Researchers have developed RLEC, a novel reinforcement learning method designed to train large language models (LLMs) to adhere to explicit confidence targets. Unlike previous approaches that struggled to differentiate between confidence levels, RLEC successfully enables a single LLM to control its response behavior based on user-specified confidence requirements. This is achieved by preserving the learning signal, introducing a stable reward family, and employing reward scheduling for training stability. Experiments demonstrate that RLEC maintains general LLM capabilities while offering competitive calibration performance and fine-grained control without post-training adjustments. AI
IMPACT Enables LLMs to reliably abstain from answering when confidence is low, improving user trust and model reliability.
RANK_REASON The item is a research paper detailing a new method for training LLMs. [lever_c_demoted from research: ic=1 ai=1.0]
- alphaXiv
- arXiv
- CatalyzeX
- Connected Papers
- DagsHub
- Gotit.pub
- Hugging Face
- Litmaps
- LLMs
- RLEC
- ScienceCast
- scite Smart Citations
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
