As you might expect, the graph is dominated by things connected to primes, Fibonacci, etc. But the universe of completely disconnected sequences is the most interesting to me, lots of interesting substructure.

And now, 448764716674

My conclusion: no variable star seems to show any statistically significant numerical sequence that isn't as well-explained as a natural phenomena.

I knew this was going to be a unlikely search, but we gotta look from time to time!

@astrowright.bsky.social Figured you might find this interesting.

I run a bunch of processing on these stars in stages to try and filter out astrophysical or instrumental issues that could mimic a real signal, including the beating signals Hippke originally associated with the Kepler RR Lyrae they studied. I also try to extract binary sequences from modulation.

It’s easier to use an MCP server to retrieve articles with my agent than to use the web to search because the MCP path requires a login and is already vetted. It sucks but I think that’s the direction things are going.

Also convinced that referees *need* to use AI to keep up. Humans cannot keep up with the flux that's going to come from AI. And to be clear, I do mean quality AI-driven work here...it's creating results that are verifiably correct but at a rate so fast that we must design around it.

I know many in academia fear for their futures, but this is a very exciting time if you purely care about advances in understanding in many fields. When used appropriately, AI is a very powerful tool in this regard. In the right hands, it supercharges humans. Let's keep AI in the right hands.

For recent AI advances in math, many of them came from when AI dug deep into the literature, pulling ideas from papers written in the 2000s by authors that were often overlooked. I think these seeds planted by humans are, at least in the short term, the ripest fields to plow with AI at the moment.

P. J. Young wrote some papers in the late 70s that were very influential on my thought process (the "black tide" model of QSOs). He tragically died in the early 80s, and his influence noticeably waned after that. It was referenced occasionally, but the field gravitated to the active authors' ideas.

Because often their "threads of thought" hadn't gone anywhere, because the people who survived the academic gauntlet pushed their own ideas and left other peoples' behind. I think that often leaves the more correct theories on the table, unexplored.

And that's why I think it would have been difficult for AI to pick up this new (old) thread of thought. It's trained on the "current understanding" of things, which biases it towards those loud, active voices.

When I was in academia, I really enjoyed reading the papers of people who left the field.

When I left the field, some of my thought threads died with my departure. Academia is very much about who cheers the loudest for their ideas, and when you leave, your theories and understanding leave with you. My former competitors/colleagues pursued their own, different ideas. Which is fine!

But, I do believe AI was brought into our paper at the right phase: when humans had done ~80% of the work already. The novelty of the idea I think would have been difficult for the AI to come up with, particularly because the idea was a bit "frozen in time."