Skip to content

Show HN: Predict LLM Predictions

think-in-sync.com
3 pointsflowingfocus2 comments
On HN

Comments

Adversarial play might be interesting. I'm assuming the goal here is to use the feedback as training input, fitness to expectations and like.

So a random chance of a bad actor adversary? Or high scores for lying? "The traitor" style

Team models? I suppose I'm arguing for gamification with a leader board and hubristic bragging rights.

Adversarial play sounds fun, I might try that out!

The LLM we use is mistral-large-latest, we didn't do any training on the data yet

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.