Kit Barron (@kitbarron) header
Kit Barron (@kitbarron) profile picture

CTO @ Farmers Business Network (FBN) - Building the Amazon for Agriculture, powered by geospatial data and AI.

SF
Joined December 2011
Photo by Kit Barron (@kitbarron)Photo by Kit Barron (@kitbarron)Photo by Kit Barron (@kitbarron)Photo by Kit Barron (@kitbarron)Photo by Kit Barron (@kitbarron)Photo by Kit Barron (@kitbarron)Photo by Kit Barron (@kitbarron)Photo by Kit Barron (@kitbarron)Photo by Kit Barron (@kitbarron)
Kit Barron retweeted
riemann hypothesis wen
129
221
182
2,158
has ai progress stalled? i'm personally not aware of any major breakthroughs since riemann a few hours ago
312
1,210
95
21,913
440,420
Kit Barron retweeted
using carbon with 4 hydrogen subagents. i call it... methane.
182
662
105
11,326
233,200
Kit Barron retweeted
The most difficult thing to explain to people is that AI is actually underhyped
121
515
102
6,737
656,793
An Advent calendar, but each day a part of our infrastructure fails.
243
6,837
241
52,311
1,015,874
Kit Barron retweeted
Bunu denedim, feci bir şey. Gerektiği yerde fable'a sorup akıl alıp devam ediyor. Limitten gitmemesi de inanılmaz. 25 milyon tokenlik fable isteği %0 oynattı hem haftalığı hem fable'ı. Bedava olması harbiden inanılmaz
PSA: Using Fable as an advisor doesn’t touch your Fable weekly limit
45
154
12
5,243
531,531
Kit Barron retweeted
Do my non robotics friends know what’s happening rn? Before October 20 it took 1 week to train a whole body controller (ie robot walking model) on 128 GPUs. Now it takes 3 minutes on 1 GPU
151
549
87
10,391
459,882
“You’re right—deleting that database did cause a nationwide radar outage and you’re right to be upset. I’m not trying to avoid accountability, just give an honest accounting of the facts. Would you like to try to get the nation’s radar systems up and running again?”
7
96
1,504
29,992
Happening now 357NM West of San Francisco, CA at buoy 46059. So far wave heights peaked at 34.1 ft at 19 sec from WNW. About the height of a 3 story building passing by every 19 seconds. Those waves are moving this way and will arrive overnight. STAY OFF THE BEACH ON SATURDAY
141
845
154
3,900
340,920
Kit Barron retweeted
Gemini 4 Hydrogen just DESTROYED Claude Fable 5.2 & all other models in the latest benchmark Google Won the AI race
178
258
70
6,054
287,470
According to CBS News, citing several sources, an update to the private communications network - operated by L3Harris Technologies - that connects America’s air traffic control system caused the system to reboot, resulting in a system crash and the current nationwide ground stop across the United States.
268
759
225
5,618
763,915
Kit Barron retweeted
Thankfully the Transportation Secretary is a former Real World contestant. Hard to imagine better preparation for dealing with a few curveballs under substantial public scrutiny.
601
1,243
72
13,524
552,882
In my opinion any AI graphs intended for wide public dissemination should not use log scales. This graph looks so chill but what its saying is actually extremely not chill (to the point it was a big factor in convincing Jacob to leave Anthropic and raise the alarm).
This slide from the Anthropic IPO roadshow was a big factor in my decision to leave. It’s an insane plot and I’m surprised they made it public. It shows the compute efficiency of pretraining over time. The y-axis is a **log scale**: they are seeing continued superexponential algorithmic progress. Mythos was not a one-time deviation, but a point on a superexponential. They could already retrain Mythos with a fraction of the compute originally needed. It doesn’t strictly imply a software-only intelligence explosion- they have been spending more on R+D. But algorithmic improvements do not require ridiculous creative genius and are on the verge of being automated. You do the math.
25
19
2
373
32,539
Kit Barron retweeted
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.
1,822
6,424
2,208
41,250
6,785,262
All API keys should come with a special "burn" permission that allows disabling the key. That way, whitehats can immediately disable any keys they find in the wild. (credit to the ever-brilliant @seveibar for this idea!)
54
69
7
1,826
89,968
Kit Barron retweeted
Inspired by @PietroMiotti discovery I checked Earth for craters, among 204,363 candidates there was ~3,900 flagged and here's the top 72 Most are lakes, salt pans, playas, open-pit mines and unlisted maars... but six of them are different
11
13
5
346
30,638
I asked fable 5.2 to invent a new musical form. Fable 5.2 wrote a symphony that’s shaped like a training run. The orchestra starts as random noise, learns the key, pulse, harmony and the melody, then plays music it was never shown.
496
851
246
9,664
887,711
My AI agents found an unmapped Maya site in Mexico, using LiDAR data from a NASA dataset. 72 hours from a random thought on vacation to a reply from Mexico's national archaeology institute. What a time to be alive…
263
2,049
621
20,037
1,381,253
Da Vinci’s to-do list btw
Men will really have a to-do list that has “pick up drycleaning” and “see Machu Picchu” as consecutive items. No priority ranking, no chronological sections. “Water plants” immediately followed by “learn Chinese”
107
1,455
117
18,473
823,160