michaelcohen comments on Value Learning is only Asymptotically Safe

[michaelcohen]

Linking this, I meant “with probability strictly greater than 0, the agent is not safe”. Sorry for the confusion.