Efficient LLM training may reduce reasoning faithfulness but maintain monitorability

A new arXiv paper investigates the impact of efficient reasoning training on large language models (LLMs). Researchers explored three methods for applying length pressure to Chain-of-Thought (CoT) reasoning, finding that while faithfulness generally decreases due to reduced model consistency, monitorability remains more robust. The study suggests that even with shorter CoTs, models can still indicate how input changes affect their outputs. AI

IMPACT This research could inform the development of more efficient LLMs without significantly compromising their ability to explain their reasoning.

RANK_REASON The cluster contains a research paper published on arXiv discussing LLM training methods. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• alphaXiv
• arXiv
• CatalyzeX
• CORE Recommender
• CoT faithfulness
• DagsHub
• Gotit.pub
• Hugging Face
• Influence Flower
• large-language models
• monitorability
• ScienceCast

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →