Thanks for this link. I didn't see it when it was posted to HN last month.
replyGood companion read to the OP thanks for sharing
replycan you elaborate?
replyI guess the connection is that if you assume neural networks are in fact filled with meaningful features instead of superposition blackbox stuff, then what the LLM is doing here isn't quite magic but rather methodical interpretability. And if you can isolate the neuron that is e.g. responsible for steering the paddle then you can probably tweak it to steer more/less aggressively.
replyIt is still very surprising though, considering it apparently displays an "intuitive" understanding of the weights, that make whatever Goodfire did look like primitive monkeys poking at it by comparison.
> It is still very surprising though, considering it apparently displays an "intuitive" understanding of the weights, that make whatever Goodfire did look like primitive monkeys poking at it by comparison.
replyFYI, Nick/Goodfire used the models extensively in their work, so it's very possible that there were some similarities in their work. I'm really curious to see his response to this.
I think this work does vindicate their hypothesis that alignment/superalignment might actually turn out to be "easy", in that you can just train LLMs do to the interpretability work. Maybe this is in fact what OpenAI did, or if it's purely "emergent" then it seems they have great priors that can be made more powerful with some RL
reply