Stephen Sutton-Brown

yesss

it leaked??

SABR Analytics Conference logo

Congratulations to all winners of the 2026 SABR Analytics Conference Research Awards!

@michaelrosen.bsky.social of @fangraphs.com
@richstaff.bsky.social of @defector.com
@rjandersonwrites.com of CBS Sports
@depstein1983.bsky.social of @baseballprospectus.com

sabr.org/latest/rosen...

YouTube video by Lance Brozdowski Does AI Make Me A Better Baseball Analyst?

Cool video from @lancebroz.bsky.social on using AI tools to create baseball apps
youtu.be/rTBJjHosIgg?...

was about to say the same, i don't see them showing a measurement here like in what was shown above for mlb

no opinion there, i can't remember the last time i watched a pro tennis match. i assume they have similar discussions though

& tbc i agree that i don't think the buffer zone is the right approach, but i think MLB & broadcast teams should think carefully about presentation & be careful not to present the system as being more accurate than it is, regardless if more precise than a human. arguably the "0.1 inch" isn't best

i think that kinda talks past things here tho. to me it's a common question of getting buy in & establishing legitimacy when handing judgement off to automated system. we see similar debates in automated driving, it's not just about precision

the issue with this though is all it effectively does is expand the zone by that amount. you still will end up with pitches like this that are a hair outside that boundary + margin of error

very important app development activities happening

one thing i've come to appreciate since becoming a parent is 100 piece puzzles are a lot of fun for all ages

living her best life

movement plots inspired by classic car gauges

relatedly assuming you're using claude to help like you mentioned in the intro i've found it to be soooo much better at coding up calculations and data manipulations than coding up plots. i feel like if i vaguely describe code or math i want it's fine, but i have to hold its hand for graphics

Logan Webb’s Backwards Sweeper Webb’s sweeper popped against lefties and fell flat against righties in 2025. Why didn’t his overall results follow the same pattern?

cool use of their new tools
blogs.fangraphs.com/logan-webbs-...

Personally never understood this concern. Who cares if it is a mess? The tariffs were straight-up theft. Do we look the other way on robberies because the criminal was too successful?

I hope it *is* tremendously painful and makes the government never want to try this again.

a glimpse at a powerpoint slide i’m making

Introducing the FanGraphs Lab The Lab is a collaboration between the editorial team and the engineering team here at FanGraphs, a joint effort to create more ways to sort through and visualize the huge crush of data that pervades…

this is so cool, well done by the FG team blogs.fangraphs.com/introducing-...

webbing up the bad guys!

claude's just like me

I still think the internet is the best comp for AI. It was a bubble that popped, but the underlying tech turned out to be real. It enabled a bunch of psychosis and slop, but also some genuinely cool stuff. Massive labor market disruptions, but employment levels look about the same.

meet Mug Guy

can't believe they cracked this code-

i've been going through and updating my passwords on all of my accounts, and good lord i forget just how many websites i have logins for. this is like a week long task

Graphs of sensitivity, showing LLMs outperforming humans

We coded our ~100k articles using LLMs. Should you believe them? To answer this, we benchmarked 4 human RAs against 3 LLMs on their ability to recover ground truth article data. Details in the paper and appendices, but the LLMs did well and handily beat the highly trained humans.

Alphafold continues to be my biggest ex. for drug discovery. Personal experience + keeping up w/ latest evals makes me confident the impact on scientific discovery generally will be huge (even if it doesn't come up with better ideas it could streamline processes). And stuff like SB 53 gives me hope.

deleted my last response b/c it was too snarky & dismissive. At the end of the day, I get the skepticism. But I think the capabilities & potential benefits are real & huge, and I think shaping its impacts to reap rewards while mitigating risks is worth pursuing, even if it partially fails.

A legitimately useful exercise here would be to ask Opus 4.6 or GPT 5.2 Pro to Steelman the counter argument to you. I think it would both provide a strong response and in doing so speak to how good the frontier model capabilities have gotten

Skepticism is fine but the alternative Ed et. al. are selling is "deny what it does, claim it's a bubble, sit and let it blow itself up." That's not a plan. If you think there's a chance Ed's wrong & this will be hugely impactful and disruptive then we should work toward doing something about it