x
Linear steerability in continuous chain-of-thought reasoning — LessWrong