IEEE is sunk. They embarrassingly just had to pull 1417 papers[0] because they published slop, so they're going to make as many excuses as they can/try and SEO the keywords another way.
This article has a bit more depth and talks about how it's been a problem far longer than GPT. arXiv has caused it's fair share (by making publishing easier/cheaper not because it encourages fraud). Then there's Zenodo - seemingly just a spam source on HN.
this is a big problem i see at work where a lot of papers are read all the time. more and more its trying to sift through things written by LLM and ridden with mistakes, clearly reviewed also using LLMs. reading it is frustrating, u find 25 papers on a topic and only 1 has good info, but u read like 200+ pages and 5 were worth it. its put me off reading papers all together and i see a lot of colleagues growing frustrated. reluctant to write papers themselves seeing their results get swamped under slop.
we use payed platforms for papers, the usual ones, and they are still riddled with this slop. its incredible. U really need to find good researchers and only stick to their outputs, which sadly makes it very hard for new, honest, researchers to make any impression. A lot are simply skipped with a 'i dont know that guy and cant be bothered anymore' kind of tiredness.
I can only compared it to 'alert fatigue' in a security operations center, so many false positives a true positive might be mindlessly skipped as false. But with papers reading the things is much more time consuming... so its perhaps even more annoying and frustrating.
the worst one i saw was a 'paper' where people generafed a questionaire with an LLM, sent it to 6 'field experts' (random friends in the field) and then ran the answers through LLM and poof, paper done. made me want to chuck my laptop out the window.
Is this a kind of campaign? Just saw a similar article posted on HN: https://news.ycombinator.com/item?id=50364330. Science was drowning in slop for at least a decade before Gpt.
IEEE is sunk. They embarrassingly just had to pull 1417 papers[0] because they published slop, so they're going to make as many excuses as they can/try and SEO the keywords another way.
This article has a bit more depth and talks about how it's been a problem far longer than GPT. arXiv has caused it's fair share (by making publishing easier/cheaper not because it encourages fraud). Then there's Zenodo - seemingly just a spam source on HN.
[0]: (37 points, 4 days ago) https://news.ycombinator.com/item?id=50323925
this is a big problem i see at work where a lot of papers are read all the time. more and more its trying to sift through things written by LLM and ridden with mistakes, clearly reviewed also using LLMs. reading it is frustrating, u find 25 papers on a topic and only 1 has good info, but u read like 200+ pages and 5 were worth it. its put me off reading papers all together and i see a lot of colleagues growing frustrated. reluctant to write papers themselves seeing their results get swamped under slop.
we use payed platforms for papers, the usual ones, and they are still riddled with this slop. its incredible. U really need to find good researchers and only stick to their outputs, which sadly makes it very hard for new, honest, researchers to make any impression. A lot are simply skipped with a 'i dont know that guy and cant be bothered anymore' kind of tiredness.
I can only compared it to 'alert fatigue' in a security operations center, so many false positives a true positive might be mindlessly skipped as false. But with papers reading the things is much more time consuming... so its perhaps even more annoying and frustrating.
the worst one i saw was a 'paper' where people generafed a questionaire with an LLM, sent it to 6 'field experts' (random friends in the field) and then ran the answers through LLM and poof, paper done. made me want to chuck my laptop out the window.