LLM Performance Predictors: Learning When to Escalate in Hybrid Human-AI Moderation Systems
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866917195722784768 |
|---|---|
| author | Bachar, Or Levi, Or Mishra, Sardhendu Levi, Adi Minhas, Manpreet Singh Miller, Justin Ben-Porat, Omer Sheetrit, Eilon Morra, Jonathan |
| author_facet | Bachar, Or Levi, Or Mishra, Sardhendu Levi, Adi Minhas, Manpreet Singh Miller, Justin Ben-Porat, Omer Sheetrit, Eilon Morra, Jonathan |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_07006 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | LLM Performance Predictors: Learning When to Escalate in Hybrid Human-AI Moderation Systems Bachar, Or Levi, Or Mishra, Sardhendu Levi, Adi Minhas, Manpreet Singh Miller, Justin Ben-Porat, Omer Sheetrit, Eilon Morra, Jonathan Artificial Intelligence I.2.6; I.2.7; I.5.1; H.1.2 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. |
| title | LLM Performance Predictors: Learning When to Escalate in Hybrid Human-AI Moderation Systems |
| topic | Artificial Intelligence I.2.6; I.2.7; I.5.1; H.1.2 |
| url | https://arxiv.org/abs/2601.07006 |