News for hackers. Stuff that matters.

A comment on Weight-sparse transformers have interpretable circuits [pdf]. Parent comment

Re:Weight-sparse transformers have interpretable circuits [pdf] by leogao on Saturday November 22, 2025 @01:55PM (#46017242)
There's been some work (e.g RASP - https://arxiv.org/abs/2106.06981) on taking logical computations and compiling them into transformer weights.
[ Parent | on HN ]
Re:Weight-sparse transformers have interpretable circuits [pdf] by astrange on Sunday November 23, 2025 @01:23AM (#46021260)
Sakana AI is also working on merging different transformer models together to combine skills.

https://sakana.ai/evolutionary-model-merge/

[ Parent | on HN ]