CRITERIA FOR TRAINING AI MODELSÂ
Supervised Fine Tuning (SFT) and Reinforcement Learning from Human Feedback (RLHF)
Embedded Files
Ranking of model responses and rewriting
Remove unnecessary fluffiness, chatbot type tone and pleasantries
Stump models to make them capable of solving complex problems
Teach model to make refusals when required (Hard/Soft punt)
Removing any preachiness from model responses
Detection of sensitive content: PII, Privacy violations, Harmfulness. Promoting violence, Adult Content
Make it better than existing AI models
Page updated
Google Sites
Report abuse