this stuff is fascinating; it will lead to better 'lens' technology than the blind reconstructions. with luck, it'll prove far superior than CoT monitoring ever was
roon says circuit-level interpretability tools may beat CoT monitoring
AIroon says interpretability work on neuron-level circuits will lead to better "lens" technology than blind reconstructions, and may prove more effective than chain-of-thought monitoring. The post links to @_a9lim's Neuron Lens, which now shows circuits and traces how activations flow across features.
roonVerified on X
@tszzl
https://neuron-lens.com now shows circuits! we traced how activations flow across features, and it turns out that they're often very legible. as always, this was done with @treetowntree!
Source details
roon · x.comPublished · added here
Original title: this stuff is fascinating; it will lead to better 'lens' technology than the blind reconstructions. with luck, it'll prove far superior ...