analiza aiqbench
This research introduces TAPS, a modular framework that combines accuracy-focused training with post-hoc confidence estimation to achieve superior risk reduction while maintaining controllable answer coverage. By decoupling answering ability from distinguishing ability, the study demonstrates that strong confidence estimators can match the performance of complex abstention training methods without sacrificing flexibility. This approach offers a practical baseline for reducing LLM hallucinations in factual QA tasks through standard post-training adjustments rather than specialized learning objectives.
Strong yet Controllable LLM Abstention without Learning to Abstain
arXiv:2610.16318v1 Announce Type: new Abstract: 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. To compare the two, we match their coverage and decompose risk (error rate) into answering ability and distinguishing ability, i.e., how well the model answers and how accurately it abstains, respectively. The decomposition shows that abstention training performs better mainly by answering better: with strong confidence estimators, the two have nearly equal distinguishing ability. This motivates Train for Accuracy, then Post-hoc Select (TAPS), a modular approach that keeps coverage controllable, needs only standard post-training and an off-the-shelf confidence estimator, and benefits from advances in either component. 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. Within TAPS, accuracy training often weakens the model's own confidence signals, yet refitting the estimator largely preserves distinguishing ability. 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.
Abstract: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. To compare the two, we match their coverage and decompose risk (error rate) into answering ability and distinguishing ability, i.e., how well the model answers and how accurately it abstains, respectively. The decomposition shows that abstention training performs better mainly by answering better: with strong confidence estimators, the two have nearly equal distinguishing ability. This motivates Train for Accuracy, then Post-hoc Select (TAPS), a modular approach that keeps coverage controllable, needs only standard post-training and an off-the-shelf confidence estimator, and benefits from advances in either component. 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. Within TAPS, accuracy training often weakens the model's own confidence signals, yet refitting the estimator largely preserves distinguishing ability. 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.
Computation and Language (cs.CL)
arXiv:2610.16318 [cs.CL]
(or arXiv:2610.16318v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2610.16318 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 317 cuvinte
Subiecte conexe
Cross-Agent Learning Signals Enable Coordinated Role-Decomposed LLM Training
arXiv · 8 oct. 2026
[AINews] AI Engineer NYC: Gemini 4 Argon, Vals AI Benchmarks
Latent Space · 14 oct. 2026
Causal Evidence that Language Models use Confidence to Drive Behavior
arXiv · 21 sept. 2026
TwistedMerge: Certified Higher-Order Diagnostics and Abstention for Model Merging
arXiv · 24 iul. 2026