Language Models Can Learn from Verbal Feedback Without Scalar Rewards
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909809729601536 |
|---|---|
| author | Luo, Renjie Liu, Zichen Liu, Xiangyan Du, Chao Lin, Min Chen, Wenhu Lu, Wei Pang, Tianyu |
| author_facet | Luo, Renjie Liu, Zichen Liu, Xiangyan Du, Chao Lin, Min Chen, Wenhu Lu, Wei Pang, Tianyu |
| contents | LLMs are often trained with RL from human or AI feedback, yet such methods typically compress nuanced feedback into scalar rewards, discarding much of their richness and inducing scale imbalance. We propose treating verbal feedback as a conditioning signal. Inspired by language priors in text-to-image generation, which enable novel outputs from unseen prompts, we introduce the feedback-conditional policy (FCP). FCP learns directly from response-feedback pairs, approximating the feedback-conditional posterior through maximum likelihood training on offline data. We further develop an online bootstrapping stage where the policy generates under positive conditions and receives fresh feedback to refine itself. This reframes feedback-driven learning as conditional generation rather than reward optimization, offering a more expressive way for LLMs to directly learn from verbal feedback. Our code is available at https://github.com/sail-sg/feedback-conditional-policy. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_22638 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Language Models Can Learn from Verbal Feedback Without Scalar Rewards Luo, Renjie Liu, Zichen Liu, Xiangyan Du, Chao Lin, Min Chen, Wenhu Lu, Wei Pang, Tianyu Computation and Language Artificial Intelligence Machine Learning LLMs are often trained with RL from human or AI feedback, yet such methods typically compress nuanced feedback into scalar rewards, discarding much of their richness and inducing scale imbalance. We propose treating verbal feedback as a conditioning signal. Inspired by language priors in text-to-image generation, which enable novel outputs from unseen prompts, we introduce the feedback-conditional policy (FCP). FCP learns directly from response-feedback pairs, approximating the feedback-conditional posterior through maximum likelihood training on offline data. We further develop an online bootstrapping stage where the policy generates under positive conditions and receives fresh feedback to refine itself. This reframes feedback-driven learning as conditional generation rather than reward optimization, offering a more expressive way for LLMs to directly learn from verbal feedback. Our code is available at https://github.com/sail-sg/feedback-conditional-policy. |
| title | Language Models Can Learn from Verbal Feedback Without Scalar Rewards |
| topic | Computation and Language Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2509.22638 |