Not many and even fewer if you consider folks that have a good grasp of themselves, their psychology, their emotions — and how they can mislead them, and their heart.
IME most folks at Anthropic, OpenAI or whatever that are freaking out about things never defined the problem well and typically were engaging with highly theoretical models as opposed to the real capabilities of a well-defined, accomplished (or clearly accomplishable) system. It was too triggering for me to consider roles there in the past given that these were typically the folks I knew working there.
Sam may have added a lot of groundedness, but idk ofc bc I wasn’t there.
It's a way of saying that what has been historically been considered "studying AI safety" in fact bears little relation to real life AIs and what may or may not make them more or less "safe".
Yes, with the addition that I do feel that we deserve something better than I perceive we’ve gotten so far and that safety is super important; but also I don’t work at OpenAI and am not Ilya so idk
IME most folks at Anthropic, OpenAI or whatever that are freaking out about things never defined the problem well and typically were engaging with highly theoretical models as opposed to the real capabilities of a well-defined, accomplished (or clearly accomplishable) system. It was too triggering for me to consider roles there in the past given that these were typically the folks I knew working there.
Sam may have added a lot of groundedness, but idk ofc bc I wasn’t there.