Strong yet Controllable LLM Abstention without Learning to Abstain

Key points

• Abstention can be implemented outside the model (selective answering) using a confidence threshold that controls the answer rate (coverage), or learned inside it (abstention training), which can perform better but fixes coverage once trained.1
• The decomposition shows that abstention training performs better mainly by answering better: with strong confidence estimators, the two have nearly equal distinguishing ability.1
• Across three LLMs and three factual QA benchmarks, TAPS with a probing-based confidence estimator attains 1.5--9.4 percentage points lower mean risk than three representative abstention training methods at matched coverage.1
• Together, these results establish TAPS as a strong yet controllable baseline and raise the bar for future abstention training: sacrificing control over coverage should yield gains beyond what can be achieved by optimizing answering and abstaining separately.1

Sources (1)

• [1]Strong yet Controllable LLM Abstention without Learning to Abstain
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 12, 07:41 AM

  Large language models (LLMs) can reduce hallucination by abstaining from questions they would likely get wrong.

  Abstention can be implemented outside the model (selective answering) using a confidence threshold that controls the answer rate (coverage), or learned inside it (abstention training), which can perform better but fixes coverage once trained.

Extractive summary: sentences quoted from the sources.