@emilymbender and which reduces the connections in the brain

https://arxiv.org/abs/2506.08872

@Gargron I'm liking it

ICYMI, I dug into the delays that Apple inflicted on 75+% of web features announced for iOS at WWDC. Apple is fighting browser choice tooth-and-nail while demanding applause for dropping scraps from the $20BN/yr it skims from the web with a 90+% profit rate. Arrogance and hubris incarnate:

https://infrequently.org/2025/06/the-ghost-of-christmas-past/

@simon @jacob plus, do they include the amount spent on training or is this only inference time?

"There are protests, your majesty."

"Good," said the Emperor. "What are they protesting?"

"Grain prices. But why are protests good?"

"What would you call a society with no protests?"

"Happy, your majesty."

"I would call it oppressed. Prepare a report on their grievances."

#MicroFiction #TootFic #SmallStories

LLMs are designed to mimic the way people use language, first through pre-training on next-word prediction and then through additional rounds of redistribution of probability mass, called RLHF and the like.

It only becomes apparently mysterious when we do a lot of (reflexive, sure) interpretive work on the output and tell ourselves stories that involve the machines doing anything other than repeatedly calculating a likely next word.

Lots of folks fell for the "DOM is slow" marketing of certain frameworks, but DOM isn't slow. *Uncontrolled style read-back* is. But what if that wasn't a thing?

Looking for feedback on a new proposal to control layout thrashing here:

https://github.com/MicrosoftEdge/MSEdgeExplainers/blob/main/EventPhases/explainer.md

I had o4-mini write most of the plugin for me, based on an example existing plugin and this prompt. Transcript here: https://gist.github.com/simonw/4f545ecb347884d1d923dbc49550b8b0#response

@simon yeah 😔 I am being forced to look into AG Grid & infragistics' web component. Currently using sheet js for xlsx import/export and converting to JSON/CSV, which is what is usable with duckdb or LLMs.😒

@simon on a tangent ... Do you know of any good grid UI library that's performant, open source and handles large tables (say, 250 x 5000). For both display and editing, and good keyboard support?

Could be a wasm or js/css library. But should be able to bind with SQLite WASM or DuckDB WASM.

@djh interesting new term! would python wheel files help beat slopsquatting (via pip)? @simon

Anybody who's read or tried out debug-gym? 🤔🧐

I'm not a Python guy, so need to make do with others' impressions of it. 😄

https://arxiv.org/abs/2503.21557

@simon ok, so prompt injection continues to exist, while we now also have tool poisoning vulnerability added to the list. Why not memory overflow etc if npx is being used?

You can hardly find a studio the works of which are so thoughtful, kind, and intentional as Studio Ghibli. To stripmine that for its aesthetics, to take a piece of cardboard and paint it like food and say "See, doesn't this taste just as good?" is more than missing the point, it's barbaric, dystopian. It's an insult to life itself. #OpenAI

@simon yeah, we faced it back in the day when we coined the term Social CRM.

GTC felt more bullish than ever, but Nvidia’s challenges are piling up https://t.co/7VMqInGWvS

@simon waiting for onnx version to be available ... Might reduce VRAM requirement

@Gargron meta announced that on 1st April 2016, IIRC.

Just trying to get my time lines right. Nothing else. 😊

@Gargron was it before or after the announcement by meta about their FB messenger bots?

@kylelwiggers congratulations, Kyle! And all the best.