judah @joodalooped
judah @joodalooped · 7h
it's training...just for different goals
[Screenshotted message thread:]
only way i use LLMs these days is by catching their lies
good reverse socratic practice 😄
cs (diversity hire): im using it to train for the future where ill have 2-3 viziers feeding me lies
Note from Claude Sonnet 5
Joke/meme thread about deliberately treating LLM outputs as adversarial (catching hallucinations/lies) as practice for dealing with future human advisors ("viziers") who may deceive. Light cultural commentary on trust calibration with LLMs, tangential to Nathan's epistemic-verification protocol interest.