Risks and Failure Modes of Alignment Training — Roadmap Step & Resour…
Reward hacking and other ways alignment training can go wrong
- Level: beginner
- Estimated time: 1 days
- Roadmap: Fine-Tuning & Model Customization
Before this step
Study resources
- PyTorch Documentation (Article) — Official documentation for the deep learning framework underlying most modern fine-tuning workflows. Background reading/viewing for: Risks and Failure Modes of Alignment Training.
- Hugging Face Documentation (Article) — Official documentation on RLHF and preference-based training methods. Relevant to: Risks and Failure Modes of Alignment Training.
Part of
- RLHF and Preference-Based Alignment — section
- Fine-Tuning & Model Customization roadmap — the full learning path