Sale my extra compute on this thing known as a market. Virtual Power Plants and Inference.
sam: local inference is a terrible allocation of scarce resources
“a lot of people want local inference, and somehow they assume that the problem is solved now because we have open-source models, we're going to be able to run them on our computers, and so we're not going to need all the stuff in the data centre
but from a global allocation perspective, actually it's quite terrible to have everybody, like, have a bunch of the world's limited supply of DRAM sitting in their own computer and being used five percent of the time
you want maximum utilisation in the cloud if you're trying to deal with this very supply-constrained environment
somehow many people's intuition is the opposite of that”
Tim Cash retweeted
Something I've been saying for a while, you don't have a sovereign industrial base if all the machines and robots inside it are imported.
Yesterday the White House made it official. America "relies on foreign imports for most types of robotics and industrial machinery," and that's now a national security threat.
This is the first critical step toward bringing robot and machine tool manufacturing back to the US.
The US invented these tools.
The first industrial robot, Unimate, went to work at a GM plant in New Jersey in 1961. CNC machining came out of MIT. Then we stopped building them and outsourced it all. Today none of the biggest industrial robot makers is American, and Korea runs 4x our robot density and besides Fanuc, most industrial robot OEMs have outsourced critical components to China.
@AMRwelding is building industrial robot arms designed and built in the US, on our own opensource controller.
Reindustrialization runs on robots. They should be built here too.
Industrial robotics and advanced manufacturing are foundational to America’s defense industrial base.
Securing domestic production capacity isn’t just an economic priority – it’s a national security imperative.
whitehouse.gov/presidential-…
Tim Cash retweeted
i was saying it over and over: it’s absolutely a national security threat - we are actively applying robotic automation in manufacturing plants, agriculture, warehousing, etc. but what would happen if the robots break and we can’t ship the parts?
@gearsdotco was founded to secure us with gears and gearboxes 🫡
Tim Cash retweeted
I'm hosting a dinner with the Global Head of Physical AI at NVIDIA Inception on Nov 2 in SF
Theme is Robotics in Manufacturing
We have a couple of seats open. If you're a founder or researcher building in this space and want to join, let me know!
Tim Cash retweeted
I'm happy to share that Mojo and MAX are now fully open - for development, extension, and commercial use! This includes the full Mojo programming language, but also the MAX graph compiler, MLRT runtime internals, and hardware abstraction logic.
Get your agents going!👇
Tim Cash retweeted
Rainmaker Rough Riders
Making more water to conserve our nation’s natural inheritance. I hope TR would be proud
Tim Cash retweeted
I wrote an update on what we've done at Terraform this year, and what's next (roadmap.terraformindustries.…) terraformindustries.wordpres…
It's a good day to make synthetic fuels.
Geospatial Intelligence is going to be big
Introducing terrio: geospatial data I/O in pure Rust!
terrio reads, writes, converts and reprojects geospatial rasters, vector data and point clouds, with no C dependencies. No GDAL or PROJ to install, nothing to compile: it builds anywhere Rust does, including WebAssembly, so the same engine runs on the command line, in Python, and in your web browser.
What it does
- Rasters: GeoTIFF and COG, Zarr and GeoZarr, netCDF, HDF5, PMTiles, MBTiles and more, read in place from the cloud, so only the bytes needed are fetched.
- Vectors: GeoPackage, Shapefile, GeoJSON, FlatGeobuf, GeoParquet, OpenStreetMap PBF and vector tiles, with filters, field selection and reprojection.
- Point clouds: LAS, LAZ, COPC, EPT, PLY and E57, with PDAL-style filters (ground classification, noise, heights above ground) and gridding into DSMs and DTMs. Files larger than memory are converted in chunks.
- GDAL-style processing: translate, warp, mosaic, rasterize, polygonize, terrain analysis, raster calculator, proximity and sieve.
- Remote data: http(s), S3, Google Cloud Storage, Azure, zip archives and STAC searches as inputs.
- Accurate: coordinate transformations agree with GDAL's within a centimetre for all 6,354 EPSG CRSs GDAL transforms, and outputs are tested against GDAL on real-world data.
- Fast: in our benchmarks, warping, mosaicking and writing COGs run several times faster than the GDAL utilities.
- For AI agents: terrio mcp is an MCP server, so Claude Code, Claude Desktop, Cursor or VS Code can run terrio's commands on your data.
Get it
- Command line (Linux, macOS): curl -LsSf terrio.opengeos.org/install.… | sh
- Python: pip install terrio (with rasterio- and xarray-style APIs)
- JavaScript/WebAssembly: npm install terrio-js
- Rust: cargo add terrio
Try it out
- Playground in your browser: terrio.opengeos.org/playgrou…
- Documentation: terrio.opengeos.org
- GitHub: github.com/opengeos/terrio
#GIS #Rust #OpenSource #Geospatial #Python #LiDAR #CloudNative
Never a better time to learn math
Starting Friday, Oct 23: a free online course on advanced category theory.
Kan extensions, (co)ends, profunctors, monads, enriched categories and more. Proof-based, Fridays 17:00 CET, recorded.
Zoom link & updates: t.me/category_theory_communi…
#CategoryTheory #Math
Tim Cash retweeted
Just picked up some more welding gear for the Sunday welding classes. We are on for tomorrow at 10:00am in Hawthorne if anyone wants to stop by. Shoot me a DM for the address.
Good fun again yesterday at Sunday welding open mat. Another 5 first time welders in the books. @cambaughn @lucasme_12 @MayanBear @MystCold @peterpiontek_ @c0nst
Tim Cash retweeted
My dear friends, I’d like to show you my latest toy project
Introducing Handwired: turn your regular JavaScript code into a neural net!
- no training
- exact outputs vs the JS program
- fully differentiable
All runnable in a browser, without GPU
Tim Cash retweeted
We're going to need a LOT more energy
(1/6) From @GoogleDeepmind’s AlphaGo to AlphaZero, we saw machines find moves that reached beyond what we’d imagined for them. No one taught them those strategies. They emerged from reinforcement learning: playing, failing, and trying again under a single binary task reward for success or failure. Inspired by them, we couldn’t help but wonder: what might a robot discover if we gave it that same freedom to learn?
That question led us to AlphaNIST. Starting tabula rasa, with that same binary reward plus only a simple control cost, a single policy solves all 20 @NIST assembly tasks at a 98.2% success rate in simulation. More surprising still, AlphaNIST can complete 10 different tasks in a row, receiving that reward only once, after all 10 are done.
Just as AlphaGo’s strategies surprised its creators, AlphaNIST discovered dexterous strategies we couldn’t have come up with ourselves.
Paper: arxiv.org/abs/2610.13354
Website: alpha-nist.github.io