dorsal/arxiv
View SchemaLLM Performance Predictors: Learning When to Escalate in Hybrid Human-AI Moderation Systems
| Authors | Or Bachar, Or Levi, Sardhendu Mishra, Adi Levi, Manpreet Singh Minhas, Justin Miller, Omer Ben-Porat, Eilon Sheetrit, Jonathan Morra |
|---|---|
| Categories | |
| ArXiv ID | 2601.07006vv1 |
| URL | https://arxiv.org/abs/2601.07006 |
| License | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ |
Abstract
As LLMs are increasingly integrated into human-in-the-loop content moderation systems, a central challenge is deciding when their outputs can be trusted versus when escalation for human review is preferable. We propose a novel framework for supervised LLM uncertainty quantification, learning a dedicated meta-model based on LLM Performance Predictors (LPPs) derived from LLM outputs: log-probabilities, entropy, and novel uncertainty attribution indicators. We demonstrate that our method enables cost-aware selective classification in real-world human-AI workflows: escalating high-risk cases while automating the rest. Experiments across state-of-the-art LLMs, including both off-the-shelf (Gemini, GPT) and open-source (Llama, Qwen), on multimodal and multilingual moderation tasks, show significant improvements over existing uncertainty estimators in accuracy-cost trade-offs. Beyond uncertainty estimation, the LPPs enhance explainability by providing new insights into failure conditions (e.g., ambiguous content vs. under-specified policy). This work establishes a principled framework for uncertainty-aware, scalable, and responsible human-AI moderation workflows.
{
"annotation_id": "7c1a9b82-22ed-4290-9432-be3d196c27df",
"date_created": "2026-02-17T05:53:08.669000Z",
"date_modified": "2026-02-17T05:53:08.669000Z",
"file_hash": "f9405200508703f2f421d77b5cfb06de4472897c424932b114274ddb22eb846e",
"private": false,
"record": {
"abstract": "As LLMs are increasingly integrated into human-in-the-loop content moderation systems, a central challenge is deciding when their outputs can be trusted versus when escalation for human review is preferable. We propose a novel framework for supervised LLM uncertainty quantification, learning a dedicated meta-model based on LLM Performance Predictors (LPPs) derived from LLM outputs: log-probabilities, entropy, and novel uncertainty attribution indicators. We demonstrate that our method enables cost-aware selective classification in real-world human-AI workflows: escalating high-risk cases while automating the rest. Experiments across state-of-the-art LLMs, including both off-the-shelf (Gemini, GPT) and open-source (Llama, Qwen), on multimodal and multilingual moderation tasks, show significant improvements over existing uncertainty estimators in accuracy-cost trade-offs. Beyond uncertainty estimation, the LPPs enhance explainability by providing new insights into failure conditions (e.g., ambiguous content vs. under-specified policy). This work establishes a principled framework for uncertainty-aware, scalable, and responsible human-AI moderation workflows.",
"arxiv_id": "2601.07006",
"authors": [
"Or Bachar",
"Or Levi",
"Sardhendu Mishra",
"Adi Levi",
"Manpreet Singh Minhas",
"Justin Miller",
"Omer Ben-Porat",
"Eilon Sheetrit",
"Jonathan Morra"
],
"categories": [
"cs.AI"
],
"license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
"title": "LLM Performance Predictors: Learning When to Escalate in Hybrid Human-AI Moderation Systems",
"url": "https://arxiv.org/abs/2601.07006",
"version": "v1"
},
"schema_id": "dorsal/arxiv",
"source": {
"execution_id": "2f5c19b5-4e62-4bb9-a8a9-ff5ac2d49f34",
"id": "arXiv Dataset",
"type": "Model",
"variant": "snapshot-2026-01-17",
"version": "0.1.0"
},
"user_id": 1000002
}