Dusty Pomerleau

@dpom.bsky.social

(he/him)

If you're designing a website for a restaurant, my tip would be: making it easy to find the menu is important, a 30-year breakdown of the founder's original vision and evolving philosophy less important.

I find it so strange when people write blog posts explaining TS intersection types using set theory. The whole reason people are confused by the name 'intersection types' is because they expected them to mirror set theory and they don't.

Been testing this 'sign-language-to-text' (SL2T) for almost a year & it's so awesome to see this out! Looked at this almost a decade ago & knew we'd get the tech & ML models there eventually (but super challenging). Talking w/ some Deaf folks, an equivalent of 'voice-to-text' was huge for them

Putting sign language AI into users’ hands

Introducing sign-language-to-text (SL2T), our breakthrough model powering new sign language features for Deaf and hard of hearing users.

deepmind.google

As a hobbyist, the thing I like least about LLMs is having non-deterministic interactions with computers. Many people take up programming as a hobby because communicating in code is precise/reproducible in a way that everyday human communication isn't. LLMs reintroduce natural language ambiguity.

Andrew Lilley Brinker@alilleybrinker.com · 2mo ago

IMO, software devs have long had a cultural undercurrent of self-perception as *artisans* doing highly-detailed small batch or bespoke work, which is both wrong as a matter of primary economic context and sets up the adherents for a bad mismatch with AI's utility for software engineering.

day two without internet. I sat in the rocker on the porch unmoving til sunset, watching the router light blink error like candlelight. the buffer wheel comes to me in dreams now pa says he's going in to town to get answers, but I saw him pack his rifle god willing, it will return in the morrow

The current approach to AI is so brutish and inelegant. If someone invented a clothes dryer that dries your clothes in 10 seconds, but it cost 100 USD worth of electricity to run it each time, you would not buy that product. The dryer company paying your electric bill to boost sales is unsustainable

For me, the most exciting thing about today is that we will learn whether Remco's approach to #tdf2026 has paid off. I expect Lipowitz to have his usual quiet consistency in the mountains, but depending on how others ride, quiet could be the new loud.

I don't see this going well. They will inevitably end up with an overwhelming mountain of data and turn to AI models for analysis, leading to an opaque process where nothing is ever really proven. It will be terrible for the sport if they sow doubt that they can't back up.

Cycling Weekly@cyclingweekly.com · 3mo ago

'Its potential contribution to anti-doping has remained largely unexplored' – trial to use power data to fight cheats in pro cycling www.cyclingweekly.com/racing/its-p...

We at TanStack just had a major attack against our Router packages. We have a postmortem out now explaining what happened and how we mitigated the response: tanstack.com/blog/npm-sup... No other TanStack packages outside of the Router monorepo were impacted

Postmortem: TanStack npm supply-chain compromise | TanStack Blog

On 2026-05-11, an attacker chained a pull_request_target Pwn Request, GitHub Actions cache poisoning across the fork↔base trust boundary, and OIDC token extraction from runner memory to publish 84 mal...

tanstack.com