A photo of Doc Brown from Back to the Future holding jumper cables labelled "Deranged Activist" with a grant of NIH funding curves below it.
My weekly update on NIH funding
This includes awards made through 2/27/26
Warning: This thread contains data compiled by a known deranged_activist™. Do not read while operating heavy machinery.
1/9
With this war on Iran costing $1B per day, we are now at 4$B, an amount that would fund 3,200 five year NIH biomedical research grants.
But I agree with you on paper or grant writing.
I can confidently say that I don’t think a single one of my clients would care if I generated my invoices using AI.
Depends on the writing. I would absolutely love genAI to do my invoices and email them out. Grant checklists? Generic supporting documents that we can modify? I can think a lot of writing that I would categorize to be as much fun as clothes washing.
My first rotation in grad school was in a testis lab. I did a lot of testes isolation, sperm isolation, and IVF. The discomfort on people's faces did nothing to stop me from sharing.
My kids can verify. It's a running joke in our house.
As NIH moves to parent NOFOs another resource for finding out what topics are of interest. grants.nih.gov/funding/find...
I have seen too many scientists use it to fill in their deficiencies in knowledge (which is the worst possible way to use it because you cannot tell when it gets something wrong). So I have spent way too much time thinking about this
I think we're very much in agreement here, even if we are looking at this from different perspectives. I'm actually trying to develop a whole workshop on how to use genAI for grant writing without thinking that it can write the actual grant (if that makes sense).
The whole point of this thread is: I think it's ok to think about genAI as a tool. What can it do? Is there something that it can help me do faster? Can it relieve my workload? But what genAI can do, what it is built to do, is way more limited than the people like Kustov (OP) are willing to admit.
Yeah, but I am also certain that bubble is about to pop (also not looking forward to that bubble popping because hello widespread financial destabilization).
You might want to reread my thread.
But how it is used and what it can help with, is way more limited than its cheerleaders want to admit.
But my point, and maybe I need to spell that out more clearly is that I think it's ok to think about genAI as a tool. What can it do? Is there something that it can help me do faster? Can it relieve my workload? I'm not anti genAI and I don't think using it is some sort of moral failing.
It's definitely a status threat piece, isn't it?
"I have figured out how to produce papers super quickly with genAI and if the rest of you don't come on board, you will be left behind in the dust."
You're right, I shouldn't have taken the bait. 😂
If as a PI you think your only job is to come up with ideas and everything else is done by others... well, I would be worried about the quality of science produced.
If accuracy matters, and in science it absolutely does, then I don't see how genAI makes a big impact on the most crucial parts of scientific work without diluting field accuracy
And this is just a freaking grant checklist!!!
The idea that genAI making "hallucinating" is ok because humans make mistakes too concerns me about this scientist's approach to scientific accuracy. I cannot imagine anything more devastating to a scientist than analysis in their paper being wrong
Public data is great - but man, so much is missing. Context is everything in biology.
I agree, I think it should change too. Having seen how big labs produce papers, I just don't share others' reverence for them. That said, what makes a paper high impact is not the actual writing, but the data, and I'll die on that hill.
I can't imagine any grant administrator taking those odds. Getting an application administratively withdrawn is every grants persons worst nightmare. It's why we have 3 layers of institutional review and why pushing that submit button is so stressful.
I am into my 2nd week trying to get genAI to produce a reliable checklist from an FOA number and we are into 2 pages of instructions and multiple checklist examples from me, and it still fails to do so. It gets about 95% correct but that 5% wrong can lead to your application getting withdrawn
Finally, I vehemently disagree with his dismissal of "hallucinations" or inherent genAI inaccuracy. The one thing I have learned from my work with genAI so far, is that if your work depends on accuracy (and most science does), you are not going to get very far relying on genAI.
... time point, I don't see that changing. Even in social sciences, for qualitative research for example, you need multiple coders. I don't see how genAI replaces that.
I am also confused as to his dismissal of "research assistants." I cannot imagine anyone who is going to be more helpful moving forward than someone who can do the grunt work of generating data (which is what most research assistants do). Until Claude can show up in the lab at 3 am to do a 12 hr...
What makes you a scientist is generating data and then trying to understand how that data fits within the context of what we currently know. Having someone or something help you put those findings into words makes no difference to me.
Either way, it doesn't matter. What matters is the data. Is it robust? Does it inform the field? Does it help us understand something better today than we knew yesterday? Can it help us improve society?
Given that jobs like "scientific writer" always existed, I don't know that anybody in science that thought that the act of writing out the paper was the science part. So maybe Claude replaces scientific writers or maybe it evens out the playing field between those who can or cannot afford them?
I have worked as a professional scientific writer before, writing both grants and papers for a research group without claiming authorship. I was comfortable with that arrangement because while I "wrote" the paper, I had no part in the actual generation of data or analyses.
