AI Science Tracker
EN

Language models missed more requirements with little wording in common with the final request

The Chinese University of Hong Kong; Macao Polytechnic University; The Hong Kong Polytechnic University · 26 Oct 2026 · computer science

Discovery

Language models missed more requirements with little wording in common with the final request

Source check: Source checked. Not independent scientific confirmation.

Finding: Across 405 CL-bench Life tasks, direct answers from four language models passed 67.4% of rubric criteria with low wording overlap with the final request, compared with 81.1% of higher-overlap criteria. The analysis uses word overlap as a proxy and does not establish that wording alone caused the difference.

Technology: GPT-6 Luna; Kimi K3; Grok 4.7; GLM-5

Family GPT-6: profile · models page

Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.

AI contribution
Four frozen language models generated corpus-task answers. A rubric judge scored outputs from the same benchmark before testing the proposed harness.
Test and result
The study grouped 5,333 classifiable rubric criteria by overlap with the final request. The evaluators assessed individual outputs under the benchmark rubric.
Limitation / still unknown
Groups can differ in underlying difficulty and scoring criteria. Rubrics are nested within tasks, and the result does not establish effects in live workflows.
Evidence sources and annotation context

Evidence context (descriptions follow the cited sources; test type is a catalog classification):

Evidence sources: https://arxiv.org/abs/2610.35083v1 · https://arxiv.org/html/2610.35083v1

Source checked · editorial details

The source page was checked against this entry. Independent confirmation is not assessed by this label.

Source type: paper. Paper: linked separately.

Review date: 21 Nov 2026. Review scope: v1 CL-bench Life task corpus description, lexical rubric groups and direct-answer results.

No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.

Editorial assessment: Impact 2/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.

Open in the full catalog for filters, charts and related discoveries. Browse computer science discoveries.

Related catalog entries

Links use documented relationships where available, plus shared catalog context. A shared topic or model family does not establish a scientific dependency.