- Drafted a blog post
- Used an LLM to meticulously improve the argument over 4 hours.
- Wow, feeling great, it’s so convincing!
- Fun idea let’s ask it to argue the opposite.
- LLM demolishes the entire argument and convinces me that the opposite is in fact true.
- lol
The
Karpathy makes a sharp epistemic point: LLMs are excellent at constructing persuasive cases in multiple directions, so a single convincing pass should not be mistaken for truth. His takeaway is to use models adversarially—ask them to strengthen and then attack an argument—while staying alert to sycophancy. This seems squarely in your lane as a practical reminder about using LLMs for dialectic pressure-testing rather than outsourced belief formation.
