19:59.623
Um, all because of the AI demand,
00:19:59.663 -->
00:20:00.442
obviously the, the,
00:20:00.583 --> 00:20:02.483
the memory and the chips that are in
00:20:02.503 --> 00:20:05.085
your laptop also go into the AI servers.
00:20:05.105 --> 00:20:07.385
Now you've got the competition obviously
00:20:07.405 --> 00:20:09.487
happening on a shortage, um,
00:20:09.606 --> 00:20:11.147
of like memory, et cetera.
00:20:11.667 --> 00:20:13.147
And that's obviously put the demand up and
00:20:13.188 --> 00:20:14.848
obviously supply and demand it's then made
00:20:14.868 --> 00:20:15.628
the cost higher.
00:20:15.689 --> 00:20:18.230
And therefore this is a trickle down into,
00:20:18.390 --> 00:20:20.830
um, into obviously the, the,
00:20:20.951 --> 00:20:22.551
the laptop market or in this case,
00:20:22.592 --> 00:20:24.272
MacBook, but.
00:20:25.201 --> 00:20:26.442
What do you think about that as in
00:20:26.502 --> 00:20:28.565
now consumers are starting to feel the
00:20:28.625 --> 00:20:31.428
pinch on this AI hype?
00:20:32.449 --> 00:20:34.070
Yeah, John, supply and demand, right?
00:20:34.090 --> 00:20:35.912
Because the hyperscalers are buying up all
00:20:35.932 --> 00:20:37.713
this memory that goes in not just
00:20:37.753 --> 00:20:40.497
MacBooks, by the way, it's also iPads,
00:20:40.576 --> 00:20:41.917
iPad Pros and all of them.
00:20:41.998 --> 00:20:43.219
Everything has gone up by like,
00:20:43.500 --> 00:20:45.382
twenty percent of the cost.
00:20:46.041 --> 00:20:47.123
So the RAM
00:20:48.064 --> 00:20:50.484
hyperscalers are buying everything right
00:20:50.525 --> 00:20:51.986
so they're buying the gpu chips they're
00:20:52.006 --> 00:20:55.327
buying the rams so supply and demand
00:20:55.347 --> 00:20:56.848
there's not enough memory available in the
00:20:56.868 --> 00:20:59.329
market for to be put into these these
00:20:59.349 --> 00:21:01.332
machines it's costing a bit more money for
00:21:01.412 --> 00:21:04.192
apple to build it and the problem here
00:21:04.252 --> 00:21:05.054
is it's not just
00:21:05.874 --> 00:21:06.534
just Apple,
00:21:06.693 --> 00:21:08.775
pretty soon we'll see this trickle down
00:21:08.795 --> 00:21:09.796
into other devices, too,
00:21:10.155 --> 00:21:12.497
and probably phones as well at some point.
00:21:12.997 --> 00:21:16.019
So unless the bubble bursts and people
00:21:16.038 --> 00:21:21.681
start realizing that bubble is expanding
00:21:21.760 --> 00:21:23.981
way more than it should,
00:21:24.342 --> 00:21:25.462
I think the price is just going to
00:21:25.482 --> 00:21:25.962
keep going up.
00:21:26.182 --> 00:21:28.124
So this is basically what we're doing,
00:21:28.144 --> 00:21:29.565
John, you and I.
00:21:30.345 --> 00:21:32.306
If we buy an iPad or a Mac,
00:21:32.806 --> 00:21:34.346
we're subsidising the arms race.
00:21:34.366 --> 00:21:34.948
That's what we're doing.
00:21:35.107 --> 00:21:35.748
You and I, John.
00:21:36.327 --> 00:21:37.689
We are responsible for this.
00:21:37.709 --> 00:21:38.529
The booze.
00:21:38.589 --> 00:21:39.990
Do you remember the last time you were
00:21:40.030 --> 00:21:41.991
saying that they spoke to students and
00:21:42.051 --> 00:21:44.133
obviously with the job market and job
00:21:44.173 --> 00:21:45.493
market threat and I think it was Eric
00:21:45.513 --> 00:21:46.153
Schmidt, wasn't it?
00:21:46.213 --> 00:21:48.675
Or something was doing his talk and
00:21:48.695 --> 00:21:50.836
everyone was booing him every time he
00:21:50.856 --> 00:21:51.636
mentioned AI.
00:21:51.676 --> 00:21:52.117
And now...
00:21:53.055 --> 00:21:55.298
just the general public is going to start
00:21:55.337 --> 00:21:56.378
booing because they'll be like, right,
00:21:56.398 --> 00:21:57.660
I'm going to get the new iPhone or
00:21:57.680 --> 00:21:58.580
I'm going to go and get the new
00:21:58.621 --> 00:22:01.124
MacBook or I'm going to go and whatever
00:22:01.163 --> 00:22:01.585
device.
00:22:02.164 --> 00:22:03.666
And then they're going to realize it's
00:22:03.686 --> 00:22:05.128
like three or four hundred pounds more
00:22:05.169 --> 00:22:06.569
expensive than it was last year.
00:22:06.730 --> 00:22:07.751
And then they're going to be booing and
00:22:07.771 --> 00:22:08.271
be like, yeah,
00:22:08.291 --> 00:22:08.412
that's
00:22:09.298 --> 00:22:09.978
That's because of AI, mate.
00:22:09.998 --> 00:22:14.339
They were like, well,
00:22:14.740 --> 00:22:16.900
you can use AI to create all the
00:22:16.920 --> 00:22:19.040
funny images that you wanted, but sorry,
00:22:19.060 --> 00:22:19.560
unfortunately,
00:22:19.580 --> 00:22:20.461
you're going to have to pay a bit
00:22:20.541 --> 00:22:20.741
more.
00:22:20.922 --> 00:22:21.342
Exactly.
00:22:21.961 --> 00:22:24.143
Now you're feeling the pinch.
00:22:24.222 --> 00:22:24.462
Yeah,
00:22:24.542 --> 00:22:26.624
it's a bit crazy when it starts to
00:22:26.644 --> 00:22:27.703
hit consumers like that,
00:22:27.743 --> 00:22:31.744
as in the trickle down effect that's kind
00:22:31.785 --> 00:22:33.445
of happening now because of the demand.
00:22:33.506 --> 00:22:34.205
And that's a little bit.
00:22:37.717 --> 00:22:39.157
Crazy, really.
00:22:39.178 --> 00:22:39.698
Correct.
00:22:39.817 --> 00:22:43.000
I mean, we also see the models themselves.
00:22:43.060 --> 00:22:44.041
Slightly over time,
00:22:44.061 --> 00:22:45.103
they'll get more expensive.
00:22:45.403 --> 00:22:47.585
We've seen co-pilot go into usage-based
00:22:47.625 --> 00:22:48.164
consumption.
00:22:48.384 --> 00:22:48.484
Well,
00:22:48.505 --> 00:22:50.807
that's okay because you are in the market
00:22:50.846 --> 00:22:52.288
for using the model, so you understand,
00:22:52.348 --> 00:22:52.509
okay,
00:22:52.528 --> 00:22:53.388
I have to pay a little bit more.
00:22:54.450 --> 00:22:55.931
But people who are buying iPads and
00:22:55.951 --> 00:22:56.250
laptops,
00:22:56.270 --> 00:22:58.413
they're not in the market for models.
00:22:58.712 --> 00:22:59.492
Not everybody, John,
00:22:59.813 --> 00:23:00.814
as much as you and I like to
00:23:00.834 --> 00:23:02.275
think, not everybody's using AI.
00:23:02.295 --> 00:23:02.434
Exactly.
00:23:02.454 --> 00:23:04.556
Do you know, though,
00:23:04.875 --> 00:23:08.739
how you can make your money back on
00:23:08.798 --> 00:23:09.759
all of this if you want to make
00:23:09.778 --> 00:23:10.199
some money?
00:23:10.259 --> 00:23:12.181
You're a content guy,
00:23:13.961 --> 00:23:15.261
so you can produce your content.
00:23:15.301 --> 00:23:16.442
You can now monetize it.
00:23:17.423 --> 00:23:19.005
You can now monetize your content.
00:23:19.884 --> 00:23:21.065
And you can make your money back.
00:23:21.345 --> 00:23:23.767
So all the AI bots that are scraping
00:23:23.787 --> 00:23:24.846
your content and stealing it,
00:23:24.866 --> 00:23:27.249
you can gate it now and get some
00:23:27.288 --> 00:23:27.648
payment.
00:23:27.689 --> 00:23:29.931
Get that money's coming in for all that
00:23:30.010 --> 00:23:31.971
content to pay for your new MacBook that
00:23:31.991 --> 00:23:34.354
the AI was stealing to begin with.
00:23:35.153 --> 00:23:37.536
Or maybe you're using AI to make the
00:23:37.576 --> 00:23:40.156
content that's then making your MacBook
00:23:40.237 --> 00:23:41.897
expensive to obviously create it.
00:23:41.919 --> 00:23:43.398
So then you've obviously got to get the
00:23:43.440 --> 00:23:44.640
money back to pay for it.
00:23:46.260 --> 00:23:47.021
You caught me.
00:23:47.061 --> 00:23:47.741
You caught me, John.
00:23:47.961 --> 00:23:48.442
You caught me.
00:23:49.326 --> 00:23:50.407
It's an interesting time.
00:23:50.468 --> 00:23:50.749
Yeah.
00:23:51.328 --> 00:23:52.871
Interesting times and the money game.
00:23:53.171 --> 00:23:55.173
But yeah.
00:23:55.212 --> 00:23:56.454
Sorry about that.
00:23:56.515 --> 00:23:57.275
Amazon, yeah,
00:23:57.375 --> 00:23:59.076
have also released something because
00:23:59.597 --> 00:24:01.279
they've been in New York talking about
00:24:01.319 --> 00:24:01.480
that.
00:24:01.539 --> 00:24:02.320
So I just thought it was a bit
00:24:02.361 --> 00:24:07.465
of a slight interlink between costs and
00:24:07.526 --> 00:24:08.686
money and things like that.
00:24:08.707 --> 00:24:10.729
I thought I'd weave that one in.
00:24:11.769 --> 00:24:15.111
um so i guess really we need to
00:24:15.151 --> 00:24:18.633
get on quen um asap i think is
00:24:18.653 --> 00:24:21.733
the the summary save your anthropic money
00:24:22.733 --> 00:24:24.694
go and give it to alibaba because they've
00:24:24.734 --> 00:24:26.095
pretty much stolen everything from
00:24:26.115 --> 00:24:28.096
anthropic anyway so don't even need to you
00:24:28.115 --> 00:24:29.596
don't need to be spending all that cash
00:24:29.635 --> 00:24:32.978
now on uh anthropic you can now invest
00:24:32.998 -->
00:24:36.058
that into a five cents price point with
00:24:36.078 --> 00:24:38.640
uh with using quen and um
00:24:39.680 --> 00:24:41.541
save that cash so you can buy a
00:24:41.582 --> 00:24:42.622
MacBook.
00:24:42.682 --> 00:24:43.083
Exactly.
00:24:43.123 --> 00:24:44.284
That money has to come from somewhere,
00:24:44.364 --> 00:24:44.544
John.
00:24:44.983 --> 00:24:45.684
The extra two,
00:24:45.704 --> 00:24:46.904
three hundred pounds has to come from
00:24:46.944 --> 00:24:47.286
somewhere.
00:24:47.806 --> 00:24:49.827
So might as well stop using the Opus
00:24:49.847 --> 00:24:51.229
models and use the Quen models.
00:24:51.328 --> 00:24:52.849
I have used the deep seek models before.
00:24:53.549 --> 00:24:55.050
Absolutely fine.
00:24:55.090 --> 00:24:56.732
But I haven't used them in anger as
00:24:56.792 --> 00:24:56.992
much.
00:24:57.613 --> 00:24:58.753
Perhaps it's time for me to have a
00:24:58.773 --> 00:24:59.815
look into the Quen models.
00:25:00.776 --> 00:25:01.736
Yeah, to be fair, I haven't used.
00:25:01.756 --> 00:25:02.797
I do know people that have been,
00:25:02.817 --> 00:25:03.877
but I haven't used it either.
00:25:05.713 --> 00:25:07.275
cool so the next episode i think we
00:25:07.315 --> 00:25:08.556
said though we were going to break down
00:25:08.615 --> 00:25:09.958
a story and we've cheated and we've
00:25:09.978 --> 00:25:11.679
basically done news again i'm pretty sure
00:25:11.699 --> 00:25:14.060
in the last episode we said next episode
00:25:14.080 --> 00:25:16.282
we're gonna you know go deep on a
00:25:16.343 --> 00:25:18.183
topic and then we've like next episode
00:25:18.204 --> 00:25:19.766
arrives and we haven't so apologies for
00:25:19.806 --> 00:25:22.167
that that's just us uh being a bit
00:25:22.188 --> 00:25:23.689
short for time but i think next time
00:25:23.709 --> 00:25:26.391
we are going to get into a very
00:25:26.451 --> 00:25:28.292
specific topic i think we're talking about
00:25:29.153 --> 00:25:29.314
um
00:25:30.676 --> 00:25:32.799
the trust of an AI agent and what
00:25:32.819 --> 00:25:34.501
goes into it and how can you,
00:25:35.063 --> 00:25:36.826
from a security perspective,
00:25:36.865 --> 00:25:37.946
people are using it today,
00:25:39.596 --> 00:25:40.758
What does it really mean when you're
00:25:40.798 --> 00:25:43.420
giving it access to things that you have
00:25:44.000 --> 00:25:46.781
and what's the imperative risk there to
00:25:46.821 --> 00:25:46.961
you?
00:25:46.981 --> 00:25:47.823
And then what does it mean to a
00:25:47.863 --> 00:25:50.924
business as well when you have not just
00:25:51.005 --> 00:25:51.945
you as an individual,
00:25:51.986 --> 00:25:56.229
but your entire workforce all in mass
00:25:56.489 --> 00:25:58.230
using AI and connecting things up?
00:25:58.290 --> 00:26:01.673
And is it quite a risky thing to
00:26:01.692 --> 00:26:03.253
just roll out or are there ways of
00:26:03.314 --> 00:26:05.316
protecting it so you can get the value
00:26:05.355 --> 00:26:05.635
out of it?
00:26:07.224 --> 00:26:09.067
So that will be probably what we're going
00:26:09.086 --> 00:26:10.907
to discuss next time, all being well.
00:26:11.107 --> 00:26:11.969
No promises, though.
00:26:13.150 --> 00:26:14.891
But that's where we want to be for
00:26:14.931 --> 00:26:17.432
the next episode.
00:26:17.472 --> 00:26:18.354
Cool.
00:26:18.374 --> 00:26:18.673
Cool.
00:26:18.854 --> 00:26:19.134
All right.
00:26:19.255 --> 00:26:22.136
Well, a bit short and sweet today,
00:26:22.176 --> 00:26:24.499
but we should be back next week with
00:26:24.519 --> 00:26:26.500
a bit of a deep dive around Agentic.
00:26:26.519 --> 00:26:29.321
Cool.
00:26:29.362 --> 00:26:29.663
Thank you.
00:26:29.682 --> 00:26:30.002
Bye-bye.