nylonstrung
On the note of libghostty it does get me very excited about the new multiplexer thing Hashimoto is building with it
It has WASM support via Pyodide so people can use the notebooks without a python server or toolchain
It similarly offers the combo of SQL and dataframes, built in dataviz and UI widgets, great LLM integration and essentially everything else one would want from either a notebook or a BI tool
Robert Gordon pointed out that the 90s productivity gains attributable to IT where overwhelmingly attributable to the manufacturing of computer hardware itself and secondarily the computerization of retail logistics
Daron Acemoglu's prediction is that Total Factor Productivity gains from AI will be a <0.55% increase over the next 10 years
https://web.archive.org/web/20230405154054/https://www.nytim... https://web.archive.org/web/20230405154056/https://www.nytim...
Embedded and industrial computing however have had a profound impact
I see tons of devs fretting about how only Opus is "smart" enough for the React CRUD web app they are building even though Sonnet and Chinese models are demonstrably equivalent for most real world developer use cases
"What a country needs to develop is discipline more than democracy" - Lee Kwan Yue
Many French would have you believe women wearing Hijab should be banned lest it destroy their "western culture" when even the most Islamophobic Americans don't care
A 5% Muslim population in UK is supposedly threatening to send them back to the dark ages when Singapore is at 15% with codified multicultural guarantees and somehow annihilates them in every measure of modernity and development
The last release was in July 2025
I uninstalled it specifically because it was slowing down my updates and builds
It's conceptually interesting but I don't know it is instructive beyond simple projects
If the strongest "evidence" is anecdotal reports that pro-Russia posting on Reddit decreased that seems more like Redditor confirmation bias and delusion than proof of a propaganda bot farm co-located with Yandex
Are they implying that chat history of a single person is being taken an used to skew post-training results?
Why would it only be for Navier-Stokes that we see this having an effect
There are 1 million times more prompts being processed for stuff like furry roleplay than unsolved scientific problems
It's very much a game of team coordination where the power curve is exponential but also can have huge turnarounds since buildings are fragile but drive the economy entirely
So naturally you have to know how to play your specific role or it can throw the game for everyone else
It's not casual like Command and Conquer or something, but nor is it incredibly sweaty like StarCraft or DOTA
This is one reason I'm so bullish on Lean4 as a general programming language
It's now essentially proven that expressive, strong type systems like those of Rust present an advantage for LLM assisted coding
Lean's type system, in which types are first class, fully programmable objects takes it so much further. Particularly when it comes to embracing dependant types
If you're gonna vibecode based on a decomp of a native desktop app you might as well make something with Rust and WASM/WebGPU that performs well and feels native