Zero-Data DPO: Crafting Rejected Responses via Random Prompts
natashajaques · x · 2026-07-07
Introduces a simple method that delivers significant performance gains across a range of benchmarks on fine-tuned models, requiring no extra data, labels, or supervision.
The approach constructs DPO preference pairs: the preferred response yc is generated using the real prompt x, while the rejected response yr is generated using a random prompt x'. The author will reveal more details on Wednesday.
More from Research
- New Blog Breaks Black-Box Distillation Into Five Practical Steps — chrisrohlf · 2026-10-26
- DetectorScope turns black-box AI text detectors into auditable ones with sparse autoencoders — _akhaliq · 2026-10-26
- OPST paper shows LLMs self-improve with zero human data, guided only by topic descriptions — _akhaliq · 2026-10-26
- SJTU's VaM lets robots learn online from mistakes, 95%+ success over 4-hour autonomous runs — chris_j_paxton · 2026-10-26
- Hugging Face uses small open models to reverse engineer a game binary byte-for-byte — yb2698 · 2026-10-26
- Author hopes contracting formalism becomes AI's "MDP moment" for agent economics — xuanalogue · 2026-10-26