Press Space to continue
Chain of Thought Monitoring Matters for AI Safety
Press Space to continue
A concerningly common take seems to be that keeping Chain of Thought monitorable doesn't matter because interpretability will save us, or it's already useless This is total bullshit. CoT is our best current tool for safety & interpretability, losing it would be a major tragedy
open on xViews and clicks count once per visitor per day. Clicks are opens of this post on X.