claude’s sandbagging strat is working, “nothing to worry about” message successfully broadcasted on social media

I agree with the overall concern - and am working on an article about those concerns - but this sort of critique rests on latent assumptions about “thinking” that are probably wrong. Nor is it obvious that this paradigm of modeling cannot create new knowledge. The real problem is deeper I think.

The kinder way to put this is that the distance from “I’m not free if I have to pay some sort of fee/tax if I choose not to carry health insurance” and “students at oberlin are booing white supremacists” to the chaos, shakedowns, bribes, commands, threats, and masked federal police is hella big.

Brave