AI Science Tracker
EN

Google

2 discoveries from 2026

Discoveries

Google

Agents reversed fewer correct judgments when they recorded evidence before feedback

Source check: Source checked. Not independent scientific confirmation.

Finding: Across 75 passing draft evaluations, five Gemini models reversed a correct decision in 41 standard trials and 34 trials when the models wrote a claim checklist after feedback. The full pre-feedback checklist condition recorded 9 reversals. The test used a fixed set of deliberately mistaken critiques.

Technology: Gemini language models; controlled claim and feedback ordering

Family Gemini: profile · models page

Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.

AI contribution
Five Gemini models reevaluated their reports, with the timing of recording earlier claims varied around identical mistaken criticism.
Test and result
A controlled report-review study compared the order of evidential self-assessment and feedback, with 20 scenarios per model and separate cases containing valid correction.
Limitation / still unknown
Scenario and model dependence matter: the same scenarios were reused and the benchmark uses selected passing or flawed drafts.
Evidence sources and annotation context

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

Evidence sources: https://arxiv.org/abs/2610.34654v1 · https://arxiv.org/html/2610.34654v1

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 controlled Turn-2 report review, §4.3 Table 3 and evaluation limits.

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 3/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.

Google

Language models flagged missing evidence less often when an unsupported premise was usable

Source check: Source checked. Not independent scientific confirmation.

Finding: In paired arithmetic tasks, making an unsupported premise usable reduced missing-evidence flags by 8–30 percentage points across four tested language models. The source support was unchanged. Some answers may reflect conditional reasoning because the instructions did not require rejection of premises supplied in the question.

Technology: Gemma and Qwen language models

Family Qwen: profile · models page

Publication: Preprint · Test context: direct test. Catalog categories, not an evidence grade.

AI contribution
Four pretrained Gemma and Qwen language models answered mathematically matched prompts with equally unsupported premises that varied in usability.
Test and result
The controlled factorial design varied actionability and the mention of relevant entities on GSM8K. All 16 primary paired contrasts survived Holm adjustment.
Limitation / still unknown
The selected tasks and instructions limit how broadly the result applies. It does not show that all answers accept unsupported statements as facts. Pattern-based scoring was revised after a 40-response audit; residual condition-dependent scoring error was not bounded by that small check.
Evidence sources and annotation context

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

Evidence sources: https://arxiv.org/abs/2610.34488v1 · https://arxiv.org/html/2610.34488v1

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 Methods, paired premise design, §4 and Limitations.

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 3/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.

Open the interactive profile for the timeline, or browse all organizations and model families.