Post #2340226

2025-12-17 06:43 UTC

If you are an author whose work is on the ACM DL, I'd like to strongly encourage you to check the accuracy of the AI Summary of your paper, and send feedback to the ACM on this feature and the inaccuracies in your summary. See more here: https://dl.acm.org/generative-ai/summarizations Please boost, repost, and otherwise steal this post to reach other authors. In long: The ACM has begun rolling out "Digital Library Premium" features. A very notable one is AI Summaries of papers, which displace the author written abstract with an AI generated summary of the paper on the front page for the paper. Of the several I've checked, they all contain subtle inaccuracies that would VERY EASILY mislead even experts familiar with the work. This is a HUGE DISSERVICE TO SCIENCE. These summaries might appear to make science more accessible, but subtle errors introduced into the authors original text is the very opposite of making science more accessible. This will mislead rather than educate. The ACM has a feature to send feedback on these summaries. Open your DOI, and look for the "Feedback" button on the right of the page, or the "Send Feedback" link in the footer. (You might need to disable ad-blockers, as this services uses mopinion.com) The ACM is requesting this feedback, and it's important to let them know that such "hallucinations" are not solvable: These tools were designed in consultation with a diverse group of Digital Library stakeholders and will continue to evolve as Artificial Intelligence advances. We are continuously tuning our Foundational Model to optimize readability and we conduct regular audits for hallucinations and other errors. We are very interested in your thoughts and suggestions- please leave them by clicking the "Feedback" button on the far right of this page. If you find a problem with a specific AI-generated summary, please return to that summary and click the Feedback there. More aggressively, you might email dl-team@hq.acm.org (also linked in the footer). Update 1: It seems the ACM no longer displaces the abstract with these summaries, so I can hope this feedback has been helpful in affecting change. Update 2: It seems these summaries have been generated on papers whose license may forbid it, such as CC-BY-ND. If you have a restrictive license on your paper, you may check whether it’s being followed and send further feedback if not.

Replies (8)

@shironeko@fedi.tesaguri.club 2025-12-17 07:09

@wilbowma@types.pl foundational model 🤢

Open ##3015737
@arjen@idf.social 2025-12-17 07:47

@wilbowma@types.pl why would they even think about replacing a abstract with an AI generated one?! I cannot even read their explanation page, it's 404 without a premium account...

Open ##3015738
@DanielRThomas@social.coop 2025-12-17 08:51

@wilbowma@types.pl I find the use of generative AI morally repugnant in most circumstances and this action by @ACM@mastodon.acm.org to corrupt my work in this way is deeply offensive. #GenAI

Open ##3015739
@peter_sewell_@types.pl 2025-12-17 09:51

@wilbowma@types.pl Dear Jack Davidson, Wayne Graves, The ACM Digital Library provides an important service to computer science, facilitating access to the primary research literature. I am dismayed to see that it now, rather than showing the author-written and peer-reviewed abstracts of papers, displays an AI-generated summary. To add insult to injury, it moreover does so in a way that will very easily be misread to be the authors' text. This is antithetical to the practice and progress of science, and misrepresentation of the author and of the peer-review process. It is also a waste of ACM resources. The hazards should be clear to anyone involved in the scientific process. Please remove this feature immediately. best wishes, Peter Sewell Professor of Computer Science, University of Cambridge. FRS

Open ##3015740
@escape_velocity@functional.cafe 2025-12-17 12:26

@wilbowma@types.pl it says I need a premium account to see this summary.

Open ##3015747
@joey@mathstodon.xyz 2025-12-17 15:45

@wilbowma@types.pl Do you know if there's a way for authors to check their own summaries without subscribing? I'm certainly not paying for access and I haven't figured out the right combination of VPN and login voodoo to figure out if I have institutional access.

Open ##3015748
@eldaking@weirder.earth 2025-12-18 00:10

@wilbowma@types.pl Ask them if they accept full liability for inaccuracies and hallucinations, that ought to scare them.

Open ##3015752
@dif@mathstodon.xyz 2025-12-18 12:46

@wilbowma@types.pl oh god WHY

Open ##3015753