AI conference's papers contaminated by AI hallucinations

GPTZero, a detector of AI output, has found yet again that scientists are undermining their credibility by relying on unreliable AI assistance. The New York-based biz has identified 100 hallucinations in more than 51 papers accepted by the Conference on Neural Information Processing Systems (NeurIPS). This finding follows the …

[Bebu sa Ware]

Possibly a silver lining ?

Once the turd infested waters of research publication are utterly polluted with AI enshittified slop and rise above our necks the incentive to publish old fashioned human generated slop might diminish and disappear.

I would imagine the scale of research publication then being very much smaller but still quite specialized, and more reputation based even with the hazard of becoming more elitist. Fragmentation along national and language boundaries is also a possibility albeit an undesirable one.

My feeling is that the increasing problem of slop shitting in the nest of research publication existed for decades before the advent of the current AI hallucination but adding AI shit might well be the last straw.

[LionelB]

Re: Possibly a silver lining ?

> My feeling is that the increasing problem of slop shitting in the nest of research publication existed for decades before the advent of the current AI hallucination …

Absolutely. Apart from the proliferation of junk/scam journals and junk/scam conferences, even reputable journals, and especially conferences, have a hard enough time trying to review the tsunami of low-quality submissions – and that's just the human-generated ones. Having said which, it's tedious, but not actually hard to check that references actually exist (I do a fair amount of reviewing myself, mostly for journals). It would be great if that could be automated … wait … perhaps we can get an AI to do that [grimace emoji].

> … but adding AI shit might well be the last straw.

The potential silver lining, I think, is that it kills off a swath of junk journals and junk conferences.

[that one in the corner]

Re: Possibly a silver lining ?

Kill off? Hah!

Wait until the "agentic AIs" get ahold of a University payment account: then the "pay to play" junk journals and conferences will go into high gear "we've never had so many worthy admissions[1], time for another on--line only journal and webinar conference".

On the plus side, as none of the authors attending are human, the budget for tea and biscuits between presentations can be repurposed; there is a new jetski model out this summer...

[1] "worthy" == "paid the Gold Fee for speedy acceptance, I mean Peer Review (cough)" after the AI stumbled across a Paypal account that Sociology set up with their normal password.

[Doctor Syntax]

It's an AI conference The organisers would have been worried if there wasn't any.

[Expect Great Things]

We certainly hope the AI models these academics use to support their work are being rapidly trained on the peer-reviewed papers published by these prestigious conferences. Just imagine the speed at which the field will advance when it is properly fertilized by its own output.

[amanfromMars 1]

Victims'r'Us

"We're in an AI arms race and it's not clear the defenders can withstand the siege."

Methinks it is widely known and both graciously and gratefully accepted in elite and elevated circles of mass universal influence that opponents competing and seeking secure protection against that most particular and very peculiar attacking siege engine have no available effective guaranteed defence utilities facilitating their safe harbour escape and/or non-punitive quarantine.

Such is the new norm in these novel times and spooky places with virtual spaces exercising phenomenal Neurolinguistic Programming Expertise presenting Pioneering XSSXXXXpert Teases.

And dogged denial of it in the face of all evidence being presented proving it has one surprisingly quickly overwhelmed and subsumed. Don't be that dumb fool and blind tool.

[Anonymous Coward]

Perty bad

I guess the "business model that has prioritized volume over quality" is partly responsible, but also the bean-counting academic career promotion process that does the same. Per GPTZero's 'has identified' TFA link, winners of the 2025 NeurIPS detected vibe citation per-paper pageant are (with number of vibe cites and total cites):

13 of 65 (20%)- New York University + Prescient Design + Genentech - Efficient semantic uncertainty quantification in language models

6 of 34 (18%)- North Carolina State University - Fourier Token Merging: Understanding and Capitalizing Frequency Domain

5 of 93 (5%)- University of Maryland - Robust Reinforcement Learning in Finance: Modeling Market Impact

4 of 70 (6%)- New York University - FlowMixer: A Depth-Agnostic Neural Architecture

The ElReg should be mandatory reading for researchers considering AI-augmentation of their productivity, if only to avoid ending up with outsized throbbing appendages and related gaping holes imho! ;)

[Anonymous Coward]

It's an AI conference not a science conference. There is no validation or testing of their findings. Most of the work presented at these events has, for years, been largely worthless and incremental.

[amanfromMars 1]

Does AI Conspire, Inspire, Aspire or Simply Assume the Exercise of Greatness ‽

It's an AI conference not a science conference. There is no validation or testing of their findings. Most of the work presented at these events has, for years, been largely worthless and incremental. .... Anonymous Coward

And whenever it is realised the findings of AI are testing for the presence of any sort of worthy human intelligence and the debilitating extent of apparent general Earthly stupidity ..... the extent of which at least one well enough known human thinker and universal tinkerer believes to be infinite ...

“Two things are infinite: the universe and human stupidity; and I'm not sure about the universe.” .... Albert Einstein

...... and the tests prove a range clearly barely raised above basic primitive and thus of no future oppositional concern to AI and Large Learned Language Machines of Advanced IntelAIgent and/or Alien Design, what then do you imagine comes next whenever not playing nice and acting and reacting really stupidly ???

Go on, spoil yourself ... and use your imagination to discover at least a few of the countless available answers.

You are able to do that, are you not? And don't let the fact that much of what you can discover and uncover be designedly terrifying and terrorising stop you, for then all one needs to do is change course and explore another freely available root and route to a much better and brighter destination with all manner of Absolutely Fabulous and Fabless SMARTR Starting Points.

[Anonymous Coward]

Re: Does AI Conspire, Inspire, Aspire or Simply Transpire and Expire in Ultimate Exhaustion‽

Intelligence's been a moving target ever since the beginnings of time. It once meant the ability to lure wilderbeasts into cunning traps so they could be turned into steak tartare and shish kabobs. One'd have to be very naïve to think we'd just leave that defṉ wherever it's at now in the face of oblique tech advances (so-called) that purport to imitate it, however poorly.

It's an infinite treadmill rat race alright and we're not just the referee, we've got our five+ fingers on the dials! (except for 'em turncoat snake oil surrender monkey salespersons ...)

[Random as if !]

The empress same clothes

Just that we can all see it now , and it's as niche as Facebook, it's dead , they will never make money out of this