Rikiya Takehi ·
Researchers say LLMs can be good at natively reading and writing neural network weights; GPT-6 Astra edited Qwen3.5-0.8B's weights to make it better at math
TL;DR … If you were asked to handcraft a neural network that plays Go, what would you do? If you were handed raw neural network weights …
Lead Source
How this story grew
Coverage · 0
Discussion · 0
Nov 2Nov 4
More
Simon Willison's Weblog: Simon Willison's Weblog
honnibal.dev: honnibal.dev
seangoedecke.com: seangoedecke.com
VentureBeat: VentureBeat
QUE.com: QUE.com
Digit: Digit
AIM: AIM
The Register: The Register
The Rundown AI: The Rundown AI
Discussion
@alexkrentsel @kimmonismus @rikiyatakehi @csinva @yuchenj_uw @neelnanda5 @adarshsolanki @theahmadosman @nrehiew_ @kimmonismus @tszzl @karenzzhou @corsaren @yuchenj_uw @zainhas @linusmixson @dummerspast39 @signulll @fakepsyho @jcyranka @nickadobos @thorstenball @schmidhuberai @xennygrimmato_ @heyshrutimishra @realastropulse @lateinteraction @haider1 @tokengremlin @yoavgo @smokeawayyy @xeophon @hsu_steve @arjunjain @elder_plinius @emollick @tobi @lymanstoneky @adam_karvonen @martin_casado @pavel23 @0xshai @ramez @tenobrus @tunguz @dejavucoder @andrewcurran_ @a1zhang @lateinteraction @jayden_teoh_