Great question. People who've used the site heavily from the start seem to largely attribute the initial pre AI drop-off to poor management and a hostile moderator culture that grew a bit too quick to shut down new questions.

Credit to this GitHub thread where I found the data query code and supplemental discussion: gist.github.com/hopeseekr/f5...

Furthermore, do we risk running into a kind of cannibal's dilemma where the abandonment of traditional Q&A forums means less good training data for LLMs? I suppose we'll soon know the answers to all these questions.

I try to be optimistic about new tools, and I've personally found AI amazingly useful for assisting with code. Yet the risk of over-reliance can't be ignored. What happens when students who grow dependent on AI reach a technical level that AI really struggles with, e.g. when they start research?

What does this mean? We are clearly undergoing a profound change in the way students (and when it comes to Stack Overflow, even researchers) seek answers to their technical questions.

The number of monthly questions asked on Stack Overflow is down by about 90% from when ChatGPT was first introduced. The decline is about 50% over the same time period for Physics Stack Exchange.

On a personal note, this was the longest-running project of my PhD, so I am thrilled that it is finally out! I’m excited about the field-theoretic and phenomenological potential of these non-perturbative objects, and I hope our work sparks further exploration! 🚀

We construct Euclidean U(1) dyon loop field configurations and show that they carry Abelian instanton number. We UV complete these “Abelian instantons” using the Georgi-Glashow model, in which we show that dyon loops are a continuous deformation of (small) BPST instantons.

Cool! I highly recommend checking out the CERN animal shelter while you're there! It's a short walk from the antimatter factory. computer-animal-shelter.web.cern.ch/index.shtml

This is really nice, especially the physical picture! The coset construction has also recently been picked back up in the context of counting goldstones for spontaneously broken higher-form symmetries, e.g. arxiv.org/abs/2007.15901. Although I don't think there's a nice picture in the same way.

I guess your blue sky is still in beta.