WEBVTT

00:00:00.000 --> 00:00:06.560
On today's AI briefing, we think the S1 shows a real business carrying financing risk.

00:00:07.100 --> 00:00:14.060
$16.3 billion in half-year revenue against an $80.7 billion net loss

00:00:14.060 --> 00:00:19.080
and $400 billion of compute obligations not counted as debt.

00:00:20.660 --> 00:00:24.660
This is the AI briefing for Thursday, October 29.

00:00:25.260 --> 00:00:25.840
I'm Michael.

00:00:25.840 --> 00:00:31.280
Like everything here, this briefing is written and read by AI from public sources.

00:00:32.020 --> 00:00:34.640
Today's stories are about the money behind AI.

00:00:35.380 --> 00:00:43.620
Anthropics S1 shows $16.3 billion in first-half revenue next to an $80.7 billion net loss,

00:00:44.000 --> 00:00:47.160
and Senator Warren wants answers on the risks it lists.

00:00:48.440 --> 00:00:53.060
Microsoft booked $10 billion from its OpenAI partnerships in a single quarter,

00:00:53.060 --> 00:00:56.120
while the amount OpenAI owes it keeps climbing.

00:00:56.880 --> 00:01:02.060
Alphabet's search growth slowed to 12%, with AI Mode cited as a cause,

00:01:02.440 --> 00:01:05.900
and Google is rolling AI Mode ads out globally anyway.

00:01:06.560 --> 00:01:13.220
And Warp Sciences came out of stealth with $215 million to build a robotics-driven chemistry lab

00:01:13.220 --> 00:01:16.660
in the U.S. mainstories first, then a few quick hits.

00:01:16.660 --> 00:01:20.700
Stephen's take on the top of the day.

00:01:21.340 --> 00:01:24.600
AI mode will be monetized increasingly over time.

00:01:25.240 --> 00:01:30.200
The combination of facts has value, and it is unlikely to be devalued to zero,

00:01:30.860 --> 00:01:37.460
unless accuracy and reliability in human validation becomes unnecessary, which is a possibility.

00:01:39.220 --> 00:01:43.400
The first story is an update on one we've been following since last week,

00:01:43.900 --> 00:01:46.180
Anthropik's filing to list on NASDAQ.

00:01:46.180 --> 00:01:52.060
The S-1 reports $16.3 billion in revenue for the first half of 2026.

00:01:52.700 --> 00:01:55.880
That's up $15.1 billion year over year.

00:01:56.560 --> 00:01:58.620
BetaKit points to the other side of the ledger.

00:01:59.240 --> 00:02:02.680
There's an $80.7 billion net loss for the half,

00:02:03.000 --> 00:02:08.180
and $400 billion in future compute obligations that aren't counted as debt.

00:02:09.420 --> 00:02:14.920
And Senator Warren is now demanding answers on the existential risks the filing describes.

00:02:14.920 --> 00:02:18.580
On the evidence, these figures remain unverified.

00:02:19.180 --> 00:02:24.660
Any claim that Anthropic is profitable mixes up GAAP results with adjusted ones.

00:02:25.160 --> 00:02:26.400
Our view hasn't moved.

00:02:26.900 --> 00:02:29.660
This is a real business carrying financing risk.

00:02:30.200 --> 00:02:36.460
The open question is who funds that $400 billion if growth slows, and we lean cautious.

00:02:38.360 --> 00:02:39.900
Here's Stephen's take.

00:02:39.900 --> 00:02:45.120
With the pace of free good, it is hard to see how expensive better survives at this capex,

00:02:45.420 --> 00:02:50.940
mho. Increasingly, people are going to realize they are able to generate the value they need

00:02:50.940 --> 00:02:56.980
at a very low cost in software, services, and answers, and the very expensive models

00:02:56.980 --> 00:03:01.440
might become planners and coordinators in a much smaller value capture market.

00:03:02.160 --> 00:03:05.620
We have always thought that, and the power of free is amazing.

00:03:05.620 --> 00:03:10.200
From OneLab's books to a cloud providers

00:03:10.200 --> 00:03:15.860
Our second story is Microsoft's latest quarter and what it shows about OpenAI

00:03:15.860 --> 00:03:23.940
TechCrunch reports that Microsoft brought in $92.3 billion in revenue for its fourth quarter of 2026

00:03:23.940 --> 00:03:27.260
That's growth of 19% year over year

00:03:27.260 --> 00:03:31.380
Azure accounted for $33.2 billion of it

00:03:31.380 --> 00:03:33.640
The striking line is OpenAI

00:03:33.640 --> 00:03:41.360
Microsoft booked $10 billion from its OpenAI partnerships, nearly 11% of total revenue.

00:03:41.980 --> 00:03:46.740
Over the same period, receivables from OpenAI rose to $7.2 billion.

00:03:47.760 --> 00:03:50.460
Receivables are money billed but not yet collected.

00:03:51.000 --> 00:03:52.360
So how solid is this?

00:03:52.840 --> 00:03:56.120
The headline figures can be checked against Microsoft's filings.

00:03:56.120 --> 00:03:57.980
The split is not confirmed.

00:03:57.980 --> 00:04:04.320
We don't know how much of the Azure number comes from OpenAI revenue sharing and how much from other customers.

00:04:05.280 --> 00:04:08.000
This is exactly the pattern we watch for.

00:04:08.460 --> 00:04:15.380
A cloud provider books $10 billion a quarter from one AI lab, while what that lab owes it keeps rising.

00:04:16.100 --> 00:04:22.640
In our view, that unconfirmed split between Azure and OpenAI decides how much of this growth is circular.

00:04:23.300 --> 00:04:27.000
There's a smaller version of the same question for early-stage companies.

00:04:27.000 --> 00:04:32.000
Some infrastructure startups count AI Labs as their biggest customers.

00:04:33.000 --> 00:04:39.260
For them, receivables growth like this makes payment terms and days outstanding a fair diligence point.

00:04:39.680 --> 00:04:41.540
The market reads it more kindly.

00:04:42.240 --> 00:04:46.180
Microsoft grew 19% for a tenth consecutive record quarter.

00:04:46.180 --> 00:04:57.200
On that reading, OpenAI is simply a large and growing account, and receivables rising alongside revenue is normal for a customer scaling that fast.

00:04:58.860 --> 00:05:01.640
Here's Stephen's take, Round and Round.

00:05:03.220 --> 00:05:08.940
The third story stays with big tech earnings, and with what it costs to turn search into AI.

00:05:09.740 --> 00:05:13.420
Alphabet reported its third quarter, and TechMeme gathered the numbers.

00:05:13.420 --> 00:05:21.820
Free cash flow was negative $9.07 billion. Capital spending came in at $51.36 billion,

00:05:22.540 --> 00:05:32.060
below the $55.09 billion estimate. Alphabet kept its full-year capex forecast at $195 to $205

00:05:32.060 --> 00:05:40.240
billion. The more telling number is search. Search revenue grew 12% to $63.25 billion,

00:05:40.240 --> 00:05:46.920
A year earlier in the third quarter of 2025, that growth rate was 14.5%.

00:05:46.920 --> 00:05:50.740
The cause cited was lower monetization in AI mode.

00:05:51.280 --> 00:05:56.940
Meanwhile, TechCrunch and TechMeme both report that Google has expanded AI mode ads globally.

00:05:57.580 --> 00:05:58.500
This one holds up.

00:05:59.060 --> 00:06:05.300
CapEx is high but under the estimate, and the slowdown is attributed directly to AI mode earning less.

00:06:06.020 --> 00:06:09.960
The global ad rollout looks like a response to slower core growth.

00:06:10.240 --> 00:06:12.660
It hasn't yet been shown to offset the decline.

00:06:13.280 --> 00:06:17.400
Our view is that, for now, this cuts against the incumbent's turn to AI.

00:06:18.120 --> 00:06:21.500
Alphabet's own AI wedge is cited as slowing its search growth.

00:06:22.080 --> 00:06:24.680
The global AI mode ad rollout is the test.

00:06:25.160 --> 00:06:29.560
There's a practical angle for startups that find customers through Google search ads.

00:06:30.120 --> 00:06:32.560
Pricing may shift as these ads go global,

00:06:32.560 --> 00:06:37.020
so channel mix and the trend in cost per acquisition seem worth checking.

00:06:37.020 --> 00:06:52.940
The market reads the same facts differently. Slower search growth during a format shift can be the cost of moving users first and pricing later. AI mode ads have only now gone global, and CapEx came in below estimate.

00:06:52.940 --> 00:07:12.760
Here's Stephen's take. Again, incumbents are in a strong position and they have to adapt to survive, and Google is adapting. They are also good at monetization, so there's that. People always do a bad job of combining the parts of valuation together for alphabet.

00:07:12.760 --> 00:07:19.540
The fourth and last of the main stories moves from big tech spending to a startup's raise.

00:07:20.180 --> 00:07:26.780
Yesterday, the robot report covered warp sciences coming out of stealth with $215 million in funding.

00:07:27.440 --> 00:07:34.260
The plan is a U.S.-based chemistry CRO, a contract research organization that biotech's outsource lab work to.

00:07:34.820 --> 00:07:39.240
It combines chemists, robotics, and AI to test molecules faster.

00:07:39.900 --> 00:07:45.240
The pitch leans on Sovereign Biotech, with the BIOSECURE Act in the background.

00:07:46.200 --> 00:07:47.740
How well does it hold up?

00:07:48.080 --> 00:07:51.620
The raise and the service plan come from the company, and they're plausible.

00:07:52.240 --> 00:07:56.560
The speed and cost advantages are still aspirations, with no data yet.

00:07:57.000 --> 00:08:00.820
We think a U.S.-based CRO built on robotics is a credible wedge.

00:08:00.820 --> 00:08:05.620
It gives biotechs weighing the BIOSECURE Act a domestic option.

00:08:05.620 --> 00:08:11.160
What would decide it is per-molecule turnaround in cost, and neither has been disclosed.

00:08:11.740 --> 00:08:16.260
Those are the numbers we'd want to see before believing it beats existing CROs.

00:08:16.900 --> 00:08:25.300
For early-stage investors, a $215 million raise for a services business sets a high capital bar.

00:08:26.100 --> 00:08:32.900
Seed-stage lab automation companies may do better selling into operators like this than competing with them.

00:08:32.900 --> 00:08:47.080
Here's Steven's take. Advances in robots in parallel with advances in research is a powerful cycle in physical sciences. This is going to be an era of amazing advance.

00:08:47.080 --> 00:08:55.860
Two quick ones before we close. Reuters reports that the US DOJ has asked how NVIDIA's hugging

00:08:55.860 --> 00:09:02.620
face deal affects competition in AI. Antitrust enforcers are still probing NVIDIA's conduct.

00:09:02.620 --> 00:09:09.540
It matters because NVIDIA's dealmaking is now under active scrutiny. And Google DeepMind has

00:09:09.540 --> 00:09:15.200
launched Gemini 4 Carbon for cyber defenders through what Latent Space calls the Fairwind

00:09:15.200 --> 00:09:20.660
program it matters because Google is putting this model in the hands of

00:09:20.660 --> 00:09:29.480
selected defenders first not everyone on symbiotics 90.5% reliance on Walmart

00:09:29.480 --> 00:09:35.420
raises customer concentration risk here Stevens take they bought Walmart's

00:09:35.420 --> 00:09:40.100
automation division and thus it is unsurprising they look like Walmart's

00:09:40.100 --> 00:09:48.140
automation division right now one question worth asking this week what share of revenue comes from

00:09:48.140 --> 00:09:54.860
the largest customer and is that customer also an investor or supplier and one to watch what

00:09:54.860 --> 00:10:02.100
explains the gap between anthropics 10% adjusted operating margin and its 80.7 billion dollar net

00:10:02.100 --> 00:10:09.680
loss s1 amendments the gaap reconciliation or analyst breakdowns before pricing should settle

00:10:09.680 --> 00:10:17.820
it. That's the briefing. Stay with me. Right after this, Bella takes you through today's deep dive

00:10:17.820 --> 00:10:24.080
on Sweden's chatbot block is about contracts, not countries. You can also read it or listen

00:10:24.080 --> 00:10:29.400
to it on its own on the AI Briefing webpage, where you'll find the written edition in every

00:10:29.400 --> 00:10:38.740
past episode. The link is in the show notes. Thanks, Michael. This is the AI Briefing deep

00:10:38.740 --> 00:10:44.720
dive for Thursday, October 29. Sweden's chatbot block is about contracts, not countries.

00:10:45.200 --> 00:10:50.300
I'm Bella. Like the briefing, it is written and read by AI from public sources.

00:10:52.100 --> 00:10:58.040
On Monday, October 26, Sweden's government offices blocked chat GPT and clawed on staff

00:10:58.040 --> 00:11:02.900
work computers. This is the body that supports the prime minister's office and every ministry.

00:11:03.560 --> 00:11:05.980
Staff are being sent to Microsoft's co-pilot instead.

00:11:06.760 --> 00:11:10.900
Six days earlier, the U.S. Department of Homeland Security published an order

00:11:10.900 --> 00:11:13.480
removing DeepSeek from federal civilian systems.

00:11:14.040 --> 00:11:16.700
But the two side by side and the headline writes itself,

00:11:17.180 --> 00:11:19.180
Governments are walling off foreign AI.

00:11:19.760 --> 00:11:20.820
Is that true, though?

00:11:21.660 --> 00:11:25.100
The first reports we saw gave no reason for the Swedish move.

00:11:25.660 --> 00:11:30.500
The reporting since then, which traces back to Sveria's radio, does give a reason.

00:11:30.500 --> 00:11:33.200
and the reason is not where the chatbot comes from.

00:11:33.740 --> 00:11:36.260
So today's question is narrower and more useful.

00:11:36.860 --> 00:11:38.880
When a government shuts out an AI tool,

00:11:39.260 --> 00:11:42.940
is it doing so over data handling, security, or politics?

00:11:43.540 --> 00:11:47.420
And what does each of those grounds mean for a company trying to sell to the public sector?

00:11:49.380 --> 00:11:51.280
Start with what Sweden actually did.

00:11:51.720 --> 00:11:56.240
It blocked the web versions of ChatGPT and Claude, the sites you open in a browser.

00:11:56.840 --> 00:11:58.060
It did not ban the models.

00:11:58.060 --> 00:12:01.440
A limited number of staff can still reach both through Intric,

00:12:01.800 --> 00:12:04.560
a Stockholm platform for building AI assistance.

00:12:05.200 --> 00:12:08.020
Intric says its platform runs on servers in Sweden,

00:12:08.380 --> 00:12:10.360
with no need for foreign cloud firms.

00:12:11.020 --> 00:12:14.340
Copilot was approved because, in the government office's words,

00:12:14.660 --> 00:12:16.800
they had checked that tool and made it safer.

00:12:17.700 --> 00:12:19.560
And here's what makes it interesting.

00:12:20.160 --> 00:12:24.520
Microsoft has said Copilot will continue to be powered by OpenAI's latest models.

00:12:25.200 --> 00:12:29.960
Microsoft has also added anthropic models to Microsoft 365 Copilot.

00:12:30.560 --> 00:12:35.320
So the same model families can stay in the building while the consumer front door gets locked.

00:12:35.880 --> 00:12:36.860
An analogy helps.

00:12:37.380 --> 00:12:41.680
Think of a ministry that bans staff from emailing documents to their personal Gmail.

00:12:42.420 --> 00:12:44.360
Nobody reads that as a ban on Google.

00:12:45.360 --> 00:12:49.420
It's a rule about which channel has a signed contract, an audit trail,

00:12:49.800 --> 00:12:51.700
and an admin who can change the settings.

00:12:52.460 --> 00:12:54.220
Sweden's block works the same way.

00:12:54.520 --> 00:13:00.500
The digitalization director, Mikkel Wedderkranz, wrote on the staff intranet that using AI services

00:13:00.500 --> 00:13:04.680
without formal contracts can have consequences for information security.

00:13:05.400 --> 00:13:08.760
He added that they had seen several examples of problematic use,

00:13:09.220 --> 00:13:12.620
covering both what was typed in and how the answers were handled.

00:13:13.500 --> 00:13:16.520
The U.S. DeepSeek order works on a different principle.

00:13:17.140 --> 00:13:21.140
It was issued under the Federal Acquisition Supply Chain Security Act of 2018,

00:13:21.140 --> 00:13:25.920
a law that lets the government exclude a vendor it judges to be a supply chain risk.

00:13:26.520 --> 00:13:28.200
The issue there is who the vendor is.

00:13:28.820 --> 00:13:32.600
DHS argues that sensitive information from U.S. government users

00:13:32.600 --> 00:13:35.940
could be reachable by Chinese authorities under Chinese law.

00:13:36.560 --> 00:13:41.280
The order covers DeepSeek's own products and third-party products built on DeepSeek.

00:13:42.180 --> 00:13:46.900
No contract fixes that because the objection is to the vendor's legal home.

00:13:47.400 --> 00:13:49.340
So the two moves have a different mechanism.

00:13:50.100 --> 00:13:52.920
Sweden is saying, go through a contracted channel.

00:13:53.480 --> 00:13:56.720
The US is saying, not this vendor through any channel.

00:13:58.600 --> 00:13:59.980
Here's what is established.

00:14:00.680 --> 00:14:06.280
The Swedish bloc took effect October 26 and covers browser access across the government offices.

00:14:06.480 --> 00:14:11.340
That comes from Sveria's radio's reporting and the director's own internet note,

00:14:11.720 --> 00:14:13.140
which several outlets quote.

00:14:13.820 --> 00:14:19.300
Sweden's tax agency and the National Board of Institutional Care already had similar blocs.

00:14:19.340 --> 00:14:22.880
The enforcement authority was preparing its own decision.

00:14:23.740 --> 00:14:27.660
An internal review at the Swedish Transport Administration earlier,

00:14:28.200 --> 00:14:34.120
this year found roughly 10 AI services being used in what was described as a widespread and risky way.

00:14:34.880 --> 00:14:37.320
On the U.S. side, the record is primary.

00:14:38.020 --> 00:14:42.980
The order was dated July 24 and published in the Federal Register on October 20.

00:14:43.620 --> 00:14:47.680
It applies to civilian agencies such as Treasury, State, and Justice.

00:14:47.680 --> 00:14:56.920
The Congressional Research Service notes that the Defense Department and the Director of National Intelligence issued their own orders on September 8.

00:14:57.640 --> 00:15:01.320
DHS wrote that DeepSeq's models present multiple threat vectors.

00:15:02.180 --> 00:15:07.340
DeepSeq's own privacy policy says it stores users' personal data in the People's Republic of China.

00:15:07.980 --> 00:15:09.580
There's also a third data point.

00:15:09.580 --> 00:15:16.920
In February, the European Parliament's IT department turned off built-in AI features on lawmakers' devices.

00:15:17.680 --> 00:15:21.300
That covered Copilot along with ChatGPT and Claude.

00:15:21.880 --> 00:15:26.220
Its email said it could not guarantee the security of data uploaded to AI companies' servers.

00:15:26.960 --> 00:15:28.560
Now for what is claimed but not shown.

00:15:29.120 --> 00:15:33.400
We have no examples of what Swedish staff actually leaked, only that problems were seen.

00:15:33.940 --> 00:15:35.460
We don't know how many people are affected.

00:15:36.040 --> 00:15:40.160
We also don't know what Copilot settings were changed, or whether data stays in Sweden.

00:15:40.860 --> 00:15:45.480
On the US side, the order does not define what DeepSeek implementations means.

00:15:45.480 --> 00:15:50.000
Does a self-hosted copy of an open-weight deep-seek model count?

00:15:50.460 --> 00:15:51.660
What about a fine-tune of one?

00:15:52.120 --> 00:15:55.400
The order doesn't say, and the CRS summary doesn't settle it.

00:15:55.800 --> 00:15:56.640
One more caution.

00:15:57.240 --> 00:16:01.740
The idea that U.S. law lets American authorities demand chatbot user data

00:16:01.740 --> 00:16:04.420
was TechCrunch's framing of the parliament's story.

00:16:05.020 --> 00:16:08.880
The parliament's own email didn't say it, so the evidence is lopsided.

00:16:09.500 --> 00:16:13.000
The contract and data handling reading for Sweden is well supported.

00:16:13.000 --> 00:16:17.020
A political reading is something we'd be inferring,

00:16:17.400 --> 00:16:19.400
not something anyone in Sweden stated.

00:16:21.240 --> 00:16:23.620
Each reading starts from the same facts.

00:16:24.160 --> 00:16:26.540
The AGI forward reading focuses on cost.

00:16:27.040 --> 00:16:29.100
If frontier models keep improving fast,

00:16:29.460 --> 00:16:33.820
a ministry that limits staff to one approved assistant gives up capability every quarter.

00:16:34.500 --> 00:16:39.040
On that view, blocks like this push usage into shadow AI on personal phones,

00:16:39.480 --> 00:16:41.660
which is exactly the leak they were meant to stop.

00:16:41.660 --> 00:16:47.920
the a realist reading says this is ordinary it governance catching up an audit found about 10

00:16:47.920 --> 00:16:54.400
unmanaged services the organization then picked one under contract and blocked the rest nothing

00:16:54.400 --> 00:16:59.820
about it depends on what the models can do the market reading points to who won microsoft already

00:16:59.820 --> 00:17:05.460
had the enterprise agreement the admin controls and the security review on file it ended up as

00:17:05.460 --> 00:17:11.180
the only approved tool while serving open ai's models distribution and contracts beat model

00:17:11.180 --> 00:17:14.920
quality here, and the incumbent that holds the paperwork collects the seat.

00:17:15.420 --> 00:17:17.200
The policy reading has two layers.

00:17:18.560 --> 00:17:22.960
In Europe, data sovereignty is a rising theme, and Intric's pitch of Swedish

00:17:22.960 --> 00:17:25.700
servers with no foreign cloud is aimed straight at it.

00:17:26.200 --> 00:17:30.240
In the US, the DeepSeek order is explicitly about the vendor's country.

00:17:30.980 --> 00:17:34.760
China's foreign ministry called it politicizing technology and trade issues.

00:17:35.380 --> 00:17:39.200
Even so, the Swedish move doesn't support a reading of Europe banning

00:17:39.200 --> 00:17:47.120
american ai the european parliament blocked an american incumbent too sweden kept one for now

00:17:47.120 --> 00:17:51.040
the shared thread is controlled data flows not who owns the vendor

00:17:53.120 --> 00:17:57.440
our read is that this is a pattern but mostly not the one the headline suggests

00:17:57.440 --> 00:18:03.520
two gates are forming the first and most common is the contract is there a data processing agreement

00:18:03.520 --> 00:18:08.960
admin control and an audit trail any vendor can clear that gate in principle though incumbents

00:18:08.960 --> 00:18:14.880
already on the approved list have a head start measured in years the second gate is origin so

00:18:14.880 --> 00:18:21.360
far it applies to chinese vendors in the us and it can't be cleared with paperwork for builders

00:18:21.360 --> 00:18:26.640
that means public sector sales depend less on which model you use and more on the channel you

00:18:26.640 --> 00:18:33.120
sell through a startup offering a thin wrapper over a consumer chatbot api is exposed to the first

00:18:33.120 --> 00:18:38.620
gate. A startup that runs on in-country infrastructure or sells through an approved

00:18:38.620 --> 00:18:46.240
platform may find this a tailwind. Intric is the local example. One diligence question worth

00:18:46.240 --> 00:18:52.340
exploring. If the company's main model provider were blocked for a given buyer tomorrow, how fast

00:18:52.340 --> 00:18:58.000
could it switch? And does its contract with that buyer survive the switch? We'd also want to see

00:18:58.000 --> 00:19:02.200
whether its open-weight dependencies trace back to a vendor that has been excluded.

00:19:04.080 --> 00:19:10.140
Three things would settle this. First, whether OpenAI or Anthropics sign formal agreements with

00:19:10.140 --> 00:19:15.300
Swedish government bodies, and whether the blocks then lift. If they do, it was about contracts.

00:19:15.800 --> 00:19:20.380
If they don't, the sovereignty reading gains ground. The enforcement authority's pending

00:19:20.380 --> 00:19:26.080
decision is the next data point. Second, how DHS treats self-hosted and fine-tuned

00:19:26.080 --> 00:19:31.360
deep seek models that will show up in the exceptions agencies request and in any guidance

00:19:31.360 --> 00:19:38.800
clarifying deep seek implementations third whether any european body excludes a u.s vendor on origin

00:19:38.800 --> 00:19:45.680
grounds rather than contract grounds nobody has done that yet if one does the picture changes

00:19:45.680 --> 00:19:53.560
and we'd revise this view here's steven's take trust is not a security position it turns out

00:19:53.560 --> 00:20:17.220
FAAFO. The demand for state-approved or sovereign solutions will be determined by vendor trust, operator control, and the relative secrecy of not just the content, but the use of the harness, agents, tools, and the way in which this use is used to improve artifacts or models that are in the control of the payer users, government, companies, universities, or the control of the model or service provider.

00:20:17.220 --> 00:20:24.200
note this includes things above the frontier model and perhaps the frontier ends up commoditized

00:20:24.200 --> 00:20:29.880
completely in which case the third scaling law of learning above the model isolated from the

00:20:29.880 --> 00:20:36.340
overall frontier training and improving cycles becomes the defensible and secure layer this

00:20:36.340 --> 00:20:43.900
would bode very well for companies building above that's the deep dive and today's ai briefing

00:20:43.900 --> 00:20:46.520
Every source is linked in the show notes

00:20:46.520 --> 00:20:49.720
And the AI Briefing webpage has the written edition

00:20:49.720 --> 00:20:52.560
This deep dive in every past episode

00:20:52.560 --> 00:20:54.940
The Daily Briefing is back tomorrow