Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

GPTZero finds 100 new hallucinations in NeurIPS 2025 accepted papers

gptzero.me

Post reply on HN
Print view View original

481–490 of 528 posts

Re: GPTZero finds 100 new hallucinations in NeurIPS 2025 accepted papers

#481

Post by tsimionescu »

aydyn wrote: post #443
neilv wrote: post #438

Earlier quoted context omitted.

> Clearly, the authors in NeurIPS don't agree that using an LLM to help write is "plagiarism", Or they didn't consider that it arguably fell within academia's definition of plagiarism. Or they thought they could get away with it. Why is someone behaving questionably the authority on whether that's OK? > Nobody I know in real life, personally or at work, has expressed this belief. I have literally only ever encountere…

> Why is someone behaving questionably the authority on whether that's OK? Because they are not. Using AI to help writing is something literally every company is pushing for.

How is that relevant? Companies care very little about plagiarism, at least in the ethical sense (they do care if they think it's a legal risk, but that has turned out to not be the case with AI, so far at least).

Re: GPTZero finds 100 new hallucinations in NeurIPS 2025 accepted papers

#482

Post by gtirloni »

Der_Einzige wrote: post #331
gtirloni wrote: post #230

Why focus on hallucinations/LLMs and not on the authors? There are rules for submitting papers. If I drop a loaded gun and it fires, killing someone, we don't go after the gun's manufacturer in most cases.

Actually, if you’re the US navy, you DO go after the manufacturer! Go look up the P320 pistol and the tons of accidental discharges that’s it’s caused. https://stateline.org/2025/03/10/more-law-enforcement-agenci...

Thanks. But's not actually the point I'm trying to make.

What I'm saying is that the authors have a responsibility, whether they wrote the papers themselves, asked an AI to write and didn't read it thoroughly, or asked their grandparents while on LSD to write it... it all comes back to whoever put their names on the paper and submitted it.

I think AI is a red herring here.

Re: GPTZero finds 100 new hallucinations in NeurIPS 2025 accepted papers

#483

Post by necovek »

aydyn wrote: post #443
neilv wrote: post #438

Earlier quoted context omitted.

> Clearly, the authors in NeurIPS don't agree that using an LLM to help write is "plagiarism", Or they didn't consider that it arguably fell within academia's definition of plagiarism. Or they thought they could get away with it. Why is someone behaving questionably the authority on whether that's OK? > Nobody I know in real life, personally or at work, has expressed this belief. I have literally only ever encountere…

> Why is someone behaving questionably the authority on whether that's OK? Because they are not. Using AI to help writing is something literally every company is pushing for.

As long as AI companies have paid them to train on their data (see a number of licensing deals between OpenAI and news agencies and such).

Re: GPTZero finds 100 new hallucinations in NeurIPS 2025 accepted papers

#484

Post by bjourne »

thaumasiotes wrote: post #428
bjourne wrote: post #393

Earlier quoted context omitted.

There are legitimate , non-cheating ways to use LLMs for writing. I often use the wrong verb forms ("They synthesizes the ..."), write "though" when it should be "although", and forget to comma-separate clauses. LLMs are perfect for that. Generating text from scratch, however, is wrong.

> I often ... write "though" when it should be "although" That is a purely imaginary "error". Anywhere you can use 'although', you are free to use 'though' instead.

Yeah, but you cannot use although anywhere you can use though, though.

Re: GPTZero finds 100 new hallucinations in NeurIPS 2025 accepted papers

#485

Post by necovek »

rustystump wrote: post #435
falkensmaize wrote: post #429

Earlier quoted context omitted.

“anti-ai sentiment” No that’s a straw man, sorry. Skepticism is not the same thing as irrational rejection. It means that I don’t believe you until you’ve proven with evidence that what you’re saying is true. The efficacy and reliability of LLMs requires proof. Ai companies are pouring extraordinary, unprecedented amounts of money into promoting the idea that their products are intelligent and trustworthy. That marke…

The cat is out of the bag tho. AI does have provably crazy value. Certainly not the agi hype marketing spews and who knows how economically viable it would be without vc. However, i think any one who is still skeptical of the real efficacy is willfully ignorant. This is not a moral endorsement on how it was made or if it is moral to use but god damn it is a game changer across vast domains.

There were a number of studies already shared reporting on the impression of increased efficiency without the actual increase in efficiency.

Which means that it's still not a given, though there are obviously cases where individual cases seem to be good proof of it.

Re: GPTZero finds 100 new hallucinations in NeurIPS 2025 accepted papers

#486

Post by bjourne »

j2kun wrote: post #430
bjourne wrote: post #391

Earlier quoted context omitted.

Sorry, but blaming it on "AI autocomplete" is the dumbest excuse ever. Author lists come from BibTeX entries and while they often contains errors since they can come from many sources, they do not contain completely made up authors. I don't share your view that hallucinated citations are less damaging in background section. Background, related works, and introduction is the sections where citations most often show up…

I'm not blaming anything on anything, because I did not (nor did the authors) confirm the cause of any of these errors. > I don't share your view that hallucinated citations are less damaging in background section. Who exactly is damaged in this particular instance?

Trust is damaged. I cannot verify that the evidence is correct only that the conclusions follow from the evidence. I have to rely on the authors to truthfully present their evidence. If they for whatever reason add hallucinated citations to their background that trust is 100% gone.

Re: GPTZero finds 100 new hallucinations in NeurIPS 2025 accepted papers

#487

Post by naasking »

AIs are much better and hallucinate much less when they are given focused tasks, eg. instead of asking AI to write a background on complete with citations, ask the AI specifically to generate a list of citations relevant to X, programmatically check the references are correct using a true index like doi, then ask AI to use the reference list to write a background section on X.

This would be a valuable research tool that uses AI without the hallucinations.

Re: GPTZero finds 100 new hallucinations in NeurIPS 2025 accepted papers

#488

Post by mikkupikku »

gs17 wrote: post #367
ted_dunning wrote: post #357

Earlier quoted context omitted.

I can see that either way. It could also be a placeholder until the actual author list is inserted. This could happen if you know the title, but not the authors and insert a temporary reference entry.

The first Doe and Smith example I could give that to (the title is real and the arxiv ID they give is "arXiv:2401.00001", which is definitely placeholder), but the second one doesn't match a title and has fake URL/DOI that don't actually go anywhere. There's a few that are unambiguously placeholders, but they really should have been caught in review for a conference this high up.

How does a "placeholder citation" even happens? Either enter the citation properly now, or do it properly later. What role does a "placeholder citation" serve, besides giving you something to forget about and fuck up?

I do not believe the placeholder citation theory at all.

Re: GPTZero finds 100 new hallucinations in NeurIPS 2025 accepted papers

#489

Post by lallysingh »

mort96 wrote: post #217
lallysingh wrote: post #205

Earlier quoted context omitted.

On the bright side, an LLM can really help set up a reproduction environment. Perhaps repro should become the basis of peer review?

No, it can't. No LLM can purchase the equipment and chemicals and machinery you need to reproduce experiments, nor should you want it.

I was still thinking of CS :/

Re: GPTZero finds 100 new hallucinations in NeurIPS 2025 accepted papers

#490

Post by floriferous »

While this is really concerning, it feels like a small new category of errors to check for. The article mentions an increase of 220% of the amount of submissions. That's incredible news for science, probably lots of honest scientists able to produce more work and eventually lead to more science being done.
Post reply on HN