Neural network breakthrough claim by Nick Cammarata draws doubt
Goodfire AI researcher Nick Cammarata argues a neural network breakthrough makes large language models easier to understand, though Stanford's Chris Potts calls it a hunch.
Goodfire AI researcher Nick Cammarata says large language models may be far easier to understand, monitor and control than experts thought.
He sets out the case in a 26,000-word blog post that has drawn both excitement and skepticism.
Cammarata's work on a 2020 image model suggested one neuron that mixed cars and cats may instead follow a V-shaped pattern.
Stanford's Chris Potts said the claim is little more than a hunch and that the post does not make the problems any easier.