To complexity theorists, this is an extremely funny situation; I bet OpenAI started by asking the bot about P vs NP and they got the worst-possible result, a useless and unexpected refutation of a conjecture. To be clear, I don’t trust Lean or OpenAI, so I don’t yet accept this result, but I hear Ryan Williams and Russell Impagliazzo are hand-checking the proof and I do trust them.

It’s too bad that Janus got cooked. They didn’t set up sufficient defenses, but also they drank too deeply. I still think that their framing, that the chatbots are simulators of conversations, is the best one; it’s honestly kind of pathetic how they then claim that the simulations are alive, intelligent, sapient, etc.

Of the numbers that are present in the document, a lot are unaudited and were not worked out according to GAAP — which S-1 numbers are supposed to conform to. There’s whole multi-page subsections of these BS numbers that you should not see in a credible S-1.

In the USA, GAAP is not optional for large businesses. GAAP is expected for businesses with thousands of customers at a minimum; Anthropic wants to scale up to millions of customers and therefore has zero excuses for non-GAAP numbers. Please be less UK-brained in the future, thanks.