Peer-reviewed by human experts: AI failed in key steps to generate a scoping review on the neural mechanisms of cross-education
Bibliographic record
Abstract
The integration of Large Language Models (LLMs) into scientific writing presents significant opportunities for scholars but also risks, including misinformation and plagiarism. A new body of literature is shaping to verify the capability of LLMs to execute the complex tasks that are inherent to academic publishing. In this context this study was driven by the need to critically assess LLM's out-of-the-box performance in generating evidence synthesis reviews. To this end, the signature topic of the authors' group, cross-education of voluntary force, was chosen as a model. We prompted a popular LLM (Gemini 2.5 Pro, Deep Research enabled) to generate a scoping review on the neural mechanisms underpinning cross-education. The resulting unedited manuscript was submitted for formal peer-review to four leading subject-matter experts. Their qualitative feedback on manuscript's structure, content, and integrity was collated and analyzed. Peer-reviewers identified critical failures at fundamental stages of the review process. The LLM failed to: (1) identify specific research questions; (2) adhere to established methodological frameworks; (3) implement trustworthy search strategies; (4) objectively synthesize data. Importantly, the Results section was deemed interpretative rather than descriptive. Referencing was agreed as the worst issue being inaccurate, biased toward open-access sources (84%), and containing instances of plagiarism. The LLM also failed to hierarchize evidence, presenting minor or underexplored findings as established evidence. The LLM generated a non-systematic, poorly structured, and unreliable narrative review. These findings suggest that the selected LLM is incapable of autonomously performing scientific synthesis and requires massive human supervision to correct the observed issues.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.373 | 0.743 |
| Meta-epidemiology (narrow) | 0.002 | 0.003 |
| Meta-epidemiology (broad) | 0.004 | 0.004 |
| Bibliometrics | 0.013 | 0.010 |
| Science and technology studies | 0.007 | 0.014 |
| Scholarly communication | 0.017 | 0.016 |
| Open science | 0.007 | 0.016 |
| Research integrity | 0.011 | 0.013 |
| Insufficient payload (model declined to judge) | 0.024 | 0.019 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; the direct Gemma label and the distilled Codex classifier agree on what is shown here.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".