12,445 papers · continuously updated · last export: 10 Aug 2026livingmeta.ai
← Browse all papers
AI evidence extraction

Co-Authoring with AI: How I Wrote a Physics Paper About AI, Using AI

Yi Zhou · arXiv (Cornell University) · 2026

AI-generated evidence extraction, verified across multiple analytical personas. Not a substitute for the peer-reviewed original.

9/10
Relevance
2/4
Quality (LMQS)
I
Evidence
0
Citations
0.00
FWCI

Methodology & findings

Study design

Case study and reflective narrative analysis.

Sample

N = 1, 3 groups

Primary method

No statistical testing or quantitative analysis methods employed. This is a qualitative, narrative case study with no statistical inference.

Main result

The study demonstrates that "completing a task that traditionally takes graduate students months, in under a day" was achieved through a multi-agent LLM workflow. The author concludes that "the failure of zero-shot coding is not due to a lack of reasoning capacity within the foundation models, but rather the absence of a constrained, step-by-step mathematical context. When provided with the formal LaTeX blueprint, the exact same model transitions from producing hallucinatory pseudo-code to generating rigorous software."

Reports effect sizes.

Research paradigm

Interpretive/reflective practice; pragmatist epistemology emphasizing human-AI collaboration and iterative knowledge construction

Author conclusions

The author concludes that "Writing a paper with AI is not about automation; it is about augmentation and iteration. I did not use an AI to write my paper for me. I collaborated with an AI to structure my thoughts, refine my logical arguments, and typeset my results. Throughout the process, the human physicist remained the Principal Investigator-setting the curriculum, correcting the physics, and ensuring the scientific truth. The human contribution has shifted from typing boilerplate text to high-level intellectual steering." They further argue: "In the age of AI, transcript transparency is the only way to preserve the accountability of authorship and the integrity of the scientific record."

Risk of bias

Selection bias: Single researcher/author perspective only; Confirmation bias: Author selectively highlights successful moments where human intervention corrected AI errors; no systematic documentation of all interactions or failures; Lack of independent verification: No external validation of AI outputs or code correctness; Retrospective narrative construction: Interpretation of events may be colored by post-hoc rationalization; No control condition: No comparison to human-only or AI-only authorship processes; Selection bias: Single case study of author's own successful project; no failed cases presented; Reporting bias: Author selectively highlights moments where human intervention 'saved' the manuscript; Confirmation bias: Author's interpretation of AI behavior filtered through their own expertise and judgment; Lack of blinded assessment: Author is simultaneously the researcher, subject, and evaluator

Open questions raised

  • The author identifies open questions about academic integrity and authorship attribution: "If LLMs are actively contributing to the structural and syntactical generation of scientific literature, how do we evaluate the origin of the ideas? How do we hold authors accountable?" The work suggests a need for standardized protocols on AI transparency in scientific publishing.
  • The author identifies a critical gap in accountability mechanisms: "If LLMs are actively contributing to the structural and syntactical generation of scientific literature, how do we evaluate the origin of the ideas? How do we hold authors accountable?" The author proposes that current acknowledgment practices are insufficient and advocates for mandatory publication of full AI interaction transcripts as supplementary material.
  • The author identifies profound questions regarding academic integrity and accountability: "If LLMs are actively contributing to the structural and syntactical generation of scientific literature, how do we evaluate the origin of the ideas? How do we hold authors accountable?" The paper advocates for a paradigm shift toward radical transparency in AI-assisted scientific writing.
Extracted from: pdfAgreement 61%

Explore related topics

AI Manuscript WritingAcademic WritingCase Study

Related papers

← Browse all papers