3 ms·
Show HN: Predict LLM Predictions
- ggm 2y agoAdversarial 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.
- flowingfocus 2y agoAdversarial 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