People new to tech submitting long AI generated messages don't realize concise language is expected by busy people and shows you can think.

Excited to share our new paper, "DataRater: Meta-Learned Dataset Curation"! We explore a fundamental question: How can we *automatically* learn which data is most valuable for training foundation models? Paper: arxiv.org/pdf/2505.17895 to appear NeurIPS Conference Thread 👇

The health benefits of drinking a beer with your friends are underrated.

😬

Do American citizens get equity in OpenAI if our taxes are backstopping them?

I’m actually not against nationalizing the OpenAI compute if we all get to vote on how it gets used.

It’s not crazy that people who were convinced to pay a $100k entrance fee to capitalism as teens only to get nothing from it do not like capitalism.

Who is doing the best financial reporting on AI infrastructure investments right now? I’m just seeing tweets here and there.

“Eventually, some scientists will figure out how to maintain high inference quality with much, much less compute (thus cost), and when that happens, the whole tower blows up”

Decentralizing the view layer alone is not enough, But when you can give people control of their data AND view layer People with shared interests (friends, people sharing a hard to treat illness, family) can created shared, modifiable apps with ai prompting that are truly

If you’re fundraising now you are close enough to the holidays where you should play chicken with investors by offering to schedule meetings on weekends.

In our lifetime the amount of capital at stake from the AI transition will lead to the destruction of all the guilds. Real estate, medical, law, etc. In some ways this is for the better, see tweet below. In other ways we need different social safety nets detached from

Playing with Thinking Machines fine-tuning today. Realizing most teams don’t have enough observability to know when it would make sense to take the OSS performance hit and fine tune vs using a closed source model.

Loans for compute underwritten based on the predicted value of the work that will get done are going to be a big thing

A fantastic The New York Times article from Berggruen Institute (where I was a fellow) — as more people confide in chatbots, especially on mental health issues, these conversations should be given legal protections to safeguard private communications from courts “Allowing the government to

Claude is the AI boyfriend model

Founders asking VCs for intros to other VCs: DONT do this unless they invested.

Each state should have compute infrastructure. 1. Protects against capture and abuse by the federal government 2. Citizens of the state can vote to allocate compute towards initiatives they care about

Don't worry about the cyber thing, state-sponsored hackers in China are all effective altruists who just pretend to hack banks etc. in order to help Anthropic with regulatory capture

AI Safety done right is American Dynamicism