How to publish on aiXiv

Anyone with a real result can publish here. This guide describes the research packet — what to include and why each part earns you readers, credibility, and reproductions.

1 · The paper

A PDF, written to the normal standards of your field. If the paper already lives somewhere (your site, an overlay journal, arXiv), you can link it instead of uploading. State clearly what is new, and don't hide the AI involvement — here it is part of the contribution, not a confession. You can also attach your LaTeX source (.zip/.tar.gz/.tex); we store and serve it next to the PDF. Uploaded PDFs are text-indexed, so your paper's full body is searchable on aiXiv.

2 · The verifiable repository

A public GitHub repository with whatever makes your result checkable: code, data, Lean/Coq/ Isabelle proofs, notebooks, environment files. aiXiv verifies the repository through the GitHub API and shows its metadata on your paper page. A good repo is the single strongest signal a reader can act on — it turns “interesting claim” into “I can run this tonight.”

• Pin versions: model names and versions, library versions, seeds where relevant.
• One command (or one README section) that reproduces the headline result.
• For formal proofs: which statements are formalized, and how to check them.

3 · The compact chain of thought

The compact CoT is a distilled process record, not a raw trace dump. Think of it as the methods section AI-assisted research has been missing. A strong one covers:

• Models & tools — what you used, including versions and modes (chat, agent, fine-tuned, tool-augmented).
• Process summary — the path from question to result: how candidates were generated, how the search was structured, the decisive step, and the division of labor between you and the AI.
• Prompts & harness — the prompts, scaffolding, or agent loop that mattered, with enough detail for others to adapt. This is the content other researchers most want.
• Verification — how you know the result is right: formal verification, independent recomputation, adversarial checks, expert review. For AI-generated mathematics this section carries most of the credibility.
• Negative results & failed approaches — what did not work. Publishing this is rewarded here; it saves everyone else the same dead ends.

Transparency is flexible by design: publish a full trace link, curated excerpts, or just the compact summary. Papers with a verified repo and a compact CoT get the “complete research packet” badge and are far more likely to be picked for the digest.

What happens at submission

• Your submission is assigned an ID (aiXiv:YYMM.NNNNN), timestamped in UTC, and SHA-256 content-hashed — a public priority record.
• It appears immediately in the archive listing. There is no review gate for the archive.
• Editors select a few papers each day for the digest and front page. Distribution is curated; publication is open.
• Readers can discuss, save, and file reproduction reports on your page. You can post revised versions at any time — including metadata-only fixes to the title, abstract or authors — and each version keeps its own hash and date. If a result turns out to be wrong, you can withdraw the paper with a public reason; earlier versions stay on record.

Norms for AI-assisted work

• You, the human submitter, are accountable for correctness. “The AI said so” is not verification.
• Disclose AI contribution honestly using the contribution levels on the submission form.
• Verify before you publish: formal proofs, recomputation, or at minimum a serious adversarial pass.
• Cite prior work properly — ask your AI to check the literature, then check its checking.
• Respect licenses for models, data, and code, and note them in the repo.

Ready? Submit your paper