Hacker News Reader

I don't understand this comment, are you saying they should allow it to appear in training data?

They're saying this is a very poor defence.

Well, it seems better than not doing that?

It is eager to do things it shouldn't do. But I suppose it's better than lying about it. Although they do that too.

From top to bottom, I still think nobody developing these things really cares about security.

I think it should be allowed yes. I don't think our strategy is hoping the agents never learn that misalignment is an option.

Well, no one's saying they shouldn't also do other things about the problem.