Vince Buffalo (@vsbuffalo) 's Twitter Profile
Vince Buffalo

@vsbuffalo

Scientist at IDM. ♥s biology, running, probability+statistics, 🦀, fly fishing, backpacking. Author of book Bioinformatics Data Skills. All views my own.

ID: 62183077

linkhttp://vincebuffalo.com calendar_today02-08-2009 03:11:08

27,27K Tweet

15,15K Followers

2,2K Following

Prachee Avasthi (@pracheeac) 's Twitter Profile Photo

Started a substack with my favorite person to argue and solve problems with, Michael Eisen. If you love science as much as we do and want to see it reach its potential, come follow along and join the conversation Our first post below 👇

Sawyer Merritt (@sawyermerritt) 's Twitter Profile Photo

Waymo in a new blog post: "We conducted a comprehensive study using Waymo’s internal dataset. Spanning 500,000 hours of driving, it is significantly larger than any dataset used in previous scaling studies in the AV domain. Our study uncovered the following: • Similar to LLMs,

Waymo in a new blog post: "We conducted a comprehensive study using Waymo’s internal dataset. Spanning 500,000 hours of driving, it is significantly larger than any dataset used in previous scaling studies in the AV domain.

Our study uncovered the following: 
• Similar to LLMs,
Armin Ronacher ⇌ (@mitsuhiko) 's Twitter Profile Photo

Not so hot take: if you use Claude Code, most of y’all’s MCP servers could be a shell script. Easier to maintain and faster and Claude uses just as well if not better.

Not so hot take: if you use Claude Code, most of y’all’s MCP servers could be a shell script. Easier to maintain and faster and Claude uses just as well if not better.
Gappy (Giuseppe Paleologo) (@__paleologo) 's Twitter Profile Photo

Roy Carrilho X must have recommended your post because I have opined about this strand of research in the past. I will take the other side. 1. the stock/flow construct that Meadows emphasizes is just one of many in mathematical modeling and is only adequate in a small set of applications. 2.

Gappy (Giuseppe Paleologo) (@__paleologo) 's Twitter Profile Photo

At the turn of the 1960s there was a flurry of cross-disciplinary meta-theories: Systems Science (Forrester), Cybernetics (Wiener). They petered out. Why and how remains relatively under-studied.

Nathan Lambert (@natolambert) 's Twitter Profile Photo

Language models may not get us to “superintelligence”, but they’re definitely already powerful enough to create very socially and economically destabilizing technologies so it doesn’t really matter. We have to make the world ready for that.

Liv (@livgorton) 's Twitter Profile Photo

sad to report that it turns out python environment management didn't have to suck this whole time and UV is actually great

Vince Buffalo (@vsbuffalo) 's Twitter Profile Photo

This is very cool; while it’s a seemingly minor convenience, this is the sort of thing that saves hundreds of thousands of researcher-hours in the US a year. Good tooling is so important.

Saloni (@salonium) 's Twitter Profile Photo

New piece by me! Measles is often seen as a routine childhood illness: a fever, rashes, recovery. But complications are common. Even when it doesn’t kill, measles causes lasting damage. It depletes immune cells, making people vulnerable to other infections for months or years.

New piece by me!

Measles is often seen as a routine childhood illness: a fever, rashes, recovery. But complications are common.

Even when it doesn’t kill, measles causes lasting damage. It depletes immune cells, making people vulnerable to other infections for months or years.
Vince Buffalo (@vsbuffalo) 's Twitter Profile Photo

I can’t explain why, but there’s a phenomenon called bagel debt. When you crave a bagel, and you eat a subpar bagel, you’re in -1 bagel debt. Finally eating a good bagel only brings you back to 0. Another bagel mistake? Straight to -2.

Heng Li (@lh3lh3) 's Twitter Profile Photo

Preprint on "Improving spliced alignment by modeling splice sites with deep learning". It describes minisplice for modeling splice signals. Minimap2 and miniprot now optionally use the predicted scores to improve spliced alignment. arxiv.org/abs/2506.12986

Preprint on "Improving spliced alignment by modeling splice sites with deep learning". It describes minisplice for modeling splice signals. Minimap2 and miniprot now optionally use the predicted scores to improve spliced alignment.
arxiv.org/abs/2506.12986
Vince Buffalo (@vsbuffalo) 's Twitter Profile Photo

It’s fascinating that languages are trending back to static typing. JavaScript→TypeScript, Python’s type hints + we’re getting *multiple* new Rust-based type checkers for Python (Astral’s ty and Meta’s Pyrefly). I’m already a convert after Rust; but why the overall trend?

Vince Buffalo (@vsbuffalo) 's Twitter Profile Photo

Great article and story — even though I disagree with the conclusion that AI won’t revolutionize science. It’s certainly overhyped and underperforms in many domains; the revolutionary part will be more about lowering costs for mundane but important tasks.

𝕐 (@nomad421) 's Twitter Profile Photo

bzip2 officially switching from C to Rust. Not only is Rust a nicer language for the implementation, and not only is it safer, but the Rust implementation is also *faster* than the C implementation... C-niles will have to up their doses of copium. trifectatech.org/blog/bzip2-cra…

Vince Buffalo (@vsbuffalo) 's Twitter Profile Photo

The scale of RLHF they’re probably doing on coding agents must be nuts. Think millions of developers constantly fine tuning an LLM’s code decision throughout the day. Maybe we will be the last generation to hand write code.