Puzzle 1289. Prime Puzzles and AI: a pause to think together

This week, instead of a new puzzle, I would like to ask all of you for something different: your opinion.

AI tools are changing the way puzzles are solved, here and everywhere else. Several of you have already said so in your recent answers. Before deciding whether anything should change on this site, and what, I want to hear from the people who make it what it is: the puzzlers. No decision has been made yet.

To frame the question, here are three texts.

1. Bas van Dorst (from Puzzle 1288)

Asked about his view on solving puzzles with AI, Bas answered:

“It’s like using a sudoku or Rubik’s cube solver. It works, but you skip the fun part ;) For me, the fun part was improving an LLM’s reasoning. Your website is great for this: clear challenges, a good structure, clear explanations, and earlier solutions.”

2. Giorgos Kalogeropoulos (from Puzzle 1288, hia answer to Q4)

I think it was a matter of time before a massive attack would happen to this site as it has happened to all the sites that host collections of open problems.
Times are changing and we all have a new powerful tool in our hands. A very well known site with Erdos problems (erdosproblems.com) had to go through many changes lately because of the heavy use of AI.
I would recommend to everyone to read the following post by the owner of Erdos-problems: https://www.erdosproblems.com/forum/thread/blog:9
So, what are the consequences for recreational math and puzzle making/solving?
If you have read the book Recreations in the Theory of Numbers by Albert H. Beiler (1964) you will see that people at that time who used to solve puzzles with pen and paper were amazed with modern machines that could quickly solve problems that needed years of computations by hand. Can we adapt to the new era in order to keep puzzles alive?
I think we can but it might be tricky. Sites like primepuzzles that post new-unsolved problems, for the first time, have to face the fact that anybody can easily (most of the times) find the solution (or the algorithm that finds the solution) with just one click. The last few years I really enjoyed this site with all these nice problems but mostly every week I was urging to see what the other puzzlers would answer!
It was like a nice competition with people I (sort of) knew their profiles. And there was a diversity: others had a higher math education, others had access to CPUs and others had plenty of time for recreational maths. I think with the heavy use of AI we will lose all this diversity and the answers will be related to how many llm-tokens we can spend. Also, there is a problem with puzzle making: Should we make puzzles for puzzlers or for AI? I hope that we will all use this new tool in a creative way so that we can keep puzzle making/solving alive!

3. Thomas Bloom and erdosproblems.com (summary)

Thomas Bloom runs erdosproblems.com, a collection of more than 1,200 problems posed by Paul Erdős. When he added comments in 2025, a lively community grew there. Once AI tools could solve some of these problems, however, most of what the site received were AI-generated proofs posted with no explanation, just to claim priority. Guided by the idea that the site should help the human community around these problems flourish, Bloom made, as an experiment, four changes. He paused comments and proof claims on problems. He stopped showing which problems are “open” or “solved”, to discourage glory-chasing. He dropped credit language (“X proved…”) for new results. And he now gives priority to clear, well-written explanations, whether human or AI. He does not reject AI, but asks that mathematics not be abandoned as a human activity: “Be human. Let your brain be open.”

Full post: https://www.erdosproblems.com/forum/thread/blog:9

***

Our puzzles are recreational, many of them computational, and this site has always been a tacit conversation among puzzlers who, over the years, have come to know each other. That is exactly why your views matter.

Questions

Q1. Do you use AI when working on these puzzles? If so, how — for ideas, to write search code, to check results, or to produce full solutions?

Q2. What do you value most in a puzzle answer published here: the result (a number, a record), the method, or the explanation of how it was found? Has AI changed that for you?

Q3. Should contributors be asked to state briefly how AI was used in their answer? Would you be willing to do it yourself?

Q4. Should AI-assisted and unaided answers be distinguished in some way (separate mentions, notes, nothing at all)? Or would that be unfair or impossible to verify?

Q5. On puzzle making: should future puzzles try to resist AI (e.g., heavy computation, open-ended records), or should some puzzles embrace it openly — as Bas suggests, making the challenge itself about how well humans and AI reason together?

Q6. Is there anything you would like the site to start, stop, or keep doing, in view of all this?

How to answer

Answer any or all of the questions, briefly or at length, by email as usual. Please say whether I may publish your opinion with your name, or prefer it to appear anonymously, or prefer it to stay private between us. Every opinion counts, including from those of you who read the site every week but have never written to me.

I will publish the answers received in the coming weeks, and only then will I decide whether anything should change.