Fenner Tanswell
banner
fennert.bsky.social
Fenner Tanswell
@fennert.bsky.social
Philosopher of Mathematics and Logic working at TU Berlin. Also into epistemology, science, Xphi and the philosophy of extinction. Flightfree and carfree fennertanswell.com
This time just the two of you.
November 10, 2025 at 1:44 PM
So. Never let your guard down.
October 30, 2025 at 11:47 AM
I like to test out what LLMs can and cannot do, and I wrongly thought that a simple websearch and bit of rejigging to not spoil a puzzle should be easy. Indeed, this kind of thing is one of the main use cases being foisted upon us.
October 30, 2025 at 11:47 AM
The LLM had not just hallucinated a couple of answers, but entire game mechanics and puzzles. None of the hints were of any use at all!
October 30, 2025 at 11:47 AM
Problem was, once we tried to actually follow the hints, we quickly hit more dead ends, and some of the things it had described as "subtle clues" were, in fact, not present in the game at all.
October 30, 2025 at 11:47 AM
So I asked ChatGPT, and it searched the web, then seemed to do a marvellous job at pointing us in the right direction of some things to try next. I took this to be reliable because the web is literally full of walkthroughs for it to use as a basis for its answers.
October 30, 2025 at 11:47 AM
So I thought ChatGPT might be able to help with some hints without spoiling the puzzles completely, which most classic walkthroughs tend to.
October 30, 2025 at 11:47 AM
I was stuck on a computer game (Blue Prince, it's really good) that has a lot of puzzles that are quite tricky. I'd been playing with my son, and it was getting a bit frustrating not making progress.
October 30, 2025 at 11:47 AM
Reposted by Fenner Tanswell
A lady should feel safe in a castle, but a lady in a castle that has survived repeated assaults from well-provisioned armies should feel much safer. #falsifiability
October 29, 2025 at 5:50 PM
Not wanting to lend legitimacy to ten minute rule motions was the stated reason.
October 29, 2025 at 4:47 PM
I suppose what is trivial to Scholze isn't necessarily trivial to everyone else!
October 29, 2025 at 4:03 PM
Here's Cath O'Neil in 2012 on why the purported proof is impossible to follow:
mathbabe.org/2012/11/14/t...
The ABC Conjecture has not been proved
As I’ve blogged about before, proof is a social construct: it does not constitute a proof if I’ve convinced only myself that something is true. It only constitutes a proof if I can read…
mathbabe.org
October 29, 2025 at 4:03 PM
I think that's the prevailing view now, but it did take a while. The original IUT papers were uploaded in 2012, and in 2015 there was a big workshop in Oxford to try to engage with it, which wouldn't have happened if it were so clear.
October 29, 2025 at 4:03 PM
Ah sorry, the Reddit crank link didn't go in. Here you go: www.reddit.com/r/badmathema...
From the badmathematics community on Reddit
Explore this post and more from the badmathematics community
www.reddit.com
October 29, 2025 at 3:43 PM
The end.
October 29, 2025 at 3:32 PM
(My next forthcoming paper with @asgeirberg.bsky.social is partially on this problem and LLM paper. Happy to share the draft, as it should be out whenever the Annals of Maths&Phil get their act together).
October 29, 2025 at 3:32 PM