4 min read

Happy Tuesday!

OpenAI and Anthropic both say their models have knocked down a rule that computer scientists have leaned on since 1999. Anthropic also found time to file for its stock market debut and Microsoft has offered around 60,000 staff the chance to leave on New Year's Eve.

OpenAI released three new results on Monday that it calls major discoveries. The one getting the most attention is a paper titled "The Exponential Time Hypothesis is False". OpenAI's model found an algorithm that solves a classic family of logic puzzles in subexponential time, which is faster than computer scientists believed possible.

The hypothesis was first published in 1999 by Russell Impagliazzo and Ramamohan Paturi at UC San Diego. It says some problems get exponentially harder as they grow, and many papers from the past decade take it as given. Hours after OpenAI's release, Anthropic put out its own refutation. That came after Anthropic mathematician Levent Alpöge suggested OpenAI had built on existing work.

This time OpenAI handed the proof to a working group led by Impagliazzo, Paturi and MIT's Ryan Williams, who will check it and explain it. Williams says some of it is routine but "two or three alien-looking parts" need the whole field. One of his first questions to OpenAI was whether the result broke any encryption. OpenAI had checked and was reasonably confident it didn't.

Earlier this month Ryan Williams called OpenAI's proof dump a "DDOS attack on mathematicians". Now he is checking its newest one. A rule that stood for 27 years is now a race between two labs, and I find it quite funny somebody still has to do the marking!

Lucas

Anthropic has filed to go public and plans to list on the Nasdaq under the ticker ANT. It filed confidentially with the SEC in June, so Monday's S-1 is the first proper look at its books. It hasn't yet said how much it wants to raise or what the shares will cost.

Revenue for the first half of 2026 was $16.3 billion, up from $1.2 billion a year earlier. The operating loss shrank to $693 million. The net loss was $80.7 billion, mostly noncash accounting charges tied to convertible debt. About 250 customers now spend $10 million or more a year.

The seven co-founders will keep control through a "Founder LLC" holding 50.1% of shareholder votes. The risk factors include its models developing "forms of sentience", and Anthropic warns that no single insurer could cover a "catastrophic" AI incident. Dario Amodei was paid nearly $18 million in 2025.

The founders keep 50.1% of the votes, and the prospectus warns about sentient models and accidents no insurer will touch. Shareholders get the risk and very little say, to me buying ANT gets you a seat on the bus and a good view of the driver.

Callum

Microsoft is offering voluntary buyouts to around 60,000 staff in the US, Canada, India and the UK. That is about 27% of its 223,000 employees. Teams including Xbox, Windows and Copilot are taking part, while data centre, sales and marketing staff are among those left out.

Staff have until early December to decide, and anyone who accepts leaves on 31 December. Amy Coleman, its chief people officer, told staff the offer gives them more ways "to shape their next chapter".

Much of the money is going on data centres. Microsoft expected capital spending for the September quarter to top $50 billion, against $34.9 billion a year before. In April it ran a voluntary retirement scheme, which more than 30% of eligible staff took, and in July it laid off 4,800 people. It reports results on Wednesday.

Copilot's own team is on the list of people Microsoft would happily see leave. The company is building an AI future and offering more than a quarter of its staff the chance to skip it. Results come out on Wednesday, so the timing is pure coincidence, I'm sure.

Lucas