Turns out, aligning LLMs to be "helpful" via human feedback actually teaches them to bullshit—and Chain-of-Thought reasoning just makes it worse!
delichon•5h ago