3 ms·
Interesting. This is the person who "holds comparable influence to a USAF Colonel." according to "a prominent researcher in AI safety" who "discovered prompt in
by renewiltord 3y ago
Interesting. This is the person who "holds comparable influence to a USAF Colonel." according to "a prominent researcher in AI safety" who "discovered prompt injection". https://news.ycombinator.com/item?id=38330566 https://news.ycombinator.com/item?id=38330566
Well, I suppose this tells us something about the AI safety community and whether it makes sense to integrate safety into one's workflow. It seems that the best AI safetyists will scuttle your company from the inside at some moment not yet known. This does sort of imply that it is risky for an AI company to have a safetyist on board.
That does seem to be accurate. For instance, Google had the most formidable safety team, and they've got the worst AI. Meta ignored theirs and they've given us very good open models.
- comp_throw7 3y ago"AI safety" should be disentangled into "AI notkilleveroneism" and "AI ethics", which are substantially non-overlapping categories. I've looked at who works at Preamble, and there aren't any names there that I recognize from the side of things that's concerned with x-risk. Take their takes with a grain of salt.
- Seanambers 3y agoAI safety is just woke 2.0.
- tomrod 3y agoThis is incorrect.