Just wanted to say: this is might impressive research.

Really interesting breakdown, proper nerdsniped into this, thanks for the refreshing AI news outside of language models :)