Chirag Gupta

@WalentekLab @tdietterich @arxiv Ah! I see where the anti-AI stand is coming from. Never mind. I am not going to waste my time with you anymore.

@seekergupta @tdietterich @arxiv Not really. That's why many current ai "professionals" need so much to dial it up and make exaggerated claims to make their point. That's also why the companies lie, steal, build circular financial deals and try to broker with politict to get support without regulation. They know

Attention @arxiv authors: Our Code of Conduct states that by signing your name as an author of a paper, each author takes full responsibility for all its contents, irrespective of how the contents were generated. 1/

@WalentekLab @tdietterich @arxiv Yes, we are in that era where the professionals are making things reliable at scale.

@seekergupta @tdietterich @arxiv Also the claim of early adopters will win is not true. The very early adopters usually die or go broke, then the professionals make things reliable, scale it up and make the buck. Just think flight, electricity, nuclear power, and media.

@WalentekLab @tdietterich @arxiv So what happens to scientist who don’t speak YOUR favorite language and use LLMs to translate. Or you don’t think they are scientists?

@seekergupta @tdietterich @arxiv So llms and image generators should be banned from sci publication and review unless they are the subject of study.

@LucaAmb @littmath Specially with large consortium papers with sometimes >30 authors. One mistake and you are gone.

@littmath Let's see. Consider a PI with 10 students publishing 15 papers a year... statistics in not on your side...

I am quite convinced that, under these arxive guidelines, every single major PI in the field will be banned within a few years

For no reason at all, I made a simple one-page guide for self-publishing research without relying on journals or other centralized gatekeepers. It covers: – DOI deposits via Zenodo – FAIR repositories – crawlable HTML for Google + AI systems – metadata, licenses, datasets, code, and protocols

The most consequential thing a senior bioinformatician can do is build infrastructure other people use without thinking about it. The most undervalued thing a senior bioinformatician can do is build infrastructure other people use without thinking about it. Same sentence.

Yes but that’s difficult to implement. If arxiv sets an example, biorxiv and then medrxiv follows. Who’s stopping them? What’s remains are, well, blogging forums. I am not even arguing that the journals, who want exposure only through them, are behind this. That would make a solid conspiracy theory worth testing.

@seekergupta @QRDL Yes, this is why I only said -some- servers should have strict moderation policies. There should be a range of options with different standards, that have different roles.

Many people are not actually reading this. Arxiv is -not- banning the use of AI, or papers which used AI to generate proofs, code, etc. They are banning people who upload papers in which the AI content was (very clearly) not actually checked by the human author(s).

@thomasfbloom @QRDL I see. We come from two different schools of thought. If all preprint servers start implementing such policies, people from several parts of the world will have no chance of showcasing their work. Gate keeping science will not get us anywhere.

@seekergupta @QRDL On at least some public forums, yes. Moderation is sometimes necessary, to avoid the signal being drowned out by the noise. There is room for many different forums, with different moderation standards, serving different needs.

@thomasfbloom @QRDL You mean we start gate keeping who can publish on a public forum?

@seekergupta @QRDL Certainly the landscape of how papers and results are announced will have to change, with multiple 'levels' of moderation, but there will always be a need for a preprint server with high standards for moderation and accuracy. It makes sense if the arxiv becomes that.

Yes agree to that. And the policy should be clear and transparent. A completely fabricated reference list is obviously a bad actor. And it’s not just the LLM policy. Arxiv has been tightening their overall policies: pre peer preview, endorsement etc. That says a lot about where this is headed.

@seekergupta @QRDL but e.g. many wrong references could be evidence towards unvetted use of LLMs, which when combined with other evidence of this could result in a ban.

@thomasfbloom @QRDL Here

@thomasfbloom @QRDL There is a lot of trust involved here. One bad actor, or someone who genuinely misses a bad reference — which wasn’t even LLM generated— can get others banned too. Wrong citations are not a new thing. It was rampant even before LLMs.

@thomasfbloom @QRDL Do you expect 20 co-authors on a paper to all verify each reference in every paper they are a part of? If one misses, face a year long ban? Thats effective AI ban, and gate-keeping. Arxiv is not a journal.

@QRDL No, many people are using AI to help with their research, and not blindly copying the output into their papers without even reading it first. If you do not care enough about your paper to actually read it yourself before posting, why should it be made public?

Totally agree. But we have to be aware that we will be blocking a whole new technology. It’s like blocking the use of the internet in its early days and forcing people to read physical copies of manuscripts to source knowledge. Where does the buck stop with AI? Just the LLMs, or AI use in designing drugs too? Should scientists be allowed to use automated labs to test THEIR own hypotheses at scale, and publish them? The ones who do and the ones who let do — the early adopters — will win. (Yes the em dash is my own, it’s a habit)

@seekergupta @tdietterich @arxiv I think journal should disallow llm/agent use and we'd be where it worked (not perfectly but predictably) fir quite a while. Once llm gets mature and not a random word machine that makes up stuff, we can talk again imo.

Yes I do. Our agent provides a direct link to whatever paper it references. The question here is, what happens when a consortium paper is submitted to arxiv with 30 authors who co-wrote the paper, and one of them mis-cited? Do all authors get banned for a year for this? One can argue that the authors were supposed to read and verify the manuscript. Yes, and they generally do, but I can guarantee no co-author checks and verifies all references. Specially senior authors who submit multiple manuscripts a year from their labs. We have always relied on peer review to catch such inconsistencies. Plus the journals have/should have such checkpoints. IMO, the journals should disallow citations to preprints and tighten their peer review policies. Arguably, an AI reviewer pre-screen can take some of that burden off the journals (and preprint servers)

@seekergupta @tdietterich @arxiv Don't you trust the agents you build?

@pangramlabs Meaning this is not a new problem. This always existed. What happens and the Y axis is on a log scale?

For context:

We can’t agentify stupidity. We still need humans for that.

@nabla_theta @tdietterich @arxiv With a peer-review in between. Important distinction.

@seekergupta @tdietterich @arxiv arxiv is the print. the other journals are merely the postprints

If arXiv decides to do active gate checking on LLM slops, it also has the responsibility to release the stats on rejection rate and on hold delay, and explain the decision making process in more transparency.

@DGoweyAuthor @tdietterich @arxiv The “randos” are the ones who keep their shop open.

@seekergupta @tdietterich @arxiv I'm sure they really appreciate randos telling them how to run their archive so keep it up