reduce entropy
San Francisco, USA
Joined August 2014
- Tweets3.3K
- Following445
- Followers442
- Likes28.8K
Pinned Tweet
Recast launched! So happy to be part of it! It’s deeply motivating to be building something that helps get more water where it’s needed
I’ve been working primarily on the Texas and New Mexico operations software, and I also get to do physics and stretch my brain. New Mexico’s raw landscapes have a special place in my heart because it was the first US state I stayed in for an extended period. Texas is a world unto itself with a rich story and a track record for invention, engineering, and getting things done. I admire it greatly and look forward to every visit!
Like Texas, Recast is a wonderfully creative project with optimistic, unwavering resolve and a thirst for building things that work
In the past 6 months we’ve produced a cheaper, more efficient seeding material, trained a hyper-resolution AI weather model, and operated 100+ cloud seeding flights for the governments of Texas and New Mexico.
Recast Systems is coming out of stealth. Here’s some of the research we’ve been working on. (1/n)
Stephen Fay retweeted
Bacteria should have shared the 2025 chemistry Nobel Prize for metal-organic frameworks!
It’s been known for a while that there are “cable bacteria” that extrude conductive filaments up to 1 cm long, which penetrate sediments on the sea floor and allow them to draw electrons …
ALT A bacterial filament with a ring of 32 conductive fibres, each of which contains multiple intertwined nanoribbons. The ribbon is shown to comprise a series of Nickel atoms joined by Bis-Dithiolene linkers, made from sulphur and carbon.
Stephen Fay retweeted
i think this is real and a big deal. as karpathy said, the ugliest part of the transformer is the tokenizer
I've always believed that intelligence should come from simple, general building blocks that learn their own abstractions directly from raw data; in other words, the bitter lesson. I think we're finally seeing a real step towards AI that can learn from all data.
Language modeling on raw bytes is a microcosm of this ideal, and to me it's a central problem in AI: if we can't build LLMs that learn end-to-end from bytes, how will we handle the rest of the world's data that we understand even less?
Last year we introduced hierarchical networks (H-Nets) as a first step. Our new version finds up to 3-5x efficiency gains in controlled scaling law experiments, and at larger scale matches an open-source model trained on 11T tokens with ~10x less data and compute. The posts below walk through what we found and why we think it works.
Stephen Fay retweeted
Another Weather + GIS for Friday
All of my weather maps color tables + another 1000 --> forecasteagle.com/ramps
Use, share, have your bot produce beautiful weather maps. Stop the slop.
Stephen Fay retweeted
do you understand what nuclear transition's low sensitivity to environment really means? It means you can use it in SOLID STATE. No more fancy fragile optical lattice acrobatics, no vapor cells, you can have 229Th in a host crystal like CaF2, and it can be tiny.
that's what they did, picture from JILA and TU Wien folks who tested the clock transition linewidth vs doping concentration earlier this year. It shall start climbing this ladder when you echo and rephase the inhomogeneous broadening. Time to make higher power 148 nm pulses and better detectors.
OMG a team from China made a thorium 229 nuclear clock that’s amazing!!
The current SOTA timekeeping standard is atomic clocks, but nuclear clocks would be way more stable. This current clock is below SOTA but Th229 could well be the new SOTA very soon.
Nuclear clocks could be better because the nuclear transition Q is higher by 10x and it’s also shielded from the environment because it’s in the nucleus.
Th229 is rare because it has a nuclear transition that is excitable by lasers - usually these are way too high energy to do anything with so making a nuclear clock was considered impossible.
Atomic clock transitions are usually microwave/optical. The Th229 nuclear transition is ultraviolet, which we can make a laser for with a lot of work - the UV laser is one of the big technical breakthroughs here actually.
This is so cool!!!!
A nuclear clocked synchronised to Th229, Huang et al. nature.com/articles/s41586-0…
technical debt, in software, is now literally convertible to cash, if the code is sloppy it'll take more tokens to add a feature or improve it than a well designed code
Stephen Fay retweeted
All my work online for a guaranteed 100 years.
It's a new service being offered by WordPress. They tried it out on me. I think it's great.
Explained here: wordpress.com/blog/2025/02/1…
This is such a gift to the world. All of @stewartbrand’s work: writing, festival shenanigans, & 50,000(!) photos, dating back to the 50s: stewartbrand.org/
(Where else could you find Marshall McLuhan, Brian Eno, and Marlon Brando in one place!?)
Stephen Fay retweeted
“there aren’t any cryptography results”
“there aren’t any cryptography results”
Stephen Fay retweeted
I have spent far too much time on the Github integrations in T3 Code the last few days.
Cut rate limit usage by over 75% and made monitoring behaviors much more reliable. Julius moved us off the CLI for API calls we can do more efficiently directly. We rotate between the GQL and rest APIs depending on which uses your rate limits most efficiently (with fallbacks).
This was a ton of work that will largely go unnoticed, but it's the type of stuff I love. Had to do like 30 implementations of the PR watcher before I got it to a good spot.
On one hand, I'm thankful the details can be seen by those who dig into the source code. On the other, feels weird that everyone else will just copy our work and never have to think so much about this.
Yess!! Have been feinding for that auto resume feature!!
Things included in this PR:
- Pi support.
- Auto-resume when limits reset
- T3 Code MCP (create, launch, message, wait on, read, search and interrupt threads)
- `delegate_task` lets an agent start child agents on any provider or model
- ACP Registry: add any registry agent (Devin, Cline, Kimi, Droid, and more)
- Cursor runs on the official Cursor SDK, not the (broken) CLI
- OpenCode 2 support
- Agents can rename threads, regenerate titles, link PRs and settle threads
- New MCP tools for worktree handoff, and for listing and closing previews
- Proper server-side queueing and steering
- Thread forking
- Switch provider or model in the middle of a thread
- Native subagents show as child threads in a lineage view, with their model, status and history
- Use `@` with a thread name, or drag a thread from the sidebar, to attach it as context
- Scheduled tasks
- Tons more I'm forgetting
Pumped this landed. We'll have a lot of cleanup to do to get it perfect, so expect lots of updates throughout the next few days :)
