Beyond Answer Confidence: A Controlled Audit of Self-Knowledge in a Black-Box De
arXiv:2610.03894v1 Announce Type: new Abstract: A deployed LLM agent emits tool calls, queries, and code that can be silently wrong -- by the time the error surfaces, the action
arXiv:2605.27752v1 Announce Type: new Abstract: LLM confidence calibration is often evaluated by comparing two signals: token-probability scores and verbalized confidence. These
arXiv:2608.13591v1 Announce Type: new Abstract: High-confidence errors in large language models are often treated as evidence of fragile internal inference. We study a different
arXiv:2609.32035v1 Announce Type: new Abstract: Self-consistency assumes that independent samples disagree when a model is unsure, so agreement is evidence of correctness. Holding
arXiv:2610.20836v1 Announce Type: new Abstract: Evaluator-guided reasoning presents a local control problem at every step: trust the current answer, return a fallback, or spend
arXiv:2607.18278v1 Announce Type: cross Abstract: Calibration is usually evaluated in aggregate, but the most dangerous failures are often local: predictions that remain highly
NewsHive monitors these sources continuously. All signal titles above link to the original reporting.
Intelligence by NewsHive. Need help navigating what this means for your business? Contact GeekyBee →