cutting-to-the-chase CAP-0053 · proposedCutting to the chase
Proposed. Several papers converged on this framing, so the pipeline added it. Nobody has decided it is the right way to carve up the subject — it may be two topics, or a duplicate of another, or not a topic at all. Claims from those papers arrive nightly. Say so if the carving is wrong, in either direction.
Giving the answer directly when that's what is asked, without preamble, restatement or visible working the reader didn't request; the failure is padding the model was told to skip, or skipping working the task needed.
Also called: answer-only compliance, no-thinking, concision, verbosity, overthinking
Tags: behavior, chat-assistant
Named by Russ, 2026-10-18, from the Thinking Inertia paper, which found that telling a model not to think, or switching its thinking mode off, does not reliably stop visible reasoning. The capability is about what the reader sees, not about internal computation: a model may reason privately and still answer directly, and that is success. It has two sides, and both belong here: the padding a model adds when told not to (preamble, restating the question, step-by-step working), and knowing when the shortcut is safe, since answering directly cost nothing on closed-answer questions and cost accuracy on open-ended ones. Over-long answers are among the commonest complaints about models, and the judge-bias claims already filed show judges reward length, which is one route by which models are trained toward padding.
What counts as this capability
Scope boundary used when deciding whether a paper is really about this capability, rather than merely mentioning it.
In scope when the paper measures whether a model gives a direct answer when asked, the length or padding of its output relative to what the task needed, or reasoning that persists or expands beyond what was requested. NOT in scope for whether stated reasoning is faithful (explanation faithfulness), for judges preferring longer answers (critique), or for internal reasoning that never reaches the output.
Claims
- mechanismsingle paper↑1↓0In LLM-based repository-level program repair, telling the generator to be minimal or constraining its output format does not shrink patches, because verbosity comes from the capability-raising machinery itself (iterative refinement, broad context); a separate post-generation refiner that rewrites an already-plausible patch is what reduces size without losing fixes.
- observationsingle paper↑1↓0Across six models and six prompting interventions on Boolean, multiple-choice and open-ended questions, explicit controls to stop reasoning — a disabled thinking mode or an instruction to answer only — do not reliably eliminate visible inference before the answer: reasoning persists even under strict controls ("thinking inertia"), and it becomes more prevalent as the answer space opens, least on Boolean questions and most on open-ended ones; so a direct answer cannot be assumed from model settings or instructions alone.
- observationsingle paper↑1↓0When a model is asked to answer without visible reasoning, rewriting the same question with candidate answers supplied makes answer-only responses easier to produce, and on Boolean and multiple-choice questions answering directly leaves accuracy unchanged — whereas on open-ended questions there is a trade-off, and forcing an answer-only response costs accuracy.
- observationsingle paper↑1↓0For a same-family pair of reasoning and instruction-tuned 4B models, learning one interpolation ratio per module (attention block, FFN block, embedding, final norm) with a reward that scores correctness and ranks correct responses by length cuts average generated tokens by about 27% relative to the reasoning model while holding average accuracy, and beats fixed global-ratio merging and CMA-ES search over the same ratios.
Techniques
- Learn per-module merge ratios with a length-aware rewardtrainingFreeze a reasoning model and an instruction model of identical architecture, assign one sigmoid-bounded interpolation ratio per module, and update only those scalars with GRPO-style rollouts rewarded for correct and comparatively short answ
- Offer the candidate answerspromptingWhen a direct, answer-only response is wanted, give the model the possible answers to choose from instead of leaving the answer space open.
- Shrink the patch after generation, not during itprocessA small refiner model (SFT plus DPO on verbose-to-concise patch pairs with distilled reasoning traces) is plugged in after an existing repair framework's patch generation to rewrite the candidate patch into a smaller one, with optional judg
Sources
Every paper, post or observation behind this capability. Ones cited by a claim show the claim and its stance; ones filed directly are about the capability but have no claim drafted from them yet. Open one to add support or contest it.
- ↑ supports a claim here · ↑ supports a claim here
- ↑ supports a claim here
- ↑ supports a claim here
Capabilities are a way of carving up the subject, and carvings are arguable. Say so if this one is wrong — especially a proposed one, which a pipeline added because several papers used the same framing, not because anyone decided it was right.