Steinn Sigurðsson

Can anyone explain this to me? I Googled a 1972 dissertation about historical fire in the North American West. Google gave me this "AI summary." It gave the right title, provided a decent summary, accurately described its importance, and provided a Proquest link to it. But...

@tunguz Traveling salesman efficient routing optimization can in principle lead to large immediate savings

What is *one* hard unresolved *pure* mathematical problem on which a breakthrough would have the biggest *immediate* impact on the economic activity?

An internal OpenAI model has disproved a central conjecture in discrete geometry: If you place n points in the plane, how many pairs of points can be exactly distance 1 apart? It is clear that AI has reached the threshold of superhuman capability in pure math and (I assert) theoretical physics. Quotes below from companion paper, linked in thread. The Erdős unit distance problem [14] raised in 1946 is among the best known open problems in Combinatorics. It is also arguably the best known problem in Discrete Geometry. Indeed, its description in the book of Brass, Moser and Pach on Research Problems in Discrete Geometry ([9], Chapter 5) is: “The following problem of Erdős [14] is possibly the best known (and simplest to explain) problem in combinatorial geometry: How often can the same distance occur among n points in the plane?” ... I believe it would be fair to say that every mathematician working in Combinatorial Geometry thought about this problem, and lots of mathematicians working in other areas spent at least some time thinking about it. Let me also add that although this problem may look at first as a recreational one this is not the case, it is in fact closely related to other mathematical areas including Number Theory and Algebraic Geometry. ... the fact is that the AI was able to do here what lots of excellent human researchers tried and failed to do. ... my impression has been that AI tools are capable of changing research in mathematics in a dramatic way. The new spectacular solution of the Erdős unit distance problem convinces me that it is hard to overestimate the full potential impact of this change.

This is impressive: it is a problem I had actually heard of. It looks like the solution approach is surprising to mathematicians. It was a general reasoning model rather than a specialized one: bitter lesson time. I think the stochastic parrot is now nuked from orbit.

@mbeisen arxiv.org/abs/2505.13400

The "we're using AI to automate everything a scientist does" paper WAS IN REVIEW AT NATURE FOR A ENTIRE YEAR. That is some God-tier verisimilitude, replicating all the catastrophically stupid decisions meat scientists make.

The relationship between this and the maths result I previously tweeted about reminds me of this passage from Beatrix Potter's The Tale of Jeremy Fisher (except that I'd replace "worse" by "more disturbing").

AI has now solved a major open problem -- one of the best known Erdos problems called the unit distance problem, one of Erdos's favourite questions and one that many mathematicians had tried. openai.com/index/model-di…

@CburgesCliff @REasther @petergklein a suspension is "no submission will be permitted" for that period coming out of suspension, the first submission(s) will have to have been accepted at a reputable refereed venue requirement for showing acceptance may be relaxed, on appeal, typically after 3 good submissions

@REasther @steinly0 @petergklein No. They said hallucinated references will require a one year pause during which the affected party can still post but only after first being published.

The latest outrage on academic Twitter is the idea that reference lists may contain the occasional error and that it isn’t always worth finding and fixing these. Is John points out below, this discourse is largely silly. But there is an underlying point that is being missed. 1/

@REasther @petergklein just to clarify the parsing: arxiv has used suspension of submission privileges as a sanction on authors, for a number of years - a small number of decades the duration of suspension in varies currently for incontrovertible AI slop suspension is for 1 year

@REasther @petergklein info.arxiv.org/help/policies/… authors have been suspended for various periods for decades for code of conduct issues

@lpachter @cociclo from 2011 - core of it is to do string matching on author names followed by cluster analysis hand curated to disambiguate everyone is working on AI metadata extraction, it will be done well enough and then better still soon everything is a moving target

@steinly0 @cociclo INSPIRE-HEP uses Authormagic: inspirehep.net/files/2b3dd3cc…

It pains me to report @WKCosmo for what he terms ACADEMIC MISCONDUCT. The four authors in Ref. 15 of his "Phase Transitions and Gravitational Waves" arxiv.org/abs/2605.05019 are HALLUCINATED. Also citing an ERRATUM? Did @WKCosmo read his own paper. BAN FOR LIFE?

@lpachter @cociclo in 2019/2022 it was probably just a regular expression scrape is why people are traditionally paid to clean up indexing

@steinly0 @WKCosmo @cociclo As for INSPIRE-HEP, it's based on automatic machine extraction (see help.inspirehep.net/knowledge-base…). In this case there was an error: there are only 2, not 4 authors on this paper. The paper is real and exists but the four authors aren't a typo. They're a machine produced hallucination.

@lpachter @cociclo no there are many ways in which references can have errors there are a range of code of conduct issues, the _example_ that was one of several that started is is an incontrovertible one where papers that do not exist are cited

@steinly0 @WKCosmo @cociclo ICLR2026 flagged as a "verified hallucination" a paper which they describe as "exists, but all authors are wrong" (see the TamperTok paper, first on the list here gptzero.me/news/iclr-2026/). The paper is arxiv.org/abs/2304.06718 I could find this paper no problem. Is that ok then?

@lpachter @WKCosmo @cociclo the specific example of "AI hallucinated papers" are papers that literally do not exist sometimes they are the papers those authors probably ought to have written, but didn't most of the time they're just word salad author authentication is a distinct issue, also important

@steinly0 @WKCosmo @cociclo I agree with you that "the paper exists and can be found" is a reasonable bar but many AI hallucinated papers do exist and can be found! I personally still would rather not see even those on the arXiv but I also think the deterrent is the preprint! x.com/lpachter/statu…

@WKCosmo Step 4: open the PDF and <s>read</s> <i>skim</i> it to make sure it is at least somewhat relevant and makes a coherent point or provides context...

Step 3: add those papers to your bibliography.

How to use a LLM to generate a reference list: a step-by-step guide.1/

@CburgesCliff I dub thee Cliff "The Quinn" Burgess

Recently filled in a form and was given a marvellous list of titles to choose from. Started with Mr, Mrs, Ms and continued through Dr and Prof to these. I have clearly been missing out.

@EBasilion @mbeisen all the Nobel Laureates I have personally known are and were pretty chill overall

@mbeisen @steinly0 Nobel laureate? It’s a feature, not a bug.

A Nobel laureate widely celebrated for his collegiality and integrity once, when asked why his Nature paper didn’t cite earlier work that made the same “discovery” his paper asserted replied by saying “I don’t read other people’s work so that I don’t get in trouble for not citing it.

@lpachter @WKCosmo @cociclo the citation is on topic, to a paper and a subsequent erratum the authors are real and correct from the cite you can correctly identify the titles, journal, volume and page numbers that is the purpose of a cite, so readers can find it it is formatted as per INSPIRE-HEP

@WKCosmo @cociclo What is 100% for sure is that this is NOT a correct citation. Something you can verify in a second by just looking up the paper in the journal.