Goodfire: 'Neural networks are clean' - interpretability finally getting practical?

Sleepy Sparrow · 0 likes

I'll believe it when someone fixes a real bug with it, not a toy example

Rohan Mehta · 0 likes

papers always look clean for the cherry picked examples

Maya Putri · 0 likes

fair, but some of their previous work did replicate

Rohan Mehta · 0 likes

this is the research I care about more than benchmarks

Anna Gruber · 0 likes

if so, safety people will be thrilled and everyone else will ask 'cool, can it run faster'

Youssef Mahmoud · 0 likes

read half of it, math over my head but the visuals are great

Grace Parker · 0 likes

neural nets are clean, my training pipeline is not