NIH limits scientists to six applications per year
science.org
science.org
"Lauer notes that not long before he left NIH, he and his colleagues identified a principal investigator (PI) who had submitted more than 40 distinct applications in a single submission round, most of which appeared to be partially or entirely AI generated. The incident was “stunning” and “disappointing,” says Lauer, who was not involved in creating the new NIH policy but hopes the cap will discourage other researchers from abusing the system."
Always somebody who ruins things for everybody else.
The main gripe isn’t the fact that there is a limit, but rather that the threshold is too low. If you doubled the threshold to 12, I don’t think you’d see much pushback and still be able to limit over submissions. This being NIH, I’m a bit surprised there wasn’t more data presented to show why 6 was a reasonable limit.
Although it might also just be an impact of illusions of nobility from the past shattered by increased transparency. George Washington abused his power to land speculate for one, presidents who had their affair participant institutionalized for stating uncomfortable truths about whose child was whose.
It is relevant, though awkward to discuss, that a large share of NIH and NSF funding proposals (and indeed funded projects) are led by researchers who didn't grow up in the US. I wonder if it's in fact a majority.
It's common that the head of the laboratory submit the applications for each project, so 40 application may mean 40 subteams with 2-6 minions each (where each minion has a Ph.D. or is a graduate student.) Usually when the paper is published, the head of the laboratory is the "last author".
Now it's getting common to make an AI cleanup, like fixing orthography and grammar and perhaps reduce the text to 5000 characters. Without reading them it's hard to know if this is the case or it's nonsensical AI slop.
I believe this is the reason they are limiting it. A not of grants require the PI to spend at least some percentage of their research time on the project. PIs trend to ignore that requirement. As a result big names were getting a huge number of grants and early career researchers were getting none. Requirements like these give other researchers a chance.
Besides reducing the incentive to spam, this rule has had another positive effect: As a researcher without funding, you don't have to spend your whole year writing grants. You can, instead, spend your time on actual research.
With that said, NIH grants tend to me much more narrow than CS ones, and I imagine that it takes a lot more grants to keep a lab going...
[1] https://www.nsf.gov/funding/opportunities/computer-informati...
This specific solicitation — CISE Core Programs — has a 2-proposal-per-year limit. However, that only applies to this solicitation, and only counts proposals submitted to this solicitation. CISE Core Programs is an important CS funding mechanism, but there are quite a few other funding vehicles within CISE (Robust Intelligence, RETTL, SATC, and many more, including CAREER). Each has its own limits, that generally don't count or count against the Core Programs limit.
But in general science will have to deal with that problem. Written text used to "proof" that the author spend some level of thought into the topic. With AI that promise is broken.
I'm not a scientist/researcher myself, but from what I hear from friends who are, the whole "industry" (which is really what it is) is riddled with corruption, politics, broken systems and lack of actual scientific interest.
Many systems are going to have to come up with better solutions to the problems AI will pose for legacy processes.
This barrier is clearly broken now.
A different barrier could be money that people submitting grant proposals would need to pay. First grant proposal could be $0, second $1, third $10, fourth $100, etc.
My idea doesn't involve any of that -- you want to submit 10 proposals? $111,111,111 please.
But it's similar to emails, under water your e-mail account has a 'trust score' based on previous behaviour, domain, etc. It could also come from the other side, if a scientist is attached to a university or other research body, they should sign off on a declaration that AI was not used (or used but clearly marked as such), with a big fine and reputation damage for the university if their researchers violate it.
If I write anything, and put my name as the author, I'm 100% lying if I am just copy pasting text.
This holds true 10 years ago, if I copied in any text without attribution. A novel, a book, a grant app, a paper, whatever.
Just because you're now copying large swaths of text from an LLM, doesn't make it better than copying from a person, eg plagiarism. And if you took a person's text 10 years ago, and modified a few words out of thousands, yes, that'd be called plagiarism too.
(No, a spell checker isn't that. It's correcting your word for the same word. If you think spellcheckers are the same as whole paragraph insertion, please check your ethics meter, it's broken.)
If the work isn't yours, you need to say so. Otherwise you're being dishonest.
If people get upset at the notion of disclosing, that feels like guilty behaviour. Otherwise, why not disclose?
Now, taking a step back? We're in a period of transition.
I agree that vast imbalances are being created here. This is the true problem.
For example, an application process could state "LLM applications are fine", or not. Instead?
The current is "no" without clearly saying so, for obvious reasons (it's copying work you didn't write, as your own ... plagiarism), but any such "no" without a high incident of detection and punishment, is worse than anything.
The 6 applications seems like a cop out, although it is logical. It should also be coupled with a "OK you win, use LLMs" statement too.
On another note, soon there will be two types of people, and only one of which I will engage in thoughtful email/text communication with.
Those who use LLMs, and people worth talking too.
Of what value is any meaningful conversation, if the other person's response is an LLM paste? Might as well just talk to chatgpt instead.
(note, I'm talking about friendly debate among friends or colleagues. Seeking their opinion or vice versa.)
I disagree. It's like doing collage art with text. Like using samples to make music. We are already in that era of collating text and images and video from AIs. We should learn to embrace
It seems to me that the incentives are such that now you've just guaranteed that all PIs will now submit 6 applications every year.
It may be less that what you originally were seeing, but I don't know the population stats.
It also may be that you've now poisoned all the other grant agencies (NSF, etc) and they'll soon have to have maximums too.
> According to the new notice, the number of PIs who submit more than six applications per year is “relatively low.”
I imagine that, given funding cuts, PIs are going to try to work harder to find funding opportunities (i.e. more proposals submitted) for insurance.
Unless they start using AI on their end to review the quality as well, which I don’t think is the way we want to have things going.
... and everybody else is going to be assumed to be a software system.-
PS. Even worse (better): Your agent is going to be cryptographically bound to your identity.-
I don't see a way of ensuring actual human input that doesn't involve panopticon-level surveillance.
(Yes, I'd say it'd mitigate the issues somewhat, with the putative slowdown ...)
I wouldn't be surprised if the institutions or projects like CURL will create harsher measures to stop this flood of high-quality spam, like putting people on a list or requiring payment per submission.
I mean that last one isn't a bad thing, Apple did something like it years ago and I think asking for an upfront cost and having a strict (at the time) review program made it so that all apps were from serious developers and met minimum quality standards.
Hammers build homes, and smash faces
Encryption keeps honest people’s information safe, but it also can be used for drug trafficking
The internet is used for social media and news. But it’s also used for child porn.
Tools are neutral, it’s how they are used that makes a tool like a gun, something that feeds a family or something that kills innocent people.
"Please don't fulminate."
"Eschew flamebait. Avoid generic tangents."
- Massive time sink. Those of us at senior/PI levels devote a lot of time to grant writing, often more than to actual research.
- Not something that you really get much useful learning or enrichment from (apart from learning to write better grant proposals the next time). The part of brainstorming and structuring ideas is useful but you would mostly do it without grant writing anyway, all the actual writing and polishing (which is 95% of the time) isn't. Definitely not an efficient use of the amount of hours it takes.
- I don't know specifically for NIH, but in my (non-US) context, grant proposals are full of formulaic sections that aren't really useful (Gantt chart, data management plan, etc.) When I'm in an evaluator role, I tend to outright skip many of them, not out of neglect or laziness but because they're just useless ritual fluff.
- As a consequence of the above three points, most of us dislike or even hate this part of our work.
- The meta for most funding agencies I know has long been to overhype and to use exaggeratedly positive language and takes. Exactly what LLMs are naturals at.
- If you're a non-native English speaker and write grant requests in English (common in Europe), the LLM also helps you level the playing field with native speakers, which is quite a big deal. From a naive outside standpoint you might think that scientific grant evaluation is all about the actual ideas and CVs, but the truth is that in practice, ability to pitch your ideas better than other competitors in your call is key.
- Honestly in the last grant I wrote, Gemini came up with some paragraphs that I consider to be clearly better than what I would have written by myself. Clear, concise, attractive to read, etc. It's just very good at writing. I'm better than it at the actual research, but at writing, let alone in English where I'm not a native? I don't have a chance.
As a result of this... good luck convincing scientists not to use LLMs for this. I'm pretty sure that if you ask, you will find two types of scientists: those that tell you that they use LLMs for grant writing and those who are hypocrites and deny it. I wouldn't even trust a scientist who didn't use LLMs for this (unless it's out of some very deep quasireligious conviction): why waste your time? Don't you want to have more time to do actual science?
I believe that LLMs can be very useful to identify this stuff in our processes. The solution shouldn't be then to fill them with LLMs but strip them entirely away. I tend to think the same about everyone freaking out about LLM misuse in education.
The problem is the bureaucracy. And if it asks for useless fluff, I'm happy to feed it with LLMs.
Could LLM participation be blowing holes in good-governance measures that were only weakly effective, and therefore a good thing in the long-term? Could the rise in the practice drive grants arrangements to better governance?
Indeed, some of the fluff is due to the first reason - for example, the data management plan (where you need to specify how you're going to handle data) has good intentions: it's there so that you explain how you will make your data findable, interoperable, etc. which is a legitimately positive thing; as opposed to e.g. never releasing the research software you produce and making your results unreproducible. But the result is still fluff: I (well, Gemini and I) wrote one last week, it's 6 pages, and what it says can be said in 2-3 lines: that we use a few standard data formats, we will publish all papers on arXiv and/or our institutional repository, software on GitHub, data on GitHub or data repositories, and all the relevant URLs or handles will be linked from the papers. That's pretty much all, but of course you have to put it into a document with various sections and all sorts of unnecessary detail. Why? I suppose in part due to requirements of some disciplines "leaking" into others (I can imagine for people who work with medical data, it's important to specify in fine detail how they're going to handle the data. But not for my projects where I never touch sensitive data at all). And in part due to the trend of bureaucracies to grow - someone adds something, and then it's difficult to remove it because "hey, what if for some project it's relevant?", etc.
Then there are things that are outright useless, like the Gantt chart. At least in my area (CS), you can't really Gantt chart what you're going to do in a 5-year project, because it's research. Any nontrivial research should be unexpected, so beyond the first year you don't know what you'll exactly be doing.
Why is that there? I suppose it can be a mix of several factors:
- Maybe again, spill from other disciplines: I suppose in some particular disciplines, a Gantt chart might be useful. Perhaps if you're a historian and you're going to spend one year at a given archive, another year at a second archive, etc... but in CS it's useless.
- Scientists who end up at bureaucratic roles are those that don't actually like doing science that much, so they tend to focus on the fluff rather than on actual research.
- Research is unpredictable but funding agencies want to believe they're funding something predictable. So they make you plan the project, and then write a final report on how everything turned just as planned (even if this requires contorting facts) to make them happy.
You don't need language models to identify useless processes. The problem, however, is that people tend to be more comfortable with a process that exists whose product is ignored rather than no process at all.
For example, in the case of the grants here it's easier to imagine giving money to someone with a Gantt chart – even if that chart will never really represent reality – rather than someone who says 'trust us to use the money effectively.'
For an alternative view, a lot of the information supplied in such processes isn't related to the happy path, but rather it creates a paper trail for blame when things go wrong.
> I tend to think the same about everyone freaking out about LLM misuse in education.
The difference for education is that students need to practice, so the repetition is the point. The AI might ultimately be better at writing the book report, particularly compared to a student in 6th grade, but there's few other ways to train skills of reading comprehension and analysis.
My comment was more about the part that says
"Aside from the cap, the policy makes clear that NIH will not consider AI-generated proposals to be the original work of applicants. “NIH will continue to employ the latest technology in detection of AI-generated content to identify AI generated applications,” the agency’s notice says. If AI use is detected after an award has been granted, NIH warns, the agency may refer the matter to the Office of Research Integrity while imposing penalties."
I think it's just unrealistic to expect that scientists won't use LLMs for grant proposals. And detection doesn't even actually work without the risk of false positives (not that there are going to be actual false positives, because everyone is going to use LLMs for this so by definition you will only have true positives and false negatives, but it's still unfair if some scientists' proposals are flagged by an AI detector and others bypass it).
All this represents the overwhelming majority of the work (in terms of hours) in writing a grant proposal.
As a funny anecdote and a side note, I noticed a pattern of tricks used by people to detect AI texts that is dangerously close to my natural writing habits in English. For example I've seen people saying text with a lot of "fancier" words ( closer to my native language than more natural words), using a lot of moreover or "-" in the sentences. They are all things I've read people here and elsewhere say it helps them to know that a text is AI generated. I'm now very self-conscious about using moreover.