 We're so fucked. "Our analysis of a selection of questionable GPT-fabricated scientific papers found in Google Scholar shows that many are about applied, often controversial topics susceptible to disinformation: the environment, health, and computing. The resulting enhanced potential for malicious manipulation of society’s evidence base, particularly in politically divisive domains, is a growing concern." 57 comments
@cobalt @ct_bergstrom @Blob_Calder @cobalt @ct_bergstrom Well indeed… The reason why we do what we do is the real problem. It’s not only our endeavour to describe and understand our world, as scientists do, but the competitive relations we’ve created to sustain ourselves in relation to doing so…  @cobalt @ct_bergstrom I note "A sample of scientific papers with signs of GPT-use found on Google Scholar was retrieved". So a majority of the papers that were flagged as potentially fraudulent were in fact made using the text generator. That is already quite bad enough, since the number of fraudulent papers indexed by Google Scholar with just a couple of obvious tells of ChatGPT is apparently >100.  @504DR @ct_bergstrom Now I'm gonna have to go watch it again. The perks of having an old, forgetful and covid brain - watching the same movie over and over again, and it's like the first time everytime. 😏  Is it time to find a proper Butler and begin a Jihad? I always thought Herbert was a little over the top. But now I'm not sure. He may have been on to something.  > Once men turned their thinking over to machines in the hope that this would set them free. But that only permitted other men with machines to enslave them. > The target of the Jihad was a machine-attitude as much as the machines ... Humans had set those machines to usurp our sense of beauty, our necessary selfdom out of which we make living judgments. It’s funny, both Tolkien and Herbert had strong Ludditism in their stories, and many AI nerds are likely fans.  @johntimaeus @ct_bergstrom @maegul see also the unveiling of the Torment Nexus, inspired by classic sci fi novel “don’t invent the torment nexus”  @ct_bergstrom Yet academia has embraced it, especially in certain disciplines (like business).  Perhaps some sectors have, but I don't think it's fair to assert this against academia as a whole. Most academics I listen to have been sounding the alarm about AI, not embracing it.  @ct_bergstrom Does Google maintain and publish robust upload logging so that it's possible to determine what actors are flooding the ecosystem with this poison? Or is it just "a random paper has appeared, it came out of nowhere"?  @ct_bergstrom we should start printing reliable sources on paper again to save them from ai corruption @ct_bergstrom I am starting to become quite radicalized by this; I am starting to feel that in order to save society, we must classify AI development as terrorism. @WhyNotZoidberg @ct_bergstrom I'm a teacher at a university. I strongly believe that the emergence of AI/ChatGPT etc is going to set back humanity's development by at least a decade or two. It seems that a lot of people are just becoming reliant on it and are losing the ability to think for themselves.  @artdragon86 @WhyNotZoidberg @ct_bergstrom to be fair the previous tech fads (especially crypto/NFTs) and the way social media has been weaponised by the far right have already set us back about 20 years. @ShadowInTheVoid @artdragon86 @WhyNotZoidberg @ct_bergstrom The dynamics are similar. Many people think that they are resistant to misinformation. However, we don’t see it or don’t want to see it, if it confirms our world view. Add to that, that AI often spits out nonsense in an environment where people expect facts (search engine) and you have the perfect feedback loop.  @artdragon86 @WhyNotZoidberg @ct_bergstrom hot take: people who become reliant on ai couldn't think for themselves to begin with @ct_bergstrom @ct_bergstrom while it is very disheartening, it doesn't change much in reality, does it? Like global warming - if you claim to be skeptic, no amount of valid papers gonna convince you. Same with other issues. @peteriskrisjanis Systemic reviews. Search engines. Training for future LLMs. Daubert decisions. Automated discovery. Need I continue?  The issue here is human bias being proliferated at an exponential rate. These problems were here before. How many scientists in the 50-70’s said lead in gasoline was fine? People point at AI because it’s new. I point at AI not because it's new, but because it exponentially increases the power of misinformation. Creating misinformation by hand is limited in speed and reach. Creating misinformation by automated means (of which AI is the most powerful) is essentially unlimited.  @tschenkel @altruios @ct_bergstrom this, this is the thing. It tilts the field even further, makes all the "old" problems worse, and the various ways in which they're made worse are cumulative. The lie has not only gone around the world, before truth managed to put on it's shoes, it went a dozen times and spawned two generations of offspring, all of them avid travelers. Automation and AI are not the same thing… I generally agree with your point, just not exactly where your point is aimed. The issues have been accelerating since 2016, Cambridge analytica Automation has been going on much longer than AI. The methodology of that time and now look nearly identical: besides image generation (something you probably take a few passes in Photoshop to clean up), I just don’t see where the efficiency gains are. Compared to what was already happening.  @ct_bergstrom honestly, add this to the many reasons. we stopped having functional societies in most places on earth a long time ago, IMO @ct_bergstrom just a wild guess: these same papers produced bij AI are used bij LLM to give answers to search-questions? "Garbage is even more garbage out"?  All scientific publications and journals should immediately require a declaration from authors that no LLM/AI systems were used in the generation of results or the paper being published.  @simonzerafa ... and convincing evidence of a false declaration ought to be grounds for being banned from publishing, by all credible academic journals. For at least 5 years, if not permanently. @ct_bergstrom  @ct_bergstrom@fediscience.org  Already now this is a big problem. The next big thing after the IT revolution… or a next phase of it… @ct_bergstrom you know what's bad then? The fact that they are not reviewed properly. If they are, then we wouldnt be here in the first place. @tengkuizdihar Yeah, so read the article. By and large this is not peer reviewed literature. (1/2) @ct_bergstrom BTW I'm guessing this problem is even worse in China. Where there have long been questionable attitudes towards academic publishing, influenced by decades of the CCP's MiniTru approach to information. Also where training MOLE (Machine Operated Learning Engines) is even more hyped and funded, and less regulated, unless for MiniTru purposes. (2/2) "China continues to have problems with research integrity and has the highest number of retractions of any country due to plagiarism, invented data and fake peer review, but is seeking to improve this by removing cash incentives and use of plagiarism software." #KenHyland, 2023  @ct_bergstrom It's poisoning the well, misinformation and ignorance are great for creating an underclass to exploit and who aren't smart enough to argue back. @nini @ct_bergstrom And that bit about reaching a point where there's no longer an aim of twisting or replacing the truth but destroying the notion of truth altogether. Vanishing all signal in a total wave of noise means inviting all who fear, or are overwhelmed by, the responsibility of decisionmaking to amputate all higher mental function, reason, values, offering carte blanche to run with whatever is most gratifying, easiest, most comfortable. Become livestock. @ct_bergstrom and that is just the beginning. Have you seen this? https://sakana.ai/ai-scientist/ (link to arXiv is also in the post) @ct_bergstrom funny how the people who have the most money in AI are also the ones who have funded or made money from disinformation 🤔 @ct_bergstrom ok so not only do LLMs suck more energy than we can decently produce, and destroy the internet by flooding it with more BS than we can debunk, but its next target is now science. |
Gregory's Server
@ct_bergstrom Oh wow, the scholarly papers already a majority of those cited: "Roughly two-thirds of the retrieved papers were found to have been produced, at least in part, through undisclosed, potentially deceptive use of GPT. The majority (57%) of these questionable papers dealt with policy-relevant subjects (i.e., environment, health, computing), susceptible to influence operations. Most were available in several copies on different domains (e.g., social media, archives, and repositories)"