I think Anthropic didn’t have this mindset with the recent sparse coding paper. I heard rumors about it before publishing, but after publishing there was a bunch of interest into sparse coding, with multiple teams working on circuit discovery and elsewhere in interpretability. Plausible to me that if a draft were widely shared this could have happened sooner.
I think Anthropic didn’t have this mindset with the recent sparse coding paper. I heard rumors about it before publishing, but after publishing there was a bunch of interest into sparse coding, with multiple teams working on circuit discovery and elsewhere in interpretability. Plausible to me that if a draft were widely shared this could have happened sooner.