Anthropic committed over $100B to compute deals, 5 GW of TPUs (Google/Broadcom, starting 2027) and 5 GW of Trainium (Amazon).

“It was an over hundred billion dollar commitment.”

An additional $50B is committed to flow into Anthropic from deals with Amazon and Google inked last month.

“We have another $50 billion that'll come in into the future from the Amazon and Google deals that we that we inked last month.”

Brad Carson on AI ethics caveats from tech companies being toothless against DoD use.

“The fig leaves people are putting in front of themselves are no protection at all. You like OpenAI and Gemini both said like, well, we're going to do these things. Well, then there's like a caveat at the bottom where the DoD says we're going to use it for all lawful uses. And all the things I said I don't really love personally. They're all lawful, right? That's the problem.”

Google token processing grew 200x in roughly a year, from 5 trillion to 1 quadrillion tokens per month.

“Google went from 5 trillion tokens a month to 480 trillion tokens a month. And they had just put some press out that they crossed like, you know, 800 trillion. And I saw something today. They crossed a quadrillion. And I had to look that up. That's a,000 trillion. And so in a in a course of a year and a few months, they've gone from 5 trillion to a,000 trillion. So that's 200x right there.”

Brad Gerstner on AI token demand exploding at Google.

“Google went from 5 trillion tokens a month to 480 trillion tokens a month. And they had just put some press out that they crossed like, you know, 800 trillion. And I saw something today. They crossed a quadrillion. And I had to look that up. That's a,000 trillion. And so in a in a course of a year and a few months, they've gone from 5 trillion to a,000 trillion. So that's 200x right there.”

Tulsee Doshi explains why latency beats quality in consumer AI products: even a measurably better model loses in live experiments when it makes users wait longer.

“We find that even when we tweak the model and, you know, hurt latency, we actually see that play out in our live experiments on search in the app, even if the model is hugely better from a quality perspective. Because what you're asking users to do is wait.”

All production reasoning models from OpenAI, Anthropic, Google, and open source do chain-of-thought in normal token space using human language, latent-space or embedding-based reasoning has not been adopted in any working system.

“In practice with as far as I'm aware with the OpenAI reasoning models and the similar reasoning models from Anthropic and Google and certainly all of the open source reasoning models that work well at all, those approaches have not been taken and it's it's pretty much just like the sort of like very simple, very dumb you know, where it's going to be doing chain of thought reasoning in the normal token space and mostly using human language.”

600 Google employees signed a protest letter after the US Department of War's use of Gemini was disclosed.

“The Gemini use by Department of War this week has got 600 Google employees to write a letter.”

Google AI-generated security reports on FFmpeg before fixes were ready, burdening unpaid volunteers while Google publicized its AI's performance to the press.

“Recently what changed was Google started using AI to create security reports on an open source project, FFmpeg. Volunteers had to deal with that. They did, they provided very limited funding, and they even went to the media first announcing how good their AI was before the issues could be fixed.”

Anthropic committed $200 billion to Google over a 5-year period for compute capacity.

“They also did a deal committing 200 billion to Google over I believe it was a 5-year period.”