The purpose of citations

As irritated as I was with Zdeborova's posts, there is good news that comes with it.

https://x.com/zdeborova/status/2056135395591491632

If you want to read, you can. Her complaint centers on the year-long publishing ban, likening it to having your license suspended for a year if you got caught speeding.

Then she suggests that maybe she should be banned because she cited a paper in her PhD dissertation that did not say what she said it did. 

Maybe she shouldn't be banned solely for that, but if it is her habit (as she indicates) to cite sources without reading them, that might indicate shoddy work that shouldn't have passed the defense.

The issue with both Zdeborova and Miller is that they leap from some errors being inevitable -- to which there is some truth -- to apparently not finding any accountability necessary.

Errors do happen.

For my literature study I read a lot of papers. I kept notes with my sources so I could remember where key points were, and I had copied them into the document I was writing in to make sure I didn't miss anything. 

I left one of those notes in what I turned in. It doesn't make anything inaccurate; it's just something that was only for me. I was mortified to find it, but it still passed.

That being said, I read every single one of those sources, some of them more than once. I also read papers that I didn't use. I read them because I thought they might be relevant, but then they really weren't, even if they were still kind of interesting.

One paper focused on the importance of having family or community members advocating for the patients. A focus of what I was looking at was self-advocacy, which they didn't even consider as a possibility. I noted that contrast, because whether something supports your work is a separate question from whether it is related to your work.

It would be extremely obnoxious to think I know it all now, but this is an area where I am currently spending hours, giving me some perspective on it. You should know what people are saying, even if you end up thinking they are wrong. Referring to previous work is supposed to show that you know the foundation, whether you are supporting it or criticizing it. Not doing that work means your work can't be trusted.

Zdeborova's answer to this is that no one's work should be trusted; that's how science works. People need to research and experiment.

There is a level at which that is true; reproduction of results is important, if not always possible. Papers will generally have a section that clarifies what issues weren't addressed or other factors that should be investigated. However, if there is not going to be anything trustworthy that can be built on in the work of others, then there is no point in publishing or citing anything.

There is a lot more that could be said about honesty, accountability, and respect for knowledge and skill. Instead, I am going to point out the good news.

I name those two people because the vast majority of responses were strenuously objecting. Of course you read your sources! What is wrong with you? 

This is an interesting thing from the background on the new arXiv policy: the last checked rate of hallucinated citations is 1 in every 277 papers. 

https://byteiota.com/arxiv-bans-authors-1-year-for-ai-hallucinated-citations/

That is a tenfold increase in three years, so that is serious. There are also some other stats where the number is higher, but that seems to mean that the papers that have hallucinated citations have multiple, with others having none. 

Still, the majority is not doing it. At least not yet.

A lot of AI and engagement farming with fake stories is designed not just for specific deceptions, but to create a general sense of distrust.

Lots of people are still honest and diligent. They still care. 

That's worth something.

Let's build on that.