[edot]

And OpenAI didn't try to silently degrade performance of their top model if its (extremely sensitive) safety sensors went off ...

[estetlinus]

Anthropic is the silver lining keeping p(doom) below 1.0