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)
*We* let the dogs out! WOOF-NAM select imagery now available on our model page. Forecast soundings for the model hopefully coming tomorrow. -Gilbert
youtube.com/watch?v=ojULkWEU…
found out on the wifi-less plane while grepping through some 4 years worth of md files that the bash and regex part of my brain had atrophied but the python script part of my brain was still in tact
Last December, 24 hardware founders visited 15 factories with us in one week.
The next group goes to Shenzhen Dec 6 to 11. Up to 20 founders, with introductions set up before you land.
Free to apply, pay only if accepted. Link in the reply.
We gave AI one week to build a giant drone
It helped design it, order the parts, and do final assembly
Then we delivered its first payload to @jimbelosic
I have a major result to share.
I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China.
This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model.
Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves.
I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems.
I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used.
It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20.
These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs.
I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo….
There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
One of the absolute best examples of ML can be applied in weather to solve extremely tough problems (snowfall) from one of the smartest people I know, excited to this and excited for future maps!
As part of my masters thesis, I'm working on a smarter machine-learning snow-to-liquid ratio. It clearly shows that this inland snow event will be restricted to higher terrain, and even there, ratios will average well below 10:1.
My dear friends, I’ve got 3 pieces of news
First off, I’ve retired from @midjourney after 3.5 years. It’s been an honor; we've pulled off great projects, and a few open-sourced pieces like Pretext & Handwired have been well-received. I’d do it again with the team, but they’re in a great stable place now and I planned to take >6 months off
Second, unfortunately the words spread. Life comes at you fast and (this is a crazy story best told in private) I was taken hostage by the good folks at @coreauto. I started last week. Retirement will have to wait for another few years
Third, I’ve figured out a few pieces of technology I wanted to exist for a while now. Some more like Pretext, some more like Handwired, and a mysterious new thing (pic attached. First one to guess right gets a repost). Stay tuned as we start training these at Core Auto.
We’ll show the world what AI can truly achieve in the hands of kind, capable & grounded folks who choose to look up.
P.S. we’re hiring one (1) backend engineer. Only requirement is showing us your previous project’s dependencies list