Decision Sufficiency of Learned Representation Interfaces for Safe Generative Control
We study decision sufficiency: whether the best controller using a representation can match the minimum average loss attainable with full state information, under the same strategies and evaluation criteria.
ProofPaper ↗
Key points
- A controller can guide a language model toward safe and useful responses by choosing when to assist, ask for clarification, or refuse.
- For interfaces with finitely many outputs, we compute the optimal loss exactly and separate the three sources of excess loss.
- Pairwise checks can miss information loss: states with the same controller input may share an optimal strategy in every pair but not across the whole group.
- We audit selected features from sparse autoencoders, which encode model activations using relatively few active features, in Qwen2.5, SmolLM2, and Phi.
Sources (1)
- [1]Decision Sufficiency of Learned Representation Interfaces for Safe Generative ControlarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 12:49 AM
We study decision sufficiency: whether the best controller using a representation can match the minimum average loss attainable with full state information, under the same strategies and evaluation criteria.
A controller can guide a language model toward safe and useful responses by choosing when to assist, ask for clarification, or refuse.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 19, 2026Introducing Clairic, a new System One class of extraction models
- Oct 19, 2026Understanding When Reflective Prompt and Skill Optimization Helps on Locally Served Reasoning Models: A Two-Model Study of GEPA and SkillOpt
- Oct 19, 2026Measurement Validity and Representation Dynamics of Confidence Signals in Large Language Models
- Oct 18, 2026My 5070 ti/3070 Chimera build.
- Oct 18, 2026ATEST: Adaptive Resource-Aware Test-Time Scaling for LLM Inference
- Oct 17, 2026A Symmetry-Matching Approach to Blind-Spot Reduction in Sparse Autoencoders