Trusted Uncertainty in Large Language Models: A Unified Framework for Confidence Calibration and Risk-Controlled Refusal
arXiv:2509.01455v4 Announce Type: replace Abstract: Deployed language models must decide not only what to answer but also when not to answer. We present UniCR, a unified framework that turns heterogeneous uncertainty evidence including sequence likelihoods, self-consistency dispersion, retrieval compatibility, and tool or verifier feedback into a calibrated probability of correctness and then enforces a user-specified error budget via principled refusal. UniCR learns a lightweight calibration head with temperature scaling and proper scoring, supports API-only models through black-box features,
The increasing deployment of Large Language Models (LLMs) in critical applications highlights an urgent need for reliable confidence calibration and risk management, which UniCR addresses directly.
A strategic reader should care because this framework tackles a fundamental limitation of current LLMs, enabling safer, more responsible, and auditable AI deployments across various industries.
This framework offers a standardized method for LLMs to express calibrated uncertainty and refuse answers when confidence is low, shifting from opaque outputs to more transparent and controllable AI behavior.
- · AI safety researchers
- · Enterprises deploying LLMs
- · Developers of LLM-based solutions
- · Regulators
- · Providers of uncalibrated LLMs
- · Use cases requiring absolute certainty from AI
Increased trust and adoption of LLMs in high-stakes environments due to improved reliability and risk control.
Development of new AI applications that were previously too risky due to unaddressed uncertainty, expanding the market for intelligent agents.
Potential for new regulatory frameworks and industry standards to mandate similar calibration and refusal mechanisms for production AI systems.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL