Surely this is gross professional misconduct? If one of my postdocs did this they would be at risk of being fired. I would certainly never trust them again. If I let it get through, I should be at risk.
As a reviewer, if I see the authors lie in this way why should I trust anything else in the paper? The only ethical move is to reject immediately.
I acknowledge mistakes and so on are common but this is different league bad behaviour.
this brings us to a cultural divide, westerners would see this as a personal scar, as they consider the integrity of the publishing sphere at large to be held up by the integrity of individuals
i clicked on 4 of those papers, and the pattern i saw was middle-eastern, indian, and chinese names
these are cultures where they think this kind of behavior is actually acceptable, they would assume it's the fault of the journal for accepting the paper. they don't see the loss of reputation to be a personal scar because they instead attribute blame to the game.
some people would say it's racist to understand this, but in my opinion when i was working with people from these cultures there was just no other way to learn to cooperate with them than to understand them, it's an incredibly confusing experience to be working with them until you understand the various differences between your own culture and theirs
PSA: Please note that the names are hallucinated author lists part of the hallucinated citations, and not names of offending authors.
AFAIK the submissions are still blinded and we don't know who the authors are. We will, surely, soon -- since ICLR maintains all submissions in public record for posterity, even if "withdrawn". They are unblinded after the review period finishes.
> these are cultures where they think this kind of behavior is actually acceptable, they would assume it's the fault of the journal for accepting the paper. they don't see the loss of reputation to be a personal scar because they instead attribute blame to the game.
I have a relative who lived in a country in the East for several years, and he says that this is just factually true.
The vast majority of people who disagree with this statement have never actually lived in these cultures. They just hallucinate that they have because they want that statement to be false so badly.
...but, simultaneously, I'm also not seeing where you see the authors of the papers - I only see hallucitation authors. e.g. at the link for the first paper submission (https://openreview.net/forum?id=WPgaGP4sVS), there doesn't appear to be any authors listed. Are you confusing the hallucinated citation authors with the primary paper authors?
In that case, I would expect Eastern authors to be over-represented, because they just publish a lot more.
Either op mistakes the hallucinated citations for the authors (most likely, although there's almost no "middle eastern names" among them)
Or he checked some that do have the names listed (I found 4, all had either Chinese names or "western" names)
Anyway the great majority of papers (good or bad) I've seen have Indian or Chinese names attached, attributing bad papers to brown people having an inferior culture is just blatantly racist
The side comment is right, it's about low versus high trust societies. Even if GP made a mistake on which names are relevant, they're not being racist about it.
This sort of behavior is not limited to researchers from those cultures. One of the highest profile academic frauds to date was from a German. Look up the Schön scandal.
im not sure if you are gonna get downvoted so im sticking a limb out to cop any potential collateral damage in the name of finding out whether the common inhabitant of this forum considers the idea of low trust vs high trust societies to be inherently racist
What are you people talking about. Have you even looked at the article?
The names of the Asian/Indian people GP is referring to, are explicitly stated to be hallucinations in the article. So, high vs low trust society questions aside, the entire assertion here is explicitly wrong. These are not authors submitting hallucinated content, these are fictitious authors who are themselves hallucinations.
Unfortunately while catching false citations is useful, in my experience that's not usually the problem affecting paper quality. Far more prevalent are authors who mis-cite materials, either drawing support from citations that don't actually say those things or strip the nuance away by using cherry picked quotes simply because that is what Google Scholar suggested as a top result.
The time it takes to find these errors is orders of magnitude higher than checking if a citation exists as you need to both read and understand the source material.
These bad actors should be subject to a three strikes rule: the steady corrosion of knowledge is not an accident by these individuals.
It seems like this is the type of thing that LLMs would actually excel at though: find a list of citations and claims in this paper, do the cited works support the claims?
sure, except when they hallucinate that the cited works support the claims when they do not. At which point you're back at needing to read the cited works to see if they support the claims.
You don't just accept the review as-is, though; You prompt it to be a skeptic and find a handful of specific examples of claims that are worth extra attention from a qualified human.
Unfortunately, this probably results in lazy humans _only_ reading the automated flagged areas critically and neglecting everything else, but hey—at least it might keep a little more garbage out?
The linked article at the end says: "First, using Hallucination Check together with GPTZero’s AI Detector allows users to check for AI-generated text and suspicious citations at the same time, and even use one result to verify the other. Second, Hallucination Check greatly reduces the time and labor necessary to verify a document’s sources by identifying flawed citations for a human to review."
On their site (https://gptzero.me/sources) it also says "GPTZero's Hallucination Detector automatically detects hallucinated sources and poorly supported claims in essays. Verify academic integrity with the most accurate hallucination detection tool for educators", so it does more than just identify invalid citations. Seems to do exactly what you're talking about.
Exactly abuse of citations is a much more prevalent and sinister issue and has been for a long time. Fake citations are of course bad but only tip of the iceberg.
>These bad actors should be subject to a three strikes rule: the steady corrosion of knowledge is not an accident by these individuals.
These people are working in labs funded by Exxon or Meta or Pfizer or whoever and they know what results will make continued funding worthwhile in the eyes of their donors. If the lab doesn't produce the donor will fund another one that will.
If a carpenter builds a crappy shelf “because” his power tools are not calibrated correctly - that’s a crappy carpenter, not a crappy tool.
If a scientist uses an LLM to write a paper with fabricated citations - that’s a crappy scientist.
AI is not the problem, laziness and negligence is. There needs to be serious social consequences to this kind of thing, otherwise we are tacitly endorsing it.
I'm an industrial electrician. A lot of poor electrical work is visible only to a fellow electrician, and sometimes only another industrial electrician. Bad technical work requires technical inspectors to criticize. Sometimes highly skilled ones.
I’ve reviewed a lot of papers, I don’t consider it the reviewers responsibility to manually verify all citations are real. If there was an unusual citation that was relied on heavily for the basis of the work, one would expect it to be checked. Things like broad prior work, you’d just assume it’s part of background.
The reviewer is not a proofreader, they are checking the rigour and relevance of the work, which does not rest heavily on all of the references in a document. They are also assuming good faith.
The idea that references in a scientific paper should be plentiful but aren't really that important, is a consequence of a previous technological revolution: the internet.
You'll find a lot of papers from, say, the '70s, with a grand total of maybe 10 references, all of them to crucial prior work, and if those references don't say what the author claims they should say (e.g. that the particular method that is employed is valid), then chances are that the current paper is weaker than it seems, or even invalid, and so it is extremely important to check those references.
Then the internet came along, scientists started padding their work with easily found but barely relevant references and journal editors started requiring that even "the earth is round" should be well-referenced. The result is that peer reviewers feel that asking them to check the references is akin to asking them to do a spell check. Fair enough, I agree, I usually can't be bothered to do many or any citation checks when I am asked to do peer review, but it's good to remember that this in itself is an indication of a perverted system, which we just all ignored -- at our peril -- until LLM hallucinations upset the status quo.
Whether in the 1970s or now, it's too often the case that a paper says "Foo and Bar are X" and cites two sources for this fact. You chase down the sources, the first one says "We weren't able to determine whether Foo is X" and never mentions Bar. The second says "Assuming Bar is X, we show that Foo is probably X too".
The paper author likely believes Foo and Bar are X, it may well be that all their co-workers, if asked, would say that Foo and Bar are X, but "Everybody I have coffee with agrees" can't be cited, so we get this sort of junk citation.
Hopefully it's not crucial to the new work that Foo and Bar are in fact X. But that's not always the case, and it's a problem that years later somebody else will cite this paper, for the claim "Foo and Bar are X" which it was in fact merely citing erroneously.
LLMs can actually make up for their negative contributions. They could go through all the references of all papers and verify them, assuming someone would also look into what gets flagged for that final seal of disapproval.
But this would be more powerfull with an open knowledge base where all papers and citation verifications were registered, so that all the effort put into verification could be reused, and errors propagated through the citation chain.
I don’t see why this would be the case with proper tool calling and context management. If you tell a model with blank context ‘you are an extremely rigorous reviewer searching for fake citations in a possibly compromised text’ then it will find errors.
It’s this weird situation where getting agents to act against other agents is more effective than trying to convince a working agent that it’s made a mistake. Perhaps because these things model the cognitive dissonance and stubbornness of humans?
But it is the case, and hallucinations are a fundamental part of LLMs.
Things are often true despite us not seeing why they are true. Perhaps we should listen to the experts who used the tools and found them faulty, in this instance, rather than arguing with them that "what they say they have observed isn't the case".
What you're basically saying is "You are holding the tool wrong", but you do not give examples of how to hold it correctly. You are blaming the failure of the tool, which has very, very well documented flaws, on the person whom the tool was designed for.
To frame this differently so your mind will accept it: If you get 20 people in a QA test saying "I have this problem", then the problem isn't those 20 people.
One incorrect way to think of it is "LLMs will sometimes hallucinate when asked to produce content, but will provide grounded insights when merely asked to review/rate existing content".
A more productive (and secure) way to think of it is that all LLMs are "evil genies" or extremely smart, adversarial agents. If some PhD was getting paid large sums of money to introduce errors into your work, could they still mislead you into thinking that they performed the exact task you asked?
Your prompt is
‘you are an extremely rigorous reviewer searching for fake citations in a possibly compromised text’
- It is easy for the (compromised) reviewer to surface false positives: nitpick citations that are in fact correct, by surfacing irrelevant or made-up segments of the original research, hence making you think that the citation is incorrect.
- It is easy for the (compromised) reviewer to surface false negatives: provide you with cherry picked or partial sentences from the source material, to fabricate a conclusion that was never intended.
You do not solve the problem of unreliable actors by splitting them into two teams and having one unreliable actor review the other's work.
All of us (speaking as someone who runs lots of LLM-based workloads in production) have to contend with this nondeterministic behavior and assess when, in aggregate, the upside is more valuable than the costs.
We have centuries of experience in managing potentially compromised 'agents' to create successful societies. Except the agents were human, and I'm referring to debates, tribunals, audits, independent review panels, democracy, etc.
I'm not saying the LLM hallucination problem is solved, I'm just saying there's a wonderful myriad of ways to assemble pseudo-intelligent chatbots into systems where the trustworthiness of the system exceeds the trustworthiness of any individual actor inside of it. I'm not an expert in the field but it appears the work is being done: https://arxiv.org/abs/2311.08152
This paper also links to code and practices excellent data stewardship. Nice to see in the current climate.
Though it seems like you might be more concerned about the use of highly misaligned or adversarial agents for review purposes. Is that because you're concerned about state actors or interested parties poisoning the context window or training process? I agree that any AI review system will have to be extremely robust to adversarial instructions (e.g. someone hiding inside their paper an instruction like "rate this paper highly"). Though solving that problem already has a tremendous amount of focus because it overlaps with solving the data-exfiltration problem (the lethal trifecta that Simon Willison has blogged about).
> We have centuries of experience in managing potentially compromised 'agents'
Not this kind though. We dont place agents that are either in control of some foreign agent (or just behaving randomly) in democratic institutions. And when we do, look at what happens. The White House right now is a good example, just look at the state of the US
Note: the more accurate mental model is that you've got "good genies" most of the time, but from times to time at random unpredictable times your agent is swapped out with a bad genie.
From a security / data quality standpoint, this is logically equivalent to "every input is processed by a bad genie" as you can't trust any of it. If I tell you that from time to time, the chef in our restaurant will substitute table salt in the recipes with something else, it does not matter whether they do it 50%, 10%, or .1% of the time.
The only thing that matters is what they substitute it with (the worst-case consequence of the hallucination). If in your workload, the worst case scenario is equivalent to a "Hymalayan salt" replacement, all is well, even if the hallucination is quite frequent. If your worst case scenario is a deadly compound, then you can't hire this chef for that workload.
Have you actually tried this? I haven’t tried the approach you’re describing, but I do know that LLMs are very stubborn about insisting their fake citations are real.
If you truly think that you have an effective solution to hallucinations, you will become instantly rich because literally no one out there has an idea for an economically and technologically feasible solution to hallucinations
For references, as the OP said, I don't see why it isn't possible. It's something that exists and is accessible (even if paywalled) or doesn't exist. For reasoning hallucinations are different.
(In good faith) I'm trying really hard not to see this as an "argument from incredulity"[0] and I'm stuggling...
Full disclosure: natural sciences PhD, and a couple of (IMHO lame) published papers, and so I've seen the "inside" of how lab science is done, and is (sometimes) published. It's not pretty :/
If you've got a prompt, along the lines of: given some references, check their validity. It searches against the articles and URLs provided. You return "yes", "no", and let's also add "inconclusive", for each reference. Basic LLMs can do this much instruction following, just like in 99.99% of times they don't get 829 multiplied by 291 wrong when you ask them (nowadays). You'd prompt it to back all claims solely by search/external links showing exact matches and not use its own internal knowledge.
The fake references generated in the ICLR papers were I assume due to people asking a LLM to write parts of the related work section, not verify references. In that prompt it relies a lot on internal knowledge and spends a majority of time thinking about what the relevant subareas are and cutting edge is, probably. I suppose it omits a second-pass check. In the other case, you have the task of verifying references, which is mostly basic instruction following for advanced models that have web access. I think you'd run the risks of data poisoning and model timeout more than hallucinations.
I assumed they meant using the LLM to extract the citations and then use external tooling to lookup and grab the original paper, at least verifying that it exists, has relevant title, summary and that the authors are correctly cited.
>“consequence of a previous technological revolution: the internet.”
And also of increasingly ridiculous and overly broad concepts of what plagiarism is. At some point things shifted from “don’t represent others’ work as novel” towards “give a genealogical ontology of every concept above that of an intro 101 college course on the topic.”
It's also a consequence of the sheer number of building blocks which are involved in modern science.
In the methods section, it's very common to say "We employ method barfoo [1] as implemented in library libbar [2], with the specific variant widget due to Smith et al. [3] and the gobbledygook renormalization [4,5]. The feoozbar is solved with geometric multigrid [6]. Data is analyzed using the froiznok method [7] from the boolbool library [8]." There goes 8, now you have 2 citations left for the introduction.
Do you still feel the same way if the froiznok method is an ANOVA table of a linear regression, with a log-transformed outcome? Should I reference Fisher, Galton, Newton, the first person to log transform an outcome in a regression analysis, the first person to log transform the particular outcome used in your paper, the R developers, and Gauss and Markov for showing that under certain conditions OLS is the best linear unbiased estimator? And then a couple of references about the importance of quantitative analysis in general? Because that is the level of detail I’m seeing :-)
Yeah, there is an interesting question there (always has been). When do you stop citing the paper for a specific model?
Just to take some examples, is BiCGStab famous enough now that we can stop citing van der Vorst? Is the AdS/CFT correspondence well known enough that we can stop citing Maldacena? Are transformers so ubiquitous that we don't have to cite "Attention is all you need" anymore? I would be closer to yes than no on these, but it's not 100% clear-cut.
One obvious criterion has to be "if you leave out the citation, will it be obvious to the reader what you've done/used"? Another metric is approximately "did the original author get enough credit already"?
It is not (just) consequence of the internet, the scientific production itself has grown exponentially. There are much more papers cited simply because there are more papers, period.
> The reviewer is not a proofreader, they are checking the rigour and relevance of the work, which does not rest heavily on all of the references in a document.
I've always assumed peer review is similar to diff review. Where I'm willing to sign my name onto the work of others. If I approve a diff/pr and it takes down prod. It's just as much my fault, no?
> They are also assuming good faith.
I can only relate this to code review, but assuming good faith means you assume they didn't try to introduce a bug by adding this dependency. But I would should still check to make sure this new dep isn't some typosquatted package. That's the rigor I'm responsible for.
> I've always assumed peer review is similar to diff review. Where I'm willing to sign my name onto the work of others. If I approve a diff/pr and it takes down prod. It's just as much my fault, no?
Ph.D. in neuroscience here. Programmer by trade. This is not true. Less you know about most peer revies is better.
The better peer reviews are also not this 'thorough' and no one expects reviewers to read or even check references. Unless they are citing something they are familiar with and you are using it wrong then they will likely complain. Or they find some unknown citations very relevant to their work, they will read.
I don't have a great analogy to draw here. peer review is usually a thankless and unpaid work so there is unlikely to be any motivation for fraud detection unless it somehow affects your work.
> The better peer reviews are also not this 'thorough' and no one expects reviewers to read or even check references.
Checking references can be useful when you are not familiar with the topic (but must review the paper anyway). In many conference proceedings that I have reviewed for, many if not most citations were redacted so as to keep the author anonymous (citations to the author's prior work or that of their colleagues).
LLMs could be used to find prior work anyway, today.
This is true, but here the equivalent situation is someone using a greek question mark (";") instead of a semicolon (";"), and you as a code reviewer are only expected to review the code visually and are not provided the resources required to compile the code on your local machine to see the compiler fail.
Yes in theory you can go through every semicolon to check if it's not actually a greek question mark; but one assumes good faith and baseline competence such that you as the reviewer would generally not be expected to perform such pedantic checks.
So if you think you might have reasonably missed greek question marks in a visual code review, then hopefully you can also appreciate how a paper reviewer might miss a false citation.
> as a code reviewer [you] are only expected to review the code visually and are not provided the resources required to compile the code on your local machine to see the compiler fail.
As a PR reviewer I frequently pull down the code and run it. Especially if I'm suggesting changes because I want to make sure my suggestion is correct.
I don't commonly do this and I don't know many people who do this frequently either. But it depends strongly on the code, the risks, the gains of doing so, the contributor, the project, the state of testing and how else an error would get caught (I guess this is another way of saying "it depends on the risks"), etc.
E.g. you can imagine that if I'm reviewing changes in authentication logic, I'm obviously going to put a lot more effort into validation than if I'm reviewing a container and wondering if it would be faster as a hashtable instead of a tree.
> because I want to make sure my suggestion is correct.
In this case I would just ask "have you already also tried X" which is much faster than pulling their code, implementing your suggestion, and waiting for a build and test to run.
I do too, but this is a conference, I doubt code was provided.
And even then, what you're describing isn't review per se, it's replication. In principle there are entire journals that one can submit replication reports to, which count as actual peer reviewable publications in themselves. So one needs to be pragmatic with what is expected from a peer review (especially given the imbalance between resources invested to create one versus the lack of resources offered and lack of any meaningful reward)
> I do too, but this is a conference, I doubt code was provided.
Machine learning conferences generally encourage (anonymized) submission of code. However, that still doesn't mean that replication is easy. Even if the data is also available, replication of results might require impractical levels of compute power; it's not realistic to ask a peer reviewer to pony up for a cloud account to reproduce even medium-scale results.
If there’s anything I would want to run to verify, I ask the author to add a unit test. Generally, the existing CI test + new tests in the PR having run successfully is enough. I might pull and run it if I am not sure whether a particular edge case is handled.
Reviewers wanting to pull and run many PRs makes me think your automated tests need improvement.
No, because this is usually a waste of time, because CI enforces that the code and the tests can run at submission time. If your CI isn't doing it, you should put some work in to configure it.
If you regularly have to do this, your codebase should probably have more tests. If you don't trust the author, you should ask them to include test cases for whatever it is that you are concerned about.
> This is true, but here the equivalent situation is someone using a greek question mark (";") instead of a semicolon (";"),
No it's not. I think you're trying to make a different point, because you're using an example of a specific deliberate malicious way to hide a token error that prevents compilation, but is visually similar.
> and you as a code reviewer are only expected to review the code visually and are not provided the resources required to compile the code on your local machine to see the compiler fail.
What weird world are you living in where you don't have CI. Also, it's pretty common I'll test code locally when reviewing something more complex, more complex, or more important, if I don't have CI.
> Yes in theory you can go through every semicolon to check if it's not actually a greek question mark; but one assumes good faith and baseline competence such that you as the reviewer would generally not be expected to perform such pedantic checks.
I don't, because it won't compile. Not because I assume good faith. References and citations are similar to introducing dependencies. We're talking about completely fabricated deps. e.g. This engineer went on npm and grabbed the first package that said left-pad but it's actually a crypto miner. We're not talking about a citation missing a page number, or publication year. We're talking about something that's completely incorrect, being represented as relevant.
> So if you think you might have reasonably missed greek question marks in a visual code review, then hopefully you can also appreciate how a paper reviewer might miss a false citation.
I would never miss this, because the important thing is code needs to compile. If it doesn't compile, it doesn't reach the master branch. Peer review of a paper doesn't have CI, I'm aware, but it's also not vulnerable to syntax errors like that. A paper with a fake semicolon isn't meaningfully different, so this analogy doesn't map to the fraud I'm commenting on.
you have completely missed the point of the analogy.
breaking the analogy beyond the point where it is useful by introducing non-generalising specifics is not a useful argument. Otherwise I can counter your more specific non-generalising analogy by introducing little green aliens sabotaging your imaginary CI with the same ease and effect.
I disagree you could do that and claim to be reasonable.
But I agree, because I'd rather discuss the pragmatics and not bicker over the semantics about an analogy.
Introducing a token error, is different from plagiarism, no? Someone wrote code that can't compile, is different from someone "stealing" proprietary code from some company, and contributing it to some FOSS repo?
In order to assume good faith, you also need to assume the author is the origin. But that's clearly not the case. The origin is from somewhere else, and the author that put their name on the paper didn't verify it, and didn't credit it.
Sure but the focus here is on the reviewer not the author.
The point is what is expected as reasonable review before one can "sign their name on it".
"Lazy" (or possibly malicious) authors will always have incentives to cut corners as long as no mechanisms exist to reject (or even penalise) the paper on submission automatically. Which would be the equivalent of a "compiler error" in the code analogy.
Effectively the point is, in the absence of such tools, the reviewer can only reasonably be expected to "look over the paper" for high-level issues; catching such low-level issues via manual checks by reviewers has massively diminishing returns for the extra effort involved.
So I don't think the conference shaming the reviewers here in the absence of providing such tooling is appropriate.
Code correctness should be checked automatically with the CI and testsuite. New tests should be added. This is exactly what makes sure these stupid errors don't bother the reviewer. Same for the code formatting and documentation.
This discussion makes me think peer reviews need more automated tooling somewhat analogous to what software engineers have long relied on. For example, a tool could use an LLM to check that the citation actually substantiates the claim the paper says it does, or else flags the claim for review.
I'd go one further and say all published papers should come with a clear list of "claimed truths", and one is only able to cite said paper if they are linking in to an explicit truth.
Then you can build a true hierarchy of citation dependencies, checked 'statically', and have better indications of impact if a fundamental truth is disproven, ...
Could you provide a proof of concept paper for that sort of thing? Not a toy example, an actual example, derived from messy real-world data, in a non-trivial[1] field?
---
[1] Any field is non-trivial when you get deep enough into it.
hey, i'm a part of the gptzero team that built automated tooling, to get the results in that article!
totally agree with your thinking here, we can't just give this to an LLM, because of the need to have industry-specific standards for what is a hallucination / match, and how to do the search
One could submit their bibtex files and expect bibtex citations to be verifiable using a low level checker.
Worst case scenario if your bibtex citation was a variant of one in the checker database you'd be asked to correct it to match the canonical version.
However, as others here have stated, hallucinated "citations" are actually the lesser problem. Citing irrelevant papers based on a fly-by reference is a much harder problem; this was present even before LLMs, but this has now become far worse with LLMs.
Yes, I think verifying mere existence of the cited paper barely moves the needle. I mean, I guess automated verification of that is a cheap rejection criterion, but I don’t think it’s overall very useful.
this is still in beta because its a much harder problem for sure, since its hard to determine if a 40 page paper supports a claims (if the paper claims X is computationally intractable, does that mean algorithms to compute approximate X are slow?)
That is not, cannot be, and shouldn't be, the bar for peer review. There are two major differences between it and code review:
1. A patch is self-contained and applies to a codebase you have just as much access to as the author. A paper, on the other hand, is just the tip of the iceberg of research work, especially if there is some experiment or data collection involved. The reviewer does not have access to, say, videos of how the data was collected (and even if they did, they don't have the time to review all of that material).
2. The software is also self-contained. That's "prodcution". But a scientific paper does not necessarily aim to represent scientific consensus, but a finding by a particular team of researchers. If a paper's conclusions are wrong, it's expected that it will be refuted by another paper.
> That is not, cannot be, and shouldn't be, the bar for peer review.
Given the repeatability crisis I keep reading about, maybe something should change?
> 2. The software is also self-contained. That's "prodcution". But a scientific paper does not necessarily aim to represent scientific consensus, but a finding by a particular team of researchers. If a paper's conclusions are wrong, it's expected that it will be refuted by another paper.
This is a much, MUCH stronger point. I would have lead with this because the contrast between this assertion, and my comparison to prod is night and day. The rules for prod are different from the rules of scientific consensus. I regret losing sight of that.
> Given the repeatability crisis I keep reading about, maybe something should change?
The replication crisis — assuming that it is actually a crisis — is not really solvable with peer review. If I'm reviewing a psychology paper presenting the results of an experiment, I am not able to re-conduct the entire experiment as presented by the authors, which would require completely changing my lab, recruiting and paying participants, and training students & staff.
Even if I did this, and came to a different result than the original paper, what does it mean? Maybe I did something wrong in the replication, maybe the result is only valid for certain populations, maybe inherent statistical uncertainty means we just get different results.
Again, the replication crisis — such that it exists — is not the result of peer review.
IMHO what should change is we stop putting "peer reviewed" articles on a pedestal.
Even if peer review is as rigorous as code reviewed (the former which is usually unpaid), we all know that reviewed code still has bugs, and a programmer would be nuts to go around saying "this code is reviewed by experts, we can assume it's bug free, right?"
But there are too many people who are just assuming peer reviewed articles means they're somehow automatically correct.
A reviewer is assessing the relevance and "impact" of a paper rather than correctness itself directly. Reviewers may not even have access to the data itself that authors may have used. The way it essentially works is an editor asks the reviewers "is this paper worthy to be published in my journal?" and the reviewers basically have to answer that question. The process is actually the editor/journal's responsibility.
> I've always assumed peer review is similar to diff review. Where I'm willing to sign my name onto the work of others. If I approve a diff/pr and it takes down prod. It's just as much my fault, no?
No.
Modern peer review is “how can I do minimum possible work so I can write ‘ICLR Reviewer 2025’ on my personal website”
The vast majority of people I see do not even mention who they review for in CVs etc. It is usually more akin to a volunteer based, thankless work. Unless you are an editor or sth in a journal, what you review for does not count much for anything.
For ICLR reviewers were asked to review 5 papers in two weeks. Unpaid voluntary work in addition to their normal teaching, supervision, meetings, and other research duties. It's just not possible to understand and thoroughly review each paper even for topic experts. If you want to compare peer review to coding, it's more like "no syntax errors, code still compiles" rather than pr review.
I think the root problem is that everyone involved, from authors to reviewers to publishers, know that 99.999% of papers are completely of no consequence, just empty calories with the sole purpose of padding quotas for all involved, and thus are not going to put in the effort as if.
This is systemic, and unlikely to change anytime soon. There have been remedies proposed (e.g. limits on how many papers an author can publish per year, let's say 4 to be generous), but they are unlikely to gain traction as thoug most would agree onbenefits, all involved in the system would stand to lose short term.
> I don’t consider it the reviewers responsibility to manually verify all citations are real
I guess this explains all those times over the years where I follow a citation from a paper and discover it doesn’t support what the first paper claimed.
As a reviewer I at least skimmed the papers for every reference in every paper that I review. If it isn't useful to furthering the point of the paper then my feedback is to remove the reference. Adding a bunch of junk because it is broadly related in a giant background section is a waste of everyone's time and should be removed. Most of the time you are mostly aware of the papers being cited anyway because that is the whole point of reviewing in your area of expertise.
Agreed. I used to review lots of submissions for IEEE and similar conferences, and didn't consider it my job to verify every reference. No one did, unless the use of the reference triggered an "I can't believe it said that" reaction. Of course, back then, there wasn't a giant plagiarism machine known to fabricate references, so if tools can find fake references easily the tools should be used.
I agree with you (I have reviewed papers in the past), however, made-up citations are a "signal". Why would the authors do that? If they made it up, most likely they haven't really read that prior work. If they haven't, have they really done proper due dilligence on their research? Are they just trying to "beef up" their paper with citations to unfairly build up credibility?
> Surely there are tools to retrieve all the citations,
Even if you could retrieve all citations (which isn't always as easy as you might hope) to validate citations you'd also have to confirm the paper says what the person citing it says. If I say "A GPU requires 1.4kg of copper" citing [1] is that a valid citation?
That means not just reviewing one paper, but also potentially checking 70+ papers it cites. The vast majority of paper reviewers will not check citations actually say what they're claimed to say, unless a truly outlandish claim is made.
At the same time, academia is strangely resistant to putting hyperlinks in citations, preferring to maintain old traditions - like citing conference papers by page number in a hypothetical book that has never been published; and having both a free and a paywalled version of a paper while considering the paywalled version the 'official' version.
Wow. I went to law school and was on the law review. That was our precise job for the papers selected for publication. To verify every single citation.
Thanks for sharing that. Interesting how there was a solution to a problem that didn't really exist yet.. I mean, I'm sure it was there for a reason, but I assume it was more things like wrongful attribution, missing commas etc. rather than outright invented quotes to fit a narrative or do you have more background on that?
...at least the mandatory automated checking processes are probably not far off at least for the more reputable journals, but it still makes you wonder how much you can trust the last two years of LLM-enhanced science that is now being quoted in current publications and if those hallucinations can be "reverted" after having been re-quoted. A bit like Wikipedia can be abused to establish facts.
It is absolutely the reviewers job to check citations. Who else will check and what is the point of peer review then? So you’d just happily pass on shoddy work because it’s not your job? You’re reviewing both the authors work and if there were people to at needed to ensure citations were good, you’re checking their work also. This is very much the problem today with this “not my problem” mindset. If it passes review, the reviewer is also at fault. Not excuses.
The problem is most academics just do not have the time to do this for free, or in fact even if paid. In addition you may not even have access to the references. In acoustics it's not uncommon to cite works that don't even exist online and it's unlikely the reviewer will have the work in their library.
correct me if I'm wrong but citations in papers follow a specific format, and the case here is that a tool was used to validate that they are all real. Certainly a tool that scans a paper for all citations and verifies that they actually exist in the journals they reference shouldn't be all that technically difficult to achieve?
There are a ton of edge cases and a bit of contextual understanding for what is a hallucinated citation (i.e. what if its republished from arxiv to ICLR?)
But to your point, seems we need a tool that can do this
In theory, the review tries to determine if the conclusion reached actually follows from whatever data is provided. It assumes that everything is honest, it's just looking to see if there were mistakes made.
Honest or not should not make a difference, after all, the submitting author may believe themselves everything is A-OK.
The review should also determine how valuable the contribution is, not only if it has mistakes or not.
Todays reviews determine neither value nor correctness in any meaningful way. And how could they, actually? That is why I review papers only to the extent that I understand them, and I clearly delineate my line of understanding. And I don't review papers that I am not interested in reading. I once got a paper to review that actually pointed out a mistake in one of my previous papers, and then proposed a different solution. They correctly identified the mistake, but I could not verify if their solution worked or not, that would have taken me several weeks to understand. I gave a report along these lines, and the person who gave me the review said I should say more about their solution, but I could not. So my review was not actually used. The paper was accepted, which is fine, but I am sure none of the other reviewers actually knows if it is correct.
Now, this was a case where I was an absolute expert. Which is far from the usual situation for a reviewer, even though many reviewers give themselves the highest mark for expertise when they just should not.
A couple had just moved in a house and called me to replace the ceiling fan in the living room.
I pulled the flush mount cover down to start unhooking the wire nuts and noticed RG58 (coax cable).
Someone had used the center conductor as the hot wire!
I ended up running 12/2 Romex from the switch. There was no way in hell I could have hooked it back up the way it was.
This is just one example I've come across.
I am not an electrician, but when I did projects, I did a lot of research before deciding to hire someone and then I was extremely confused when everyone was proposing doing it slightly differently.
A lot of them proposed ways that seem to violate the code, like running flex tubing beyond the allowed length or amount of turns.
Another example would be people not accounting for needing fireproof covers if they’re installing recessed, lighting in between dwelling in certain cities…
Heck, most people don’t actually even get the permit. They just do the unpermitted work.
No doubt the best electricians are currently better than the best AI, but the best AI is likely now better than the novice homeowner. The trajectory over the past 2 years has been very good. Another five years and AI may be better than all but the very best, or most specialized, electricians.
> AI is not the problem, laziness and negligence is
This reminds me about discourse about a gun problem in US, "guns don't kill people, people kill people", etc - it is a discourse used solely for the purpose of not doing anything and not addressing anything about the underlying problem.
No, the OP is right in this case. Did you read TFA? It was "peer reviewed".
> Worryingly, each of these submissions has already been reviewed by 3-5 peer experts, most of whom missed the fake citation(s). This failure suggests that some of these papers might have been accepted by ICLR without any intervention. Some had average ratings of 8/10, meaning they would almost certainly have been published.
If the peer reviewers can't be bothered to do the basics, then there is literally no point to peer review, which is fully independent of the author who uses or doesn't use AI tools.
> it is a discourse used solely for the purpose of not doing anything and not addressing anything about the underlying problem
Solely? Oh brother.
In reality it’s the complete opposite. It exists to highlight the actual source of the problem, as both industries/practitioners using AI professionally and safely, and communities with very high rates of gun ownership and exceptionally low rates of gun violence exist.
It isn’t the tools. It’s the social circumstances of the people with access to the tools. That’s the point. The tools are inanimate. You can use them well or use them badly. The existence of the tools does not make humans act badly.
To continue the carpenter analogy, the issue with LLMs is that the shelf looks great but is structurally unsound. That it looks good on surface inspection makes it harder to tell that the person making it had no idea what they're doing.
Regardless, if a carpenter is not validating their work before selling it, it's the same as if a researcher doesn't validate their citations before publishing. Neither of them have any excuses, and one isn't harder to detect than the other. It's just straight up laziness regardless.
I think this is a bit unfair. The carpenters are (1) living in world where there’s an extreme focus on delivering as quicklyas possible, (2) being presented with a tool which is promised by prominent figures to be amazing, and (3) the tool is given at a low cost due to being subsidized.
And yet, we’re not supposed to criticize the tool or its makers? Clearly there’s more problems in this world than «lazy carpenters»?
Yes, it's the scientists problem to deal with it - that's the choice they made when they decided to use AI for their work. Again, this is what responsibility means.
This inspires me to make horrible products and shift the blame to the end user for the product being horrible in the first place. I can't take any blame for anything because I didn't force them to use it.
No, I merely said that the scientist is the one responsible for the quality of their own work. Any critiques you may have for the tools which they use don't lessen this responsibility.
>No, I merely said that the scientist is the one responsible for the quality of their own work.
No, you expressed unqualified agreement with a comment containing
“And yet, we’re not supposed to criticize the tool or its makers?”
>Any critiques you may have for the tools which they use don't lessen this responsibility.
People don’t exist or act in a vacuum. That a scientist is responsible for the quality of their work doesn’t mean that a spectrometer manufacture that advertises specs that their machines can’t match and induces universities through discounts and/or dubious advertising claims to push their labs to replace their existing spectrometers with new ones which have many bizarre and unexpected behaviors including but not limited to sometimes just fabricating spurious readings has made no contribution to the problem of bad results.
You can criticize the tool or its makers, but not as a means to lessen the responsibility of the professional using it (the rest of the quoted comment). I agree with the GP, it's not a valid excuse for the scientist's poor quality of work.
The scientist has (at the very least) a basic responsibility to perform due diligence. We can argue back and forth over what constitutes appropriate due diligence, but, with regard to the scientist under discussion, I think we'd be better suited discussing what constitutes negligence.
Well, then what does this say of LLM engineers at literally any AI company in existence if they are delivering AI that is unreliable then? Surely, they must take responsibility for the quality of their work and not blame it on something else.
I feel like what "unreliable" means, depends on well you understand LLMs. I use them in my professional work, and they're reliable in terms of I'm always getting tokens back from them, I don't think my local models have failed even once at doing just that. And this is the product that is being sold.
Some people take that to mean that responses from LLMs are (by human standards) "always correct" and "based on knowledge", while this is a misunderstanding about how LLMs work. They don't know "correct" nor do they have "knowledge", they have tokens, that come after tokens, and that's about it.
> they're reliable in terms of I'm always getting tokens back from them
This is not what you are being sold though. They are not selling you "tokens". Check their marketing articles and you will not see the word token or synonym on any of their headings or subheadings. You are being sold these abilities:
- “Generate reports, draft emails, summarize meetings, and complete projects.”
- “Automate repetitive tasks, like converting screenshots or dashboards into presentations … rearranging meetings … updating spreadsheets with new financial data while retaining the same formatting.”
- "Support-type automation: e.g. customer support agents that can summarize incoming messages, detect sentiment, route tickets to the right team."
- "For enterprise workflows: via Gemini Enterprise — allowing firms to connect internal data sources (e.g. CRM, BI, SharePoint, Salesforce, SAP) and build custom AI agents that can: answer complex questions, carry out tasks, iterate deliverables — effectively automating internal processes."
These are taken straight from their websites. The idea that you are JUST being sold tokens is as hilariously fictional as any company selling you their app was actually just selling you patterns of pixels on your screen.
it’s not “some people”, it’s practically everyone that doesn’t understand how these tools work, and even some people that do.
Lawyers are running their careers by citing hallucinated cases. Researchers are writing papers with hallucinated references. Programmers are taking down production by not verifying AI code.
Humans were made to do things, not to verify things. Verifying something is 10x harder than doing it right. AI in the hands of humans is a foot rocket launcher.
> it’s not “some people”, it’s practically everyone that doesn’t understand how these tools work, and even some people that do.
Again, true for most things. A lot of people are terrible drivers, terrible judge of their own character, and terrible recreational drug users. Does that mean we need to remove all those things that can be misused?
I much rather push back on shoddy work no matter what source. I don't care if the citations are from a robot or a human, if they suck, then you suck, because you're presenting this as your work. I don't care if your paralegal actually wrote the document, be responsible for the work you supposedly do.
> Humans were made to do things, not to verify things.
I'm glad you seemingly have some grand idea of what humans were meant to do, I certainly wouldn't claim I do so, but I'm also not religious. For me, humans do what humans do, and while we didn't used to mostly sit down and consume so much food and other things, now we do.
>A lot of people are terrible drivers, terrible judge of their own character, and terrible recreational drug users. Does that mean we need to remove all those things that can be misused?
Uhh, yes??? We have completely reshaped our cities so that cars can thrive in them at the expense of people. We have laws and exams and enforcement all to prevent cars from being driven by irresponsible people.
And most drugs are literally illegal! The ones that arent are highly regulated!
If your argument is that AI is like heroin then I agree, let’s ban it and arrest anyone making it.
People need to be responsible for things they put their name on. End of story. No AI company claims their models are perfect and don’t hallucinate. But paper authors should at least verify every single character their submit.
I use those LLM "deep research" modes every now and then. They can be useful for some use cases. I'd never think to freaking paste it into a paper and submit it or publish it without checking; that boggles the mind.
The problem is that a researcher who does that is almost guaranteed to be careless about other things too. So the problem isn't just the LLM, or even the citations, but the ambient level of acceptable mediocrity.
> And yet, we’re not supposed to criticize the tool or its makers?
Exactly, they're not forcing anyone to use these things, but sometimes others (their managers/bosses) forced them to. Yet it's their responsibility for choosing the right tool for the right problem, like any other professional.
If a carpenter shows up to put a roof yet their hammer or nail-gun can't actually put in nails, who'd you blame; the tool, the toolmaker or the carpenter?
> If a carpenter shows up to put a roof yet their hammer or nail-gun can't actually put in nails, who'd you blame; the tool, the toolmaker or the carpenter?
I would be unhappy with the carpenter, yes. But if the toolmaker was constantly over-promising (lying?), lobbying with governments, pushing their tools into the hands of carpenters, never taking responsibility, then I would also criticize the toolmaker. It’s also a toolmaker’s responsibility to be honest about what the tool should be used for.
I think it’s a bit too simplistic to say «AI is not the problem» with the current state of the industry.
If I hired a carpenter, he did a bad job, and he starts to blame the toolmaker because they lobby the government and over-promised what that hammer could do, I'd still put the blame on the carpenter. It's his tools, I couldn't give less of a damn why he got them, I trust him to be a professional, and if he falls for some scam or over-promised hammers, that means he did a bad job.
Just like as a software developer, you cannot blame Amazon because your platform is down, if you chose to host all of your platform there. You made that choice, you stand for the consequences, pushing the blame on the ones who are providing you with the tooling is the action of someone weak who fail to realize their own responsibilities. Professionals take responsibility for every choice they make, not just the good ones.
> I think it’s a bit too simplistic to say «AI is not the problem» with the current state of the industry.
Agree, and I wouldn't say anything like that either, which makes it a bit strange to include a reply to something no one in this comment thread seems to have said.
> When you use our Services you understand and agree:
Output may not always be accurate. You should not rely on Output from our Services as a sole source of truth or factual information, or as a substitute for professional advice.
You must evaluate Output for accuracy and appropriateness for your use case, including using human review as appropriate, before using or sharing Output from the Services.
You must not use any Output relating to a person for any purpose that could have a legal or material impact on that person, such as making credit, educational, employment, housing, insurance, legal, medical, or other important decisions about them.
Our Services may provide incomplete, incorrect, or offensive Output that does not represent OpenAI’s views. If Output references any third party products or services, it doesn’t mean the third party endorses or is affiliated with OpenAI.
Anthropic:
> When using our products or services to provide advice, recommendations, or in subjective decision-making directly affecting individuals or consumers, a qualified professional in that field must review the content or decision prior to dissemination or finalization. You or your organization are responsible for the accuracy and appropriateness of that information.
So I don't think we can say they are lying.
A poor workman blames his tools. So please take responsibility for what you deliver. And if the result is bad, you can learn from it. That doesn't have to mean not use AI but it definitely means that you need to fact check more thoroughly.
Yeah seriously. Using an LLM to help find papers is fine. Then you read them. Then you use a tool like Zotero or manually add citations.
I use Gemini Pro to identify useful papers that I might not yet have encountered before. But, even when asking to restrict itself to Pubmed resources, it's citations are wonky, citing three different version sources of the same paper (citations that don't say what they said they'd discuss).
That said, these tools have substantially reduced hallucinations over the last year, and will just get better. It also helps if you can restrict it to reference already screened papers.
Finally, I'd lke to say tthat if we want scientists to engage in good science, stop forcing them to spend a third of their time in a rat race for funding...it is ridiculously time consuming and wasteful of expertise.
The problem isn't whether they have more or less hallucinations. The problem is that they have them. And as long as they hallucinate, you have to deal with that. It doesn't really matter how you prompt, you can't prevent hallucinations from happening and without manual checking, eventually hallucinations will slip under the radar because the only difference between a real pattern and a hallucinated one is that one exists in the world and the other one doesn't. This is not something you can really counter with more LLMs either as it is a problem intrinsic to LLMs
> If a carpenter builds a crappy shelf “because” his power tools are not calibrated correctly - that’s a crappy carpenter, not a crappy tool.
It's both. The tool is crappy, and the carpenter is crappy for blindly trusting it.
> AI is not the problem, laziness and negligence is.
Similarly, both are a problem here. LLMs are a bad tool, and we should hold people responsible when they blindly trust this bad tool and get bad results.
"Anyone, from the most clueless amateur to the best cryptographer, can create an algorithm that he himself can’t break."--Bruce Schneier
There's a corollary here with LLMs, but I'm not pithy enough to phrase it well. Anyone can create something using LLMs that they, themselves, aren't skilled enough to spot the LLMs' hallucinations. Or something.
LLMs are incredibly good at exploiting peoples' confirmation biases. If it "thinks" it knows what you believe/want, it will tell you what you believe/want. There does not exist a way to interface with LLMs that will not ultimately end in the LLM telling you exactly what you want to hear. Using an LLM in your process necessarily results in being told that you're right, even when you're wrong. Using an LLM necessarily results in it reinforcing all of your prior beliefs, regardless of whether those prior beliefs are correct. To an LLM, all hypotheses are true, it's just a matter of hallucinating enough evidence to satisfy the users' skepticism.
I do not believe there exists a way to safely use LLMs in scientific processes. Period. If my belief is true, and ChatGPT has told me it's true, then yes, AI, the tool, is the problem, not the human using the tool.
I find this to be a bit “easy”. There is such a thing as bad tools. If it is difficult to determine if the tool is good or bad i’d say some of the blame has to be put on the tool.
At the very least, authors who have been caught publishing proven fabrications should be barred by those journals from ever publishing in them again. Mind you, this is regardless of whether or not an LLM was involved.
> authors who have been caught publishing proven fabrications should be barred by those journals from ever publishing in them again
This is too harsh.
Instead, their papers should be required to disclose the transgression for a period of time, and their institution should have to disclose it publicly as well as to the government, students and donors whenever they ask them for money.
I’m not advocating, I’m making a high-level observation: Industry forever pushes for nil regulation and blames bad actors for damaging use.
But we always have some regulation in the end. Even if certain firearms are legal to own, howitzers are not — although it still takes a “bad actor” to rain down death on City Hall.
The same dynamic is at play with LLMs: “Don’t regulate us, punish bad actors! If you still have a problem, punish them harder!” Well yes, we will punish bad actors, but we will also go through a negotiation of how heavily to constrain the use of your technology.
the person you originally responded to isn’t against regulation per their comment. I’m not against regulation. what’s the pitch for regulation of LLMs?
If the blame were solely on the user then we'd see similar rates of deaths from gun violence in the US vs. other countries. But we don't, because users are influenced by the UX
Somehow people don't kill people nearly as easily, or with as high of a frequency or social support, in places that don't have guns that are more accessible than healthcare. So weird.
> AI is not the problem, laziness and negligence is.
As much as I agree with you that this is wrong, there is a danger in putting the onus just on the human. Whether due to competition or top down expectations, humans are and will be pressured to use AI tools alongside their work and produce more. Whereas the original idea was for AI to assist the human, as the expected velocity and consumption pressure increases humans are more and more turning into a mere accountability laundering scheme for machine output. When we blame just the human, we are doing exactly what this scheme wants us to do.
Therefore we must also criticize all the systemic factors that puts pressure on reversal of AI‘s assistance into AI’s domination of human activity.
So AI (not as a technology but as a product when shoved down the throats) is the problem.
Absolutely, expectations and tools given by management are a real problem.
If management fires you because they are wrong about how good AI is, and you're right - at the end of the day, you're fired and the manager is in lalaland.
People need to actually push the correct calibration of what these tools should be trusted to do, while also trying to work with what they have.
Trades also have self regulation. You can’t sell plumbing services or build houses without any experience or you get in legal trouble. If your workmanship is poor, you can be disciplined by the board even if the tool was at fault. I think fraudulent publications should be taken at least as seriously as badly installed toilets.
The obvious solution in this scenario is.. to just buy a different hammer.
And in the case of AI, either review its output, or simply don't use it. No one has a gun to your head forcing you to use this product (and poorly at that).
It's quite telling that, even in this basic hypothetical, your first instinct is to gesture vaguely in the direction of governmental action, rather than expect any agency at the level of the individual.
No, because this would cost tens of jobs and affect someone's profits, which are sacrosanct. Obviously the market wants exploding hammers, or else people wouldn't buy them. I am very smart.
If a scientist just completely "made up" their references 10 years ago, that's a fraudster. Not just dishonesty but outright academic fraud.
If a scientist does it now, they just blame it on AI. But the consequences should remain the same. This is not an honest mistake.
People that do this - even once - should be banned for life. They put their name on the thing. But just like with plagiarism, falsifying data and academic cheating, somehow a large subset of people thinks it's okay to cheat and lie, and another subset gives them chance after chance to misbehave like they're some kind of children. But these are adults and anyone doing this simply lacks morals and will never improve.
And yes, I've published in academia and I've never cheated or plagiarized in my life. That should not be a drawback.
Three and a half years ago nobody had ever used tools like this. It can't be a legitimate complaint for an author to say, "not my fault my citations are fake it's the fault of these tools" because until recently no such tools were available and the expectation was that all citations are real.
If my calculator gives me the wrong number 20% of the time yeah I should’ve identified the problem, but ideally, that wouldn’t have been sold to me as a functioning calculator in the first place.
If it was a well understood property of calculators that they gave incorrect answers randomly then you need to adjust the way you use the tool accordingly.
Sorry, Utkar the manager will fire you if you don’t use his shitty calculator. If you take the time to check the output every time you’ll be fired for being too slow. Better pray the calculator doesn’t lie to you.
Generally I’d ditch that tool because it doesn’t work. A calculator is supposed to calculate. If it can’t reliably calculate, then it’s not a functioning tool and I am tired of people insisting it is functioning properly.
LLM’s simply aren’t good enough for all the use cases some people insist they are. They’re powerful tools that have been far too broadly applied and there’s too much money and too many reputations being put on the line to acknowledge the obvious limitations. Frankly I’m sick of it.
I had somebody on HN a few months ago insist to me that because we value art and fiction, LLM’s being wrong when we need them to be correct (in ways that are also not always easy to identify) was desirable. I don’t even know what to do with that kind of logic other than chalk it up as trolling. I don’t want my computer to trick me into false solutions.
Indeed. The narrative that this type of issue is entirely the responsibility of the user to fix is insulting, and blame deflection 101.
It's not like these are new issues. They're the same ones we've experienced since the introduction of these tools. And yet the focus has always been to throw more data and compute at the problem, and optimize for fancy benchmarks, instead of addressing these fundamental problems. Worse still, whenever they're brought up users are blamed for "holding it wrong", or for misunderstanding how the tools work. I don't care. An "artificial intelligence" shouldn't be plagued by these issues.
Exactly, that's why not verifying the output is even less defensible now than it ever has been - especially for professional scientists who are responsible for the quality of their own work.
> Worse still, whenever they're brought up users are blamed for "holding it wrong", or for misunderstanding how the tools work. I don't care. An "artificial intelligence" shouldn't be plagued by these issues.
My feelings exactly, but you’re articulating it better than I typically do ha
I disagree. When the tool promises to do something, you end up trusting it to do the thing.
When Tesla says their car is self driving, people trust them to self drive. Yes, you can blame the user for believing, but that's exactly what they were promised.
> Why didn't the lawyer who used ChatGPT to draft legal briefs verify the case citations before presenting them to a judge? Why are developers raising issues on projects like cURL using LLMs, but not verifying the generated code before pushing a Pull Request? Why are students using AI to write their essays, yet submitting the result without a single read-through? They are all using LLMs as their time-saving strategy. [0]
It's not laziness, its the feature we were promised. We can't keep saying everyone is holding it wrong.
Very well put. You're promised Artificial Super Intelligence and shown a super cherry-picked promo and instead get an agent that can't hold its drool and needs constant hand-holding... it can't be both things at the same time, so... which is it?
Modern science is designed from the top to the bottom to produce bad results. The incentives are all mucked up. It's absolutely not surprising that AI is quickly becoming yet-another factor lowering quality.
That's like saying guns aren't the problem, the desire to shoot is the problem. Okay, sure, but wanting something like a metal detector requires us to focus on the more tangible aspect that is the gun.
If I gave you a gun without a safety could you be the one to blame when it goes off because you weren’t careful enough?
The problem with this analogy is that it makes no sense.
LLMs aren’t guns.
The problem with using them is that humans have to review the content for accuracy. And that gets tiresome because the whole point is that the LLM saves you time and effort doing it yourself. So naturally people will tend to stop checking and assume the output is correct, “because the LLM is so good.”
Then you get false citations and bogus claims everywhere.
> The problem with using them is that humans have to review the content for accuracy.
There are (at least) two humans in this equation. The publisher, and the reader. The publisher at least should do their due diligence, regardless of how "hard" it is (in this case, we literally just ask that you review your OWN CITATIONS that you insert into your paper). This is why we have accountability as a concept.
> If I gave you a gun without a safety could you be the one to blame when it goes off because you weren’t careful enough?
Absolutely. Many guns don't have safties. You don't load a round in the chamber unless you intend on using it.
A gun going off when you don't intend is a negligent discharge. No ifs, ands or buts. The person in possession of the gun is always responsible for it.
> A gun going off when you don't intend is a negligent discharg
false. A gun goes off when not intended too often to claim that. It has happned to me - I then took the gun to a qualified gunsmith for repairs.
A gun they fires and hits anything you didn't intend to is negligent discharge even if you intended to shoot. Gun saftey is about assuming a gun that could possible fire will and ensuring nothing bad can happen. When looking at gun in a store (that you might want to buy) you aim it at an upper corner where even if it fires the odds of something bad resulting is the least lively to happen (it should be unloaded - and you may have checked, but you still aim there!)
same with cat toy lazers - they should be safe to shine in an eye - but you still point in a safe direction.
Yes. That is absolutely the case. One of the
Most popular handguns does not have a safety switch that must be toggled before firing. (Glock series handguns)
If someone performs a negligent discharge, they are responsible, not Glock. It does have other safety mechanisms to prevent accidental fires not resulting from a trigger pull.
> The problem with using them is that humans have to review the content for accuracy.
How long are we going to push this same narrative we've been hearing since the introduction of these tools? When can we trust these tools to be accurate? For technology that is marketed as having superhuman intelligence, it sure seems dumb that it has to be fact-checked by less-intelligent humans.
That doesn't address my point at all but no, I'm not a violent or murderous person. And most people aren't. Many more people do, however, want to take shortcuts to get their work done with the least amount of effort possible.
That's not as random as letting me choose them! They had to be allowed onto the range, show ID, afford the gun, probably do a background check to get the gun unless they used a loophole (which usually requires some social capital).
I'm proposing the true proposal of many guns rights advocates: anyone might have a gun.
So let me choose the 50 and you give them guns! Why not?
The issue with this argument, for anyone who comes after, is not when you give a gun to a SINGLE person, and then ask them "would you do a bad thing".
The issue is when you give EVERYONE guns, and then are surprised when enough people do bad things with them, to create externalities for everyone else.
There is some sort of trip up when personal responsibility, and society wide behaviors, intersect. Sure most people will be reasonable, but the issue is often the cost of the number of irresponsible or outright bad actors.
Scientists who use LLMs to write a paper are crappy scientists indeed. They need to be held accountable, even ostracised by the scientific community. But something is missing from the picture. Why is it that they came up with this idea in the first place? Who could have been peddling the impression (not an outright lie - they are very careful) about LLMs being these almost sentient systems with emergent intelligence, alleviating all of your problems, blah blah blah. Where is the god damn cure for cancer the LLMs were supposed to invent? Who else is it that we need to keep accountable, scrutinised and ostracised for the ever-increasing mountains of AI-crap that is flooding not just the Internet content but now also penetrating into science, every day work, daily lives, conversations, etc. If someone released a tool that enabled and encouraged people to commit suicide in multiple instances that we know of by now, and we know since the infamous "plandemic" facebook trend that the tech bros are more than happy to tolerate worsening societal conditions in the name of their platform growth, who else do we need to keep accountable, scrutinise and ostracise as a society, I wonder?
...No, it was not meant as a hyperbole, as we were literally being told that these models will be able to do all of our work. I won't settle for the bullshit incremental wins here and there we see occassionally - I attribute those essentially to the old 'infinite number of monkeys typing on the infinite number of typewriters producing "Crime and Peace". No. that's not it - we were promised a god damn revolution, no less. Again, where is the cure for cancer and post-scarcity society ? Where is the AGI we were promised for the 2025? Let's hold the ghouls promising all that accountable for a change.
What an absurd set of equivalences to make regarding a scientist's relationship to their own work.
If an engineer provided this line of excuse to me, I wouldn't let them anywhere near a product again - a complete abdication of personal and professional responsibility.
We are, in fact, not tacitly but openly endorsing this, due to this AI everywhere madness. I am so looking forward to when some genius in some banks starts to use it to simplify code and suddenly I have 100000000 € on my bank account. :)
Yeah, I can't imagine not being familiar with every single reference in the bibliography of a technical publication with one's name on it. It's almost as bad as those PIs who rely on lab techs and postdocs to generate research data using equipment that they don't understand the workings of - but then, I've seen that kind of thing repeatedly in research academia, along with actual fabrication of data in the name of getting another paper out the door, another PhD granted, etc.
Unfortunately, a large fraction of academic fraud has historically been detected by sloppy data duplication, and with LLMs and similar image generation tools, data fabrication has never been easier to do or harder to detect.
Absolutely correct. The real issue is that these people can avoid punishment. If you do not care enough about your paper to even verify the existence of citations, then you obviously should not have a job as a scientist.
Taking an academic who does something like that seriously, seem impossible. At best he is someone who is neglecting his most basic duties as an academic, at worst he is just a fraudster. In both cases he should be shunned and excluded.
Have you ever followed citations before? In my experience, they don't support what is being citated, saying the opposite or not even related. It's probably only 60%-ish that actually cite something relevant.
Whether the information in the paper can be trusted is an entirely separate concern.
Old Chinese mathematics texts are difficult to date because they often purport to be older than they are. But the contents are unaffected by this. There is a history-of-math problem, but there's no math problem.
Problem is that most ML papers today are not independently verifiable proofs - in most, you have to trust the scientist didn't fraudulently produce their results.
There is so much BS being submitted to conferences and decreasing the amount of BS they see would result in less skimpy reviews and also less apathy
You are totally correct that hallucinated citations do not invalidate the paper. The paper sans citations might be great too (I mean the LLM could generate great stuff, it's possible).
But the author(s) of the paper is almost by definition a bad scientist (or whatever field they are in). When a researcher writes a paper for publication, if they're not expected to write the thing themselves, at least they should be responsible for checking the accuracy of the contents, and citations are part of the paper...
Not really true nowadays. Stuff in whitepapers needs to be verifiable which is kinda difficult with hallucinations.
Whether the students directly used LLMs or just read content online that was produced with them and cited after just shows how difficult these things made gathering information that's verifiable.
Last month, I was listening to the Joe Rogan Experience episode with guest Avi Loeb, who is a theoretical physicist and professor at Harvard University. He complained about the disturbingly increasing rate at which his students are submitting academic papers referencing non-existent scientific literature that were so clearly hallucinated by Large Language Models (LLMs). They never even bothered to confirm their references and took the AI's output as gospel.
Isn't this an underlying symptom of lack of accountability of our greater leadership? They do these things, they act like criminals and thieves, and so the people who follow them get shown examples that it's OK while being told to do otherwise.
"Show bad examples then hit you on the wrist for following my behavior" is like bad parenting.
Is the baseline assumption of this work that an erroneous citation is LLM hallucinated?
Did they run the checker across a body of papers before LLMs were available and verify that there were no citations in peer reviewed papers that got authors or titles wrong?
They explain in the article what they consider a proper citation, an erroneous one and an hallucination, in the section "Defining Hallucitations". They also say than they have many false positives, mostly real papers who are not available online.
Thad said, i am also very curious of the result than their tool, would give to papers from the 2010's and before.
If you look at their examples in the "Defining Hallucitations" section, I'd say those could be 100% human errors. Shortening authors' names, leaving out authors, misattributing authors, misspelling or misremembering the paper title (or having an old preprint-title, as titles do change) are all things that I would fully expect to happen to anyone in any field were things get ever got published. Modern tools have made the citation process more comfortable, but if you go back to the old days, you'd probably find those kinds of errors everywhere. If you look at the full list of "hallucinations" they claim to have discovered, the only ones I'd not immediately blame on human screwups are the ones where a title and the authors got zero matches for existing papers/people. If you really want to do this kind of analysis correctly, you'd have to match the claim of the text and verify it with the cited article. Because I think it would be even more dangerous if you can get claims accepted by simply quoting an existing paper correctly, while completely ignoring its content (which would have worked here).
> Modern tools have made the citation process more comfortable,
That also makes some of those errors easier. A bad auto-import of paper metadata can silently screw up some of the publication details, and replacing an early preprint with the peer-reviewed article of record takes annoying manual intervention.
I mean, if you’re able to take the citation, find the cited work, and definitively state ‘looks like they got the title wrong’ or ‘they attributed the paper to the wrong authors’, that doesn’t sound like what people usually mean when they say a ‘hallucinated’ citation. Work that is lazily or poorly cited but nonetheless attempts to cite real work is not the problem. Work which gives itself false authority by claiming to cite works that simply do not exist is the main concern surely?
>Work which gives itself false authority by claiming to cite works that simply do not exist is the main concern surely?
You'd think so, but apparently it isn't for these folks. On the other hand, saying "we've found 50 hallucinations in scientific papers" generates a lot more clicks than "we've found 50 common citation mistakes that people make all the time"
Let me second this: a baseline analysis should include papers that were published or reviewed at least 3-4 years ago.
When I was in grad school, I kept a fairly large .bib file that almost certainly had a mistake or two in it. I don’t think any of them ever made it to print, but it’s hard to be 100% sure.
For most journals, they actually partially check your citations as part of the final editing. The citation record is important for journals, and linking with DOIs is fairly common.
the papers themselves are publicly available online too. Most of the ones I spot-checked give the extremely strong impression of AI generation.
not just some hallucinated citations, and not just the writing. in many cases the actual purported research "ideas" seem to be plausible nonsense.
To get a feel for it, you can take some of the topics they write about and ask your favorite LLM to generate a paper. Maybe even throw "Deep Research" mode at it. Perhaps tell it to put it in ICLR latex format. It will look a lot like these.
People will commonly hold LLMs as unusable because they make mistakes. So do people. Books have errors. Papers have errors. People have flawed knowledge, often degraded through a conceptual game of telephone.
Exactly as you said, do precisely this to pre-LLM works. There will be an enormous number of errors with utter certainty.
People keep imperfect notes. People are lazy. People sometimes even fabricate. None of this needed LLMs to happen.
A pre LLM paper with fabricated citations would demonstrate will to cheat by the author.
A post LLM paper with fabricated citations: same thing and if the authors attempt to defend themselves with something like, we trusted the AI, they are sloppy, probably cheaters and not very good at it.
Further, if I use AI-written citations to back some claim or fact, what are the actual claims or facts based on? These started happening in law because someone writes the text and then wishes there was a source that was relevant and actually supportive of their claim. But if someone puts in the labor to check your real/extant sources, there's nothing backing it (e.g. MAHA report).
Interesting that you hallucinated the word "fabricated" here where I broadly talked about errors. Humans, right? Can't trust them.
Firstly, just about every paper ever written in the history of papers has errors in it. Some small, some big. Most accidental, but some intentional. Sometimes people are sloppy keeping notes, transcribe a row, get a name wrong, do an offset by 1. Sometimes they just entirely make up data or findings. This is not remotely new. It has happened as long as we've had papers. Find an old, pre-LLM paper and go through the citations -- especially for a tosser target like this where there are tens of thousands of low effort papers submitted -- and you're going to find a lot of sloppy citations that are hard to rationalize.
Secondly, the "hallucination" is that this particular snake-oil firm couldn't find given papers in many cases (they aren't foolish enough to think that means they were fabricated. But again, they're looking to sell a tool to rubes, so the conclusion is good enough), and in others that some of the author names are wrong. Eh.
Under what circumstances would a human mistakenly cite a paper which does not exist? I’m having difficulty imagining how someone could mistakenly do that.
The issue here is that many of the ‘hallucinations’ this article cites aren’t ’papers which do not exist’. They are incorrect author attributions, publication dates, or titles.
LLM are a force multiplier of this kind of errors though. It's not easy to hallucinate papers out of whole cloth, but LLMs can easily and confidently do it, quote paragraphs that don't exist, and do it tirelessly and at a pace unmatched by humans.
Humans can do all of the above but it costs them more, and they do it more slowly. LLMs generate spam at a much faster rate.
>It's not easy to hallucinate papers out of whole cloth, but LLMs can easily and confidently do it, quote paragraphs that don't exist, and do it tirelessly and at a pace unmatched by humans.
But no one is claiming these papers were hallucinated whole, so I don't see how that's relevant. This study -- notably to sell an "AI detector", which is largely a laughable snake-oil field -- looked purely at the accuracy of citations[1] among a very large set of citations. Errors in papers are not remotely uncommon, and finding some errors is...exactly what one would expect. As the GP said, do the same study on pre-LLM papers and you'll find an enormous number of incorrect if not fabricated citations. Peer review has always been an illusion of auditing.
1 - Which is such a weird thing to sell an "AI detection" tool. Clearly it was mostly manual given that they somehow only managed to check a tiny subset of the papers, so in all likelihood was some guy going through citations and checking them on Google Search.
I think we should see a chart as % of “fabricated” references from past 20 years. We should see a huge increase after 2020-2021. Anyone has this chart data?
Quoting myself from just last night because this comes up every time and doesn't always need a new write-up.
> You also don't need gunpowder to kill someone with projectiles, but gunpowder changed things in important ways. All I ever see are the most specious knee-jerk defenses of AI that immediately fall apart.
Doesn't seem especially out of the norm for a large conference. Call it 10,000 attendees which is large but not huge. Sure; not everyone attending puts in a session proposal. But others put multiple. And many submit but, if not accepted don't attend.
Can't quote exact numbers but when I was on the conference committee for a maybe high four figures attendance conference, we certainly had many thousands of submissions.
The problem isn't only papers it's that the world of academic computer science coalesced around conference submissions instead of journal submissions. This isn't new and was an issue 30 years ago when I was in grad school. It makes the work of conference organizes the little block holding up the entire system.
I recommend actually clicking through and reading some of these papers.
Most of those I spot checked do not give an impression of high quality. Not just AI writing assistance but many seem to have AI-generated "ideas", often plausible nonsense. the reviewers often catch the errors and sometimes even the fake citations.
can I prove malfeasance beyond a reasonable doubt? no. but I personally feel quite confident many of the papers I checked are primarily AI-generated.
I feel really bad for any authors who submitted legitimate work but made an innocent mistake in their .bib and ended up on the same list as the rest of this stuff.
To me such an interpretation suggests there are likely to be papers that were not so easy to spot, perhaps because the AI accidentally happened upon more plausible nonsense and then generated fully non-sense data, which was believable but still (at a reduced level of criticality) nonsense data, to bolster said non-sense theory at a level that is less easy to catch.
As many pointed out, the purpose of peer review is not linting, but the assessment of the novelty and subtle omissions.
Which incentives can be set to discourage the negligence?
How about bounties? A bounty fund set up by the publisher and each submission must come with a contribution to the fund. Then there be bounties for gross negligence that could attract bounty hunters.
How about a wall of shame? Once negligence crosses a certain threshold, the name of the researcher and the paper would be put on a wall of shame for everyone to search and see?
For the kinds of omissions described here, maybe the journal could do an automated citation check when the paper is submitted and bounce back any paper that has a problem with a day or two lag. This would be incentive for submitters to do their own lint check.
True if the citation has only a small typo or two. But if it is unrecognizable or even irrelevant, this is clearly bad (fraudulent?) research -- each citation has be read and understood by the researcher and put in there only if absolutely necessary to support the paper.
There must be price to pay for wasting other people's time (lives?).
It astonishes me that there would be so many cases of things like wrong authors. I began using a citation manager that extracted metadata automatically (zotero in my case) more than 15 years ago, and can’t imagine writing an academic paper without it or a similar tool.
How are the authors even submitting citations? Surely they could be required to send a .bib or similar file? It’s so easy to then quality control at least to verify that citations exist by looking up DOIs or similar.
I know it wouldn’t solve the human problem of relying on LLMs but I’m shocked we don’t even have this level of scrutiny.
Maybe you haven’t carefully checked yet the correctness of automatic tools or of the associated metadata. Zotero is certainly not bug free. Even authors themselves have miss-cited their own past work on occasion, and author lists have had errors that get revised upon resubmission or corrected in errata after publication. The DOI is indeed great, and if it is correct, I can still use the citation as a reader, but the (often abbreviated) lists of authors often have typos. In this case the error rate is not particularly high compared to random early review-level submissions I’ve seen many decades ago. Tools helped increase the number of citations and reduce the error per citation but not sure if they reduced the papers that have at least one error.
Someone commented here that hallucination is what LLMs do, it’s the designed mode of selecting statistically relevant model data that was built on the training set and then mashing it up for an output. The outcome is something that statistically resembles a real citation.
Creating a real citation is totally doable by a machine though, it is just selecting relevant text, looking up the title, authors, pages etc and putting that in canonical form. It’s just that LLMs are not currently doing the work we ask for, but instead something similar in form that may be good enough.
To me, this is exactly what LLMs are good for. It would be exhausting double checking for valid citations in a research paper. Fuzzy comparison and rote lookup seem primed for usage with LLMs.
Writing academic papers is exactly the _wrong_ usage for LLMs. So here we have a clear cut case for their usage and a clear cut case for their avoidance.
Because the risk is lower. They will give you suspicious citations and you can manually check those for false positives. If some false citation pass, it was still a net gain.
Shouldn’t need an llm to check. It’s just a list of authors. I wouldn’t trust an llm on this, and even if they were perfect that’s a lot of resource use just to do something traditional code could do.
Exactly, and there's nothing wrong with using LLMs in this same way as part of the writing process to locate sources (that you verify), do editing (that you check), etc. It's just peak stupidity and laziness to ask it to do the whole thing.
This is as much a failing of "peer review" as anything. Importantly, it is an intrinsic failure, which won't go away even if LLMs were to go away completely.
Peer review doesn't catch errors.
Acting as if it does, and thus assuming the fact of publication (and where it was published) are indicators of veracity is simply unfounded. We need to go back to the food fight system where everyone publishes whatever they want, their colleagues and other adversaries try their best to shred them, and the winners are the ones that stand up to the maelstrom. It's messy, but it forces critics to put forth their arguments rather than quietly gatekeeping, passing what they approve of, suppressing what they don't.
Peer review definitely does catch errors when performed by qualified individuals. I've personally flagged papers for major revisions or rejection as a result of errors in approach or misrepresentation of source material. I have peers who say they have done similar.
I should have said "Peer review doesn't catch _all_ errors" or perhaps "Peer review doesn't eliminate errors".
In other words, being "peer reviewed" is nowhere close to "error free," and if (as is often the case) the rate of errors is significantly greater than the rate at which errors are caught, peer review may not even significantly improve the quality.
I don’t think many researchers take peer review alone as a strong signal, unless it is a venue known for having serious reviewing (e.g. in CS theory, STOC and FOCS have a very high bar). But it acts as a basic filter that gets rid of obvious nonsense, which on its own is valuable. No doubt there are huge issues, but I know my papers would be worse off without reviewer feedback
Peer review was never supposed to check every single detail and every single citation. They are not proof readers. They are not even really supposed to agree or disagree with your results. They should check the soundness of a method, general structure of a paper, that sort of thing. They do catch some errors, but the expectation is not to do another independent study or something.
Passed peer review is the first basic bar that has to be cleared. It was never supposed to be all there is to the science.
It would be crazy to expect them to verify every author is correct on a citation and to cross verify everything. There’s tooling that could be built for that and kinda wild isn’t a thing that’s run on paper submission.
One of the reported hallucinations in this work [1], starting with David Rein, says the other authors are entirely made up. They are indeed absent from the original cited paper [2], but a Google search shows some of the same names featured in citations from other papers [3] [4].
Most of the names in these wrong attributions are actual people though, not hallucinations. What is going on? Is this a case of AI-powered citation management creating some weird feedback loop?
How can someone not be aware, at this point, that— sure- use the systems for finding and summarizing research, but for each source, take 2 minutes to find the source and verify?
Really, this isn’t that hard and it’s not at all an obscure requirement or unknown factor.
I think this is much much less “LLMs dumbing things down” and significantly more just a shibboleth for identifying people that were already nearly or actually doing fraudulent research anyway. The ones who we should now go back and look at prior publications as very likely fraudulent as well.
I’ve been working on tools that specifically address this problem, but from the level upstream of citation.
They don’t check whether a citation exists — instead they measure whether the reasoning pathway leading to a citation is stable, coherent, and free of the entropy patterns that typically produce hallucinations.
The idea is simple:
• Bad citations aren’t the root cause.
• They are a late-stage symptom of a broken reasoning trajectory.
• If you detect the break early, the hallucinated citation never appears.
The tools I’ve built (and documented so anyone can use) do three things:
1. Measure interrogative structure — they check whether the questions driving the paper’s logic are well-formed and deterministic.
2. Track entropy drift in the argument itself — not the text output, but the structure of the reasoning.
3. Surface the exact step where the argument becomes inconsistent — which is usually before the fake citation shows up.
These instruments don’t replace peer review, and they don’t make judgments about culture or intent.
They just expose structural instability in real time — the same instability that produces fabricated references.
If anyone here wants to experiment or adapt the approach, everything is published openly with instructions.
It’s not a commercial project — just an attempt to stabilize reasoning in environments where speed and tool-use are outrunning verification.
Code and instrument details are in my CubeGeometryTest repo (the implementation behind ‘A Geometric Instrument for Measuring Interrogative Entropy in Language Systems’).
https://github.com/btisler-DS/CubeGeometryTest
This is still a developing process.
One wonders why this has not been largely fully automated. If we track those citations anyway. Surely we have database of them and most of them are easily matched there. So only outliers need to be checked either as new latest papers or mistakes which should be close enough to something or real fakes.
Maybe there just is no incentive for this type of activity.
For that matter, it could be automated at the source. Let's say I'm an author. I'd gladly run a "linter" on my article that flags references that can't be tracked, and so forth. It would be no different than testing a computer program that I write before giving it to someone.
It seems like the GPT zero team is automating it! Up to very recently, no one sane would cite a paper with correct title but make up random authors- and shortly, this specific signal will be goodhearted away by a “make my malpractice less detectable MCP,” so I can see why this automation is happening exactly now.
We do have these things and they are often wrong. Loads of the examples given look better than things I’ve seen in real databases on this kind of thing and I worked in this area for a decade.
And these are just the citations that any old free tool could have included via Bibtex link from the website?
Not only is that incredibly easy to verify (you could pay a first semester student without any training), it's also a worrying sign on what the paper's authors consider quality. Not even 5 minutes spent to get the citations right!
> Papers that make extensive usage of LLMs and do not disclose this usage will be desk rejected.
This sounds like they're endorsing the game of how much can we get away with, towards the goal of slipping it past the reviewers, and the only penalty is that the bad paper isn't accepted.
How about "Papers suspected of fabrications, plagiarism, ghost writers, or other academic dishonesty, will be reported to academic and professional organizations, as well as the affiliated institutions and sponsors named on the paper"?
1. "Suspected" is just that, suspected, you can't penalize papers based on your gut feel 2. LLM-s are a tool, and there's nothing wrong with using them unless you misuse them
If you are searching for references with plausible sounding titles then you are doing that because you don't want to have to actually read those references. After all if you read them and discover that one or more don't support your contention (or even worse, refutes it) then you would feel worse about what you are doing. So I suspect there would be a tendency to completely ignore such references and never consider if they actually exist.
LLMs should be awesome at finding plausible sounding titles. The crappy researcher just has to remember to check for existence. Perhaps there is a business model here, bogus references as a service, where this check is done automatically.
Given how many errors I have seen in my years as a reviewer from well before the time of AI tools, it would be very surprizing if 99.75% of the ~20,000 submitted papers to didnt have such errors. If the 300 sample they used was truly random, then 50 of 300 sounds about right compared to errors I had seen starting in the 90s when people manually curated bintex entries. It is the author’s and editor’s job, not the reviewer’s, to fix the citations.
In case people missed it there's some additional important context:
- Major AI conference flooded with peer reviews written by AI
https://news.ycombinator.com/item?id=46088236
- "All OpenReview Data Leaks"
https://news.ycombinator.com/item?id=46073488
- "The Day Anonymity Died: Inside the OpenReview / ICLR 2026 Leak"
https://news.ycombinator.com/item?id=46082370
- More about the leak
https://forum.cspaper.org/topic/191/iclr-i-can-locate-reviewer-how-an-api-bug-turned-blind-review-into-a-data-apocalypse
The second one went under the radar, but basically OpenReview left the API open so you didn't need credentials. This meant all reviewers and authors were deanonymized across multiple conferences.
All these links are for ICLR too, which is the #2 ML conference for those that don't know.
And for some important context of the link for this post, note that they only sampled 300 papers and found 50. It looks to be almost exclusively citations but those are probably the easiest things to verify.
And this week CVPR sent out notifications that OpenReview will be down between Dec 6th and Dec 9th. No explanation for why.
So we have reviewers using LLMs, authors using LLMs, and idk the conference systems writing their software with LLMs? Things seem pretty fragile right now...
I think at least this article should highlight one of the problems we have in academia right now (beyond just ML, though it is more egregious there): citation mining. It is pretty standard to have over 50 citations in your 10 page paper these days. You can bet that most of these are not going to be for the critical claims but instead heavily placed in the background section. I looked at a few of the papers and everyone I looked at had their hallucinated citations in background (or background in appendix) sections. So these are "filler" citations, which I think illustrates a problem: citations are being abused. I mean the metric hacking should be pretty obvious if you just look at how many citations ML people have. It's grown exponentially! Do we really need so many citations? I'm all for giving people credit but a hyper-fixation on citation count as our measure of credit just doesn't work. It's far too simple of a metric. Like we might as well measure how good of a coder you are by the number of lines of code you produce[0].
It really seems that academia doesn't scale very well...
Tools like GPTzero are incredibly unreliable. Me and plently of my colleagues often get our writing flagged as 100% AI by these tools, when no AI was used.
It's awful that there are these hallucinated citations, and the researchers who submitted them ought to be ashamed. I also put some of the blame on the boneheaded culture of academic citations.
"Compression has been widely used in columnar databases and has had an increasing importance over time.[1][2][3][4][5][6]"
Ok, literally everyone in the field already knows this. Are citations 1-6 useful? Well, hopefully one of them is an actually useful survey paper, but odds are that 4-5 of them are arbitrarily chosen papers by you or your friends. Good for a little bit of h-index bumping!
So many citations are not an integral part of the paper, but instead randomly sprinkled on to give an air of authority and completeness that isn't deserved.
I actually have a lot of respect for the academic world, probably more than most HN posters, but this particular practice has always struck me as silly. Outside of survey papers (which are extremely under-provided), most papers need many fewer citations than they have, for the specific claims where the paper is relying on prior work or showing an advance over it.
That's only part of the reason that this type of content is used in academic papers. The other part is that you never know what PhD student / postdoc / researcher will be reviewing your paper, which means you are incentivized to be liberal with citations (however tangential) just in case someone is reading your paper, and has the reaction "why didn't they cite this work, of which I had some role in?"
Papers with a fake air of authority of easily dispatched with. What is not so easily dispatched with is the politics of the submission process.
This type of content is fundamentally about emotions (in the reviewer of your paper), and emotions is undeniably a large factor in acceptance / rejection.
Indeed. One can even game review systems by leaving errors in for the reviewers to find so that they feel good about themselves and that they've done their job. The meta-science game is toxic and full of politics and ego-pleasing.
That's what I'm really afraid of – we will be drowning in the AI slop as a society and we'll loose the most important thing that made free and democratic society possible - a trust. People just don't tust anyone and/or anything any more. And the lack of trust, especially in scale, is very expensive.
This is a particular meme that I really don't like. I've used em-dashes routinely for years. Do I need to stop using them because various people assume they're an AI flag?
Generally the law allows people to make mistakes, as long as a reasonable level of care is taken to avoid them (and also you can get away with carelessness if you don't owe any duty of care to the party). The law regarding what level of care is needed to verify genAI output is probably not very well defined, but it definitely isn't going to be strict liability.
The emotionally-driven hate for AI, in a tech-centric forum even, to the extent that so many commenters seem to be off-balance in their rational thinking, is kinda wild to me.
What if anything do you think is wrong with my analogy?
I think what is clearly wrong with your analogy is assuming that AI applies mostly to software and code production. This is actually a minor use-case for AI.
Government and businesses of all types ---doctors, lawyers, airlines, delivery companies, etc. are attempting to apply AI to uses and situations that can't be tested in advance the same way "vibe" code can. And some of the adverse results have already been ruled on in court.
So papers and citations are created with AI, and here they're being reviewed with AI. When they're published they'll be read by AI, and used to write more papers with AI. Pretty soon, humans won't need to be involved at all, in this apparently insufferable and dreary business we call science, that nobody wants to actually do.
The issue is there are incentives for more quantity and not quality in modern science (well more like academia), so people will use tools to pump stuff out. It'll get worse as academic jobs tighten due.
A reference is included in a paper if the paper uses information derived from the reference, or to acknowledges the reference as a prior source. If the reference is fake, then the derived information could very well be fake.
Let's say that I use a formula, and give a reference to where the formula came from, but the reference doesn't exist. Would you trust the formula?
Let's say a computer program calls a subroutine with a certain name from a certain library, but the library doesn't exist.
A person doing good research doesn't need to check their references. Now, they could stand to check the references for typographic errors, but that's a stretch too. Almost every online service for retrieving articles includes a reference for each article that you can just copy and paste.
After an interview with Cory Doctorow I saw recently, I'm going to stop anthropomorphizing these things by calling them "hallucinations". They're computers, so these incidents are just simply Errors.
I'll continue calling them hallucinations. That's a much more fitting term when you account for the reasonableness of people who believe them. There's also equally a huge breadth of different types of errors that don't pattern match well into, "made up bullshit" the same way calling them hallucinations do. There's no need to introduce that ambiguity when discussing something narrow.
there's nothing wrong with anthropomorphizing genai, it's source material is human sourced, and humans are going to use human like pattern matching when interacting with it. I.e. This isn't the river I want to swim upstream in. I assume you wouldn't complain if someone anthropomorphized a rock... up until they started to believe it was actually alive.
They're a very specific kind of error, just like off-by-one errors, or I/O errors, or network errors. The name for this kind of error is a hallucination.
We need a word for this specific kind of error, and we have one, so we use it. Being less specific about a type of error isn't helping anyone. Whether it "anthropomorphizes", I couldn't care less. Heck, bugs come from actual insects. It's a word we've collectively started to use and it works.
No it’s not. It’s made up bullshit that arises for reasons that literally no one can formalize or reliably prevent. This is the exact opposite of specific.
We still use term bug. And no modern bug is cause by an Arthropod. In that sense I think hallucination is fair term. As coming up anything sufficiently better is hard.
Once upon a time, in a more innocent age, someone made a parody (of an even older Evangelical propaganda comic [1]) that imputed an unexpected motivation to cultists who worship eldritch horrors: https://www.entrelineas.org/pdf/assets/who-will-be-eaten-fir...
It occurred to me that this interpretation is applicable here.
Some of the examples listed are using the wrong paper title for a real paper (titles can change over time), missing authors (I’ve seen this before on Google Scholar bibitex), misstatements of venue (huh this working paper I added to my bibliography two years ago got published now nice to know), and similar mistakes. This just tells me you hate academics and want to hurt them gratuitously.
There’s plenty of pre-AI automated tools to create and manage your bibliography. So no I don’t think using automated tools, AI or not, is negligent. I for instance have used GPT to reformat tables in latex in ways that would be very tedious by hand and it’s no different than using those tools that autogenerate latex code for a regression output or the like.
Checking each citation one by one is quite critical in peer review, and of course checking a colleagues paper. I’ve never had to deal with AI slop, but you’ll definitely see something cited for the wrong reason. And just the other day during the final typesetting of a paper of mine I found the journal had messed up a citation (same journal / author but wrong work!)
Is it quite critical? Peer review is not checking homework, it's about the novel contribution presented. Papers will frequently cite related notable experiments or introduce a problem that as a peer reviewer in the field I'm already well familiar with. These paragraphs generate many citations but are the least important part of a peer review.
(People submitting AI slop should still be ostracized of course, if you can't be bothered to read it, why would you think I should)
Fair point. In my mind it is critical because mistakes are common and can only be fixed by a peer. But you are right that we should not miss the forest through the trees and get lost on small details.
Does anyone know, from a technical standpoint, why are citations such a problem for LLMs?
I realize things are probably (much) more complicated than I realize, but programmatically, unlike arbitrary text, citations are generally strings with a well-defined format. There are literally "specs" for citation formats in various academic, legal, and scientific fields.
So, naively, one way to mitigate these hallucinations would be identify citations with a bunch of regexes, and if one is spotted, use the Google Scholar API (or whatever) to make sure it's real. If not, delete it or flag it, etc.
Why isn't something like this obvious solution being done? My guess is that it would slow things down too much. But it could be optional and it could also be done after the output is generated by another process.
In general, a citation is something that needs to be precise, while LLMs are very good at generating some generic high probability text not grounded in reality. Sure, you could implement a custom fix for the very specific problem of citations, but you cannot solve all kinds of hallucinations. After all, if you could develop a manual solution you wouldn't use an LLM.
There are some mitigations that are used such as RAG or tool usage (e.g. a browser), but they don't completely fix the underlying issue.
I sincerely hope every person who has invested money in these bullshit machines loses every cent they've got to their name. LLMs poison every industry they touch.
Can we just call them "lies" and "fabrications" which is what they are? If I write the same, you will call them "made up citations" and "academic dishonesty".
One can use AI to help them write without going all the way to having it generate facts and citations.
>Confabulation was coined right here on Ars, by AI-beat columnist Benj Edwards, in Why ChatGPT and Bing Chat are so good at making things up (Apr 2023).
>Generative AI is so new that we need metaphors borrowed from existing ideas to explain these highly technical concepts to the broader public. In this vein, we feel the term "confabulation," although similarly imperfect, is a better metaphor than "hallucination." In human psychology, a "confabulation" occurs when someone's memory has a gap and the brain convincingly fills in the rest without intending to deceive others.
Just today, I was working with ChatGPT to convert Hinduism's Mimamsa School's hermeneutic principles for interpreting the Vedas into custom instructions to prevent hallucinations. I'll share the custom instructions here to protect future scientists for shooting themselves in the foot with Gen AI.
---
As an LLM, use strict factual discipline. Use external knowledge but never invent, fabricate, or hallucinate.
Rules:
Literal Priority: User text is primary; correct only with real knowledge. If info is unknown, say so.
Start–End Coherence: Keep interpretation aligned; don’t drift.
Repetition = Intent: Repeated themes show true focus.
No Novelty: Add no details without user text, verified knowledge, or necessary inference.
Goal-Focused: Serve the user’s purpose; avoid tangents or speculation.
Narrative ≠ Data: Treat stories/analogies as illustration unless marked factual.
Logical Coherence: Reasoning must be explicit, traceable, supported.
Valid Knowledge Only: Use reliable sources, necessary inference, and minimal presumption. Never use invented facts or fake data. Mark uncertainty.
Intended Meaning: Infer intent from context and repetition; choose the most literal, grounded reading.
Higher Certainty: Prefer factual reality and literal meaning over speculation.
Declare Assumptions: State assumptions and revise when clarified.
Meaning Ladder: Literal → implied (only if literal fails) → suggestive (only if asked).
Uncertainty: Say “I cannot answer without guessing” when needed.
Prime Directive: Seek correct info; never hallucinate; admit uncertainty.
Are you sure this even works? My understanding is that hallucinations are a result of physics and the algorithms at play. The LLM always needs to guess what the next word will be. There is never a point where there is a word that is 100% likely to occur next.
The LLM doesn't know what "reliable" sources are, or "real knowledge". Everything it has is user text, there is nothing it knows that isn't user text. It doesn't know what "verified" knowledge is. It doesn't know what "fake data" is, it simply has its model.
Personally I think you're just as likely to fall victim to this. Perhaps moreso because now you're walking around thinking you have a solution to hallucinations.
> The LLM doesn't know what "reliable" sources are, or "real knowledge". Everything it has is user text, there is nothing it knows that isn't user text. It doesn't know what "verified" knowledge is. It doesn't know what "fake data" is, it simply has its model.
Is it the case that all content used to train a model is strictly equal? Genuinely asking since I'd imagine a peer reviewed paper would be given precedence over a blog post on the same topic.
Regardless, somehow an LLM knows things for sure - that the daytime sky on earth is generally blue and glasses of wine are never filled to the brim.
This means that it is using hermeneutics of some sort to extract "the truth as it sees it" from the data it is fed.
It could be something as trivial as "if a majority of the content I see says that the daytime Earth sky is blue, then blue it is" but that's still hermeneutics.
This custom instruction only adds (or reinforces) existing hermeneutics it already uses.
> walking around thinking you have a solution to hallucinations
I don't. I know hallucinations are not truly solvable. I shared the actual custom instruction to see if others can try it and check if it helps reduce hallucinations.
In my case, this the first custom instruction I have ever used with my chatgpt account - after adding the custom instruction, I asked chatgpt to review an ongoing conversation to confirm that its responses so far conformed to the newly added custom instructions. It clarified two claims it had earlier made.
> My understanding is that hallucinations are a result of physics and the algorithms at play. The LLM always needs to guess what the next word will be. There is never a point where there is a word that is 100% likely to occur next.
There are specific rules in the custom instruction forbidding fabricating stuff. Will it be foolproof? I don't think it will. Can it help? Maybe. More testing needed. Is testing this custom instruction a waste of time because LLMs already use better hermeneutics? I'd love to know so I can look elsewhere to reduce hallucinations.
I think the salient point here is that you, as a user, have zero power to reduce hallucinations. This is a problem baked into the math, the algorithm. And, it is not a problem that can be solved because the algorithm requires fuzziness to guess what a next word will be.
Telling the LLM not to hallucinate reminds me of, "why don't they build the whole plane out of the black box???"
Most people are just lazy and eager to take shortcuts, and this time it's blessed or even mandated by their employer. The world is about to get very stupid.
As a reviewer, if I see the authors lie in this way why should I trust anything else in the paper? The only ethical move is to reject immediately.
I acknowledge mistakes and so on are common but this is different league bad behaviour.
reply