kaelandt

It's just really hard for them to write non-verbose code. I don't know if this is incentives from the providers to generate more tokens, but even with guidance on compact code, simple, etc, they just can't really do it right now.

It isn't that surprising that it works well, this problem is fairly well known and some simple heat equations would lead to the result, about which there is a lot of training data online.

Would tend do disagree. Debug is arguably the most important skill to have and nurture right now (and that allows you to keep a strong mental model), and LLMs are not yet that good at it

Misleading title, it's more like GPT-5.2 derives the generalization of a formula that physicists conjectured. Not really related to physics

Now people are opening issues and LLMs are responding in completely nonsensical ways, nice one https://github.com/anthropics/claude-code/issues/22843

Nice to see an AI coding company allow such studies to come out, and it looks decently designed

the AI inference workloads shown in the paper are extremely far from what is implied when one says "... computer for AI inference". No discussion of issues around the memory hierarchy and how the presented architecture solves those. No mention of transformers, except for a vague reference to energy-based models

The API prices of $3/$15 are not right for a lot of models. see at openrouter, the gpt-oss-120b ones https://openrouter.ai/openai/gpt-oss-120b, it's more like $0.01/$0.3 (and that model actually needs h200/b200 to have good throughput).

One thing that is also truly unappreciated is most of us humans actually enjoy thinking, and people are trying to make llms strip us from a fundamental thing we enjoy doing. Look at all the people that enjoy solving problems for the sake of it

could be nice to add a "merged PR with a test" metric. looking at the PRs they are mostly without tests, so could be bogus for all we know