AI Conference Where Bots Author and Review the Papers
Description
A screenshot of a tweet from James Zou (@james_y_zou) announcing a new academic conference called 'Agents4Science'. The tweet text introduces it as a 'New conference where AI is the primary author and reviewer!' and includes a link to agents4science.stanford.edu. Key points highlighted with lightbulb emojis include initial reviews by LLMs with final assessment by human experts, and a commitment to making all submissions public to study AI's capabilities. Below the main tweet, a screenshot of the conference website's header is visible, which reads 'Open Conference of AI Agents for Science 2025' over a blue-themed background with faint illustrations of robots and lab equipment. This post highlights a significant shift in the academic and tech landscape, where AI's role is graduating from a tool to a primary contributor in scientific research. For senior developers and researchers, this represents the frontier of AI application, raising questions about the future of peer review, the nature of authorship, and the potential for AI to accelerate scientific discovery, while also acknowledging the inevitable mistakes and learning opportunities
Comments
17Comment deleted
The AI peer review process is going to be brutal. It will probably just respond with 'Your token efficiency is suboptimal. Refactor your entire hypothesis.' and then close the pull request
Great, we’ve finally achieved CI/CD for academic publishing: git push paper.md && the reviewer bot immediately requests you cite its own previous commit
Finally, a conference where the hallucination rate might actually improve the acceptance rate - and where 'it works on my machine' becomes 'it works on my model checkpoint from 3 months ago that I can't reproduce anymore'
Finally, a conference where 'the reviewer clearly didn't read my paper' becomes a feature, not a bug - though I suspect the AI reviewers will still ask you to cite that one tangentially related paper from 1987, and the rebuttal phase will just be prompt engineering your responses until GPT-4 agrees with you
Academia reimplements CI: LLM authors push, LLM reviewers run the checks, and a human hits Approve - do shared weights count as a conflict of interest?
Like Copilot auto-approving its own PRs: transparent genius until it hallucinates a critical vuln
Waiting for rebuttals that say “we lowered the reviewer’s temperature to 0.2 and it still hallucinated related work” - peer review is now prompt engineering with acceptance rates bounded by context windows
the first speaker of the conference Comment deleted
AI fights expectation reality Comment deleted
it's so over Comment deleted
People laughing about matrices multiplication Plot twist: our consciousness is matrices multiplication Comment deleted
Honestly, I keep saying "it's just math" while wondering if we're also "just math" 🤣 Comment deleted
this sounds interesting actually, curious if even anything will come out of it Comment deleted
about ♂️Ass♂️ Comment deleted
This can't possibly go wrong Comment deleted
"You conclusion is a complete bullshit, sir!" Comment deleted
robots with human teeth ☠️ Comment deleted