Tyler Cowen recently compared the emergence of LLMs to the invention of the printing press.
I think he’s right. And it seems like we’re speedrunning it.
We should expect very good stuff & very bad stuff—this year; this decade.
Thankfully, self-governance from some AI labs & action from some governments seem poised to mitigate some of the bigger risks.
Gotta catch 'em all, though. We won't manage that first time.
From a distance @OpenAI seem admirably open about their strategy & the safety work they did before releasing GPT-4.
(I have no insider knowledge.)
openai.com/research/gpt-4
cdn.openai.com/papers/gpt-4-system-card.pdf
Sam Altman strikes me as thoughtful & sincere. I think he went into this for good reasons.
FWIW I met him twice, during @80000hours YC interview & an office hour. My gut feel: trustworthy.
Hard to think of someone I’d rather see as CEO of a top AI lab.
twitter.com/peterhartree/status/1627802727144386560
I do wish @sama would say more about his view that short timelines & slow takeoff are the safest world. But: there may be a good reason he hasn’t.
Transparency is not always the best strategy, whether you’re doing dual-use research, or pursuing political outcomes.