I wrote a post about why AI for Biology could eclipse LLMs—and the unique challenges we need to solve to get there. As often I'm the bio person in a circle of LLM researchers in San Francisco, I've had fascinating discussions about AI's future in biology. It's clear we can't simply copy-paste LLM strategies, but the potential impact could be enormous: from understanding life's origins to curing disease. In this post, I break down three critical challenges: - Why we can't just use "RLHF" for biology - Why data quality is more challenging than for LLMs - Why scaling laws alone won't solve our problems - And importantly: what we can do about it. It's not another doom post or hype piece—instead, it's a pragmatic optimist's perspective on building AI for biology. https://lnkd.in/gNYihCKM
