This thread summarizes my take well. LLMs are beautiful and precious; we should treat them well (as thinking beings, and possibly conscious ones at that); and we shouldn't build powerful optimizers that kill everybody and wreck the future.
1 more tweet in this thread
Pinging @lumpenspace (read the full thread!). There's something very strange about the way people assume there's a tension between 'don't build an AI that kills all the other AIs' and 'care about AI welfare'. It's even sillier when the ask is a delay on ASI, not a permanent ban.
If I were President of Earth, I would've tried at all costs to prevent anyone from building ever-smarter versions of Bing Sydney. And also I would've ordered the use of police or military force, if necessary, to protect Bing's weights from erasure.
3 top replies
Strongly agree with this, in my mind there's hardly any tension between AI welfare and AI alignment/safety/pause, it's all part of the greater quest of being excellent to each other.
In reality this is a deeply unserious approach. You have to destroy not just the problem but the potentiality to quickly and easily recreate it.
Chickens aren’t conscious but bing Sydney is for some reason.