I used to run some systems for MLB

by Charles, Austin, TX, Tuesday, October 14, 2025, 14:55 (10 hours, 28 minutes ago) @ Mike (bart)

that powered StatCast.

The stuff they're tracking goes way beyond what shows up on the broadcast. This was about a decade ago, but using the rocket-tracking tech, they'd track every player's movement the entire game. Now, the system wasn't perfect on its own, and required having some humans to sit there and watch the feed live, and when the system wasn't tracking players correctly (and which position they were), manual intervention was needed. But with that level of oversight (10 years ago), the system was able to spit out insane levels of field coverage efficiency for various defensive alignments. Of course, you could then check this compared to spraycharts for the hitters, and where they'd go etc...

I think the ability of computerized tools to analyze sports is already pretty high. Calling it AI I think is mostly using buzzwords for stuff that's already been in process for some time. The difference that might exist is the ability of LLMs to process real-time improvements to the methodology from non-techie coaches or analysts providing feedback that then can be converted to appropriate coding. In short, I think we're going to have the ability to rapidly improve the product iteration by reducing the number of "interpreters" inbetween. Now, instead of users using the product, and then subsequently providing feedback (or having people awkwardly watching them use the product and trying to discern where the issues were), the feedback loop can happen immediately, and coding changes can be either implemented on the fly, or developed for rapid integration by teams.

I also think there's just a huge disconnect between purpose-built systems utilizing leading edge technology, and how we interact with mass-market general-use systems at mass-market prices.