3 ms·
> The fundamental challenge of AI alignment is generalization. ... > We do not have a satisfactory theory of generalization, and it seems unlikely that we can
by andai 1mo ago
> The fundamental challenge of AI alignment is generalization.
...
> We do not have a satisfactory theory of generalization, and it seems unlikely that we can develop one soon, at least without the help of more powerful AI.
-- From another OpenAI article in a sister thread:
An Alien Mind
https://news.ycombinator.com/item?id=49588080 https://news.ycombinator.com/item?id=49588080
- ahartmetz 1mo agoThat's a bit bullshit, isn't it? They basically redefined "needs more R&D" as "needs stronger AI". Maybe so - maybe AI won't help much with that problem.