Steven Byrnes comments on Does reducing the amount of RL for a given capability level make AI safer?

[Steven Byrnes]

I have to admit that I’m struggling to find these arguments at the moment

I sometimes say things kinda like that, e.g. here.