paulirish

`latentflip-test` is from the same fellow who did the "What the heck is the event loop anyway?" JSConf EU talk that many have seen. https://youtu.be/8aGhZQkoFbQ

Beyond all the wasm/webgpu approaches other folks have linked (mostly in the transformers.js ecosystem), there's been a standardized API brewing since 2019: https://webmachinelearning.github.io/webnn-intro/

Demos here: https://webmachinelearning.github.io/webnn-samples/ I'm not sure any of them allow you to select a model file from disk, but that should be entirely straightforward.

Turns out, the original TiddlyWiki used a java jar to handle the file persistence. (I remember it being so magically automatic, but recently investigated how it was done)

The comment thread: https://m.slashdot.org/thread/65583466

So glad to find some like-minded folk who care about this! Nice post, Shane.

Thanks for calling out my research on this. Really glad you found it.

https://github.com/paulirish/lite-youtube-embed/pull/167 has my solution. Basically we only attempt the hqdefault.jpg and the sddefault.webp; no other ones. Based on my research this is (probably) the best solution that optimizes for both latency and coverage. Though, I could see switching the order of two in the future.

Pretty sure the oEmbed solutions folks are mentioning are only reasonable if you're doing this work server-side/eagerly. Obviously that changes things quite a bit.

You can install webapps "as an app" which solves that problem... its own icon in the dock, cmd-tabable, etc. In Chrome this is under the "Cast, save, share" menu.

We're working on this within Chrome DevTools, similar to an MCP server with the signals you're describing.

An approach I've been using recently is to rely on pyannote/tinydiarize only for the speaker_turn timestamps, but prefer the larger model (or in this case YT's autotranscript) for the actual text.

Same, yup. OP is indeed already using yt-dlp for the video download. (Then Whisper for transcribing, Ollama/lmstudio/OpenAI for summarizing)

Can also just fetch the subs already in YouTube rather than retranscribing. eg:

yt-dlp --write-auto-subs --skip-download "https://www.youtube.com/watch?v=7xTGNNLPyMI"

It's coming from zcbenz who created Electron among others https://zcbenz.com/ Nice.

Ah! My bad. thx

It would lend credibility to publish your system prompt.

That mighta been Paul Lamere, but.. Then again I did fuck around with Echo Nest stuff forever ago.

You build from the release branches yeah? So, right now that'd be the 7204 branch. https://chromiumdash.appspot.com/releases?platform=Mac

If the target is a GitHub readme, then you can embed a video directly. eg https://github.com/paulirish/git-recent#readme

That said, OP's SVG trick may be a smarter choice if the content is a terminal capture.

Here's the top 500 servers in the dataset: https://pastebin.com/pANc3WkT

What does one of the GPU profiling reports look like?

Edit: oh is it this? https://youtu.be/b-yh3FFpSX8?t=28

> Gemini 2.5 Pro now ranks #1 on the WebDev Arena leaderboard

It'd make sense to rename WebDev Arena to React/Tailwind Arena. Its system prompt requires [1] those technologies and the entire tool breaks when requesting vanilla JS or other frameworks. The second-order implications of models competing on this narrow definition of webdev are rather troublesome.

[1] https://blog.lmarena.ai/blog/2025/webdev-arena/#:~:text=PROM...

First Contentful Paint at 12.0 seconds here, and I'm on fiber.

Going back a lil further to 2011… We had a lil fight[1][2] about the value of using perfect HTML semantics in the community. The outline algorithm was the big defense for <section>, but its problems were clear then: https://www.smashingmagazine.com/2011/11/pursuing-semantic-v...

[1] https://www.smashingmagazine.com/2011/11/our-pointless-pursu...

[2] https://www.smashingmagazine.com/2011/11/pursuing-semantic-v...

In the same vein, I've long had a soft-spot for the JS-enhanced https://pancake-charts.surge.sh/ (developed by NYT graphics team and used for the covid charts).

Impressive work! Their main demo there pays some homage to Mike Bostock's d3 transitions showcase from 2011: https://mbostock.github.io/d3/talk/20111116/transitions.html (click to start)

Vaguely related and also fun: https://www.koalastothemax.com/ (2011)

I love playful interaction too, but you can get all this for free: SVGs are natively interactive.

Here's your demo in 45 lines of SVG and CSS: https://codepen.io/paulirish/pen/gbOZXJp

A recent CCC talk covered the current state of EUV lithography (really well): https://youtu.be/NdppYYfQJgg?t=1393 This timecode has rendering of the ASML machine in action. (edit: the same one madspindel posted, oops! The rest of the talk def worth a watch)

(In case it's not obvious, parent poster is author of ripgrep.)

Let's add a "read comment" button to HN to make this discussion easier.

UX whine: Why do I have to click "Check compatibility"? After type and RAM, you instantly know all the models. Just list the compatible ones!

I went digging in W3C archives to find relevant discussions. There are 4 reserved target names: _blank, _parent, _self, _top.

- 1995 Sept https://web.archive.org/web/19990202141025/http://home.mcom.... Netscape Navigator 2.0a2 release notes including rollout of TARGET.

- 1995 Sept https://lists.w3.org/Archives/Public/www-html/1995Sep/0034.h... First definition of them; spec text provided by "the Netscape Navigator marketing guy".

- 1997 Jan https://www.w3.org/TR/2018/SPSD-html32-20180315/ HTML 3.2 published without frame target names specced.

- 1997 Aug https://lists.w3.org/Archives/Public/www-html/1997Aug/0010.h... Roles of these target names clarified

- 1997 Dec https://www.w3.org/TR/html401/types.html#h-6.16 HTML 4.0 published with target names.

I can't find any discussion about the choice of character. My guess is that discussion was internal to Netscape as they shipped _then_ specced link target names.