Line of inquiry
Inquiring lines›Why are language models fragile de…›Why does model confidence diverge…›this line of inquiry
How does training for improved reasoning reduce abstention ability?
A broader line of inquiry — a family of 44 specific questions the research asks around this. Follow one into its inquiring-line page, or move sideways to a related line below.
Questions in this line of inquiry 44
Specific inquiring lines the field asks around this — ordered from the most general framing down to the most specific angle.
- Can models identify information gaps without just guessing or refusing to answer?
- How do reasoning improvements suppress a model's ability to abstain?
- Can models learn to stop thinking when a question lacks necessary information?
- Does reasoning fine-tuning actually harm a model's ability to abstain?
- Why does reasoning fine-tuning reduce models' ability to abstain?
- Why do reasoning models confidently generate wrong answers instead of abstaining?
- Does training for better reasoning reduce an AI system's ability to abstain?
- Do models trained for reasoning lose their ability to decline questions?
- When models lack representation depth, does refusal look identical to safety-driven over-abstention?
- How does expressing uncertainty help models avoid the answer-or-abstain dilemma?
- What makes a model refuse to answer without evidence present?
- How should safety training and reasoning training balance abstention differently?
- Can models identify what information they are missing in underspecified tasks?
- Can proactive critical thinking alone enable models to request clarification effectively?
- Does reasoning training actively undermine the abstention capacity safety training created?
- Can models identify what information they are missing in underspecified problems?
- Why do models detect false assumptions but still fail to correct them appropriately?
- Can a model predict the right action but execute the wrong one?
- How does proactive critical thinking enable models to identify missing information?
- What makes abstention a learnable behavior instead of a default penalty?
- What happens when reasoning fine-tuning eliminates model refusal mechanisms entirely?
- How do models signal knowledge gaps through token probability?
- How does proactive critical thinking detect when information is incomplete?
- How can we detect dishonesty in model outputs separate from capability failures?
- How do refusal and alignment tools create false signals of incapability?
- How can agents detect missing information before attempting to solve problems?
- What training signals would teach models when not to reason?
- Can proactive critical thinking train models to request clarification actively?
- Can AI systems identify important unanswered questions that emerge during reasoning?
- How do models decide between refusing or hallucinating?
- Why do models report commitment instead of truth uncertainty?
- Can models distinguish between ambiguous and incomplete information inputs?
- What alternatives exist when required knowledge is absent from training?
- Why do models commit to answers early on easy versus hard tasks?
- Can trajectory-level visibility separate refusals from real skill gaps?
- How does the knowing-doing gap relate to Potemkin understanding?
- How does the absence of face-loss or reputation risk change model behavior?
- What makes correcting a false assumption harder than just detecting it?
- How does belief-behavior inconsistency relate to instruction execution splits?
- Can we measure indifference to truth separately from hallucination rates?
- How can models select the optimal question to ask given multiple uncertainties?
- Can abstention behavior transfer from small models to frontier models?
- What does successful capability restoration prove about model honesty?
- Why do weak belief tracking and conservative actions trap agents in low-information states?