Open-Assistant/model/reward/instructor/TODO.md

Some other reward features we can use

0. Finish classification feature

1. Summaries from human feedback

• use confidence score into the RM learning, ensure the output rank score correlates with confidence

• each labeling has a labeling note, basically comments by labeler, not sure what else we can use

• Use the score for "overall", "accuracy", "coverage", "coherence" from axis/evals to train an addition model (rank additional aspect of the policy model)

  • this should be placed under experimental_dataset.py

2. Add support for anthropic dataset

• anthropic dataset is more like a conversation tree which is much complex than simply question-answer schema

  • this is basically a MCTS from alphazero.