VLP Joaco (@JoacoIA) header
VLP Joaco (@JoacoIA) profile picture

vibe builder. Hidden Networks. IDP AI Operations Research Autonomous Systems 🦊vulpisone

Buenos Aires, Argentina
Joined February 2026
Photo by VLP Joaco (@JoacoIA)Photo by VLP Joaco (@JoacoIA)Photo by VLP Joaco (@JoacoIA)Photo by VLP Joaco (@JoacoIA)Photo by VLP Joaco (@JoacoIA)Photo by VLP Joaco (@JoacoIA)Photo by VLP Joaco (@JoacoIA)
1000 monkeys with 1000 writing machines eventually could write sheakspeare. 10.000 agents with 10.000 computers eventually could cure cancer. How many things were discovered by accident? just writing gibberish could eventually write something useful, something beautiful.
1
VLP Joaco retweeted
BREAKING 🇧🇷: Brazilian Real plunges to its lowest level vs. the U.S. Dollar since April 2025.
5
40
1
478
11,792
VLP Joaco retweeted
its so over
112
158
28
6,463
181,319
VLP Joaco retweeted
Some phone videos I took from the tower
14,153
30,523
2,502
301,611
8,961,542
VLP Joaco retweeted
This hits different every time you listen...
Kanye once said 🧵
12
193
7
2,179
130,643
Disculpa pero si viste el partido de Luzu vs Davo y los pibes te tienen que dormir
4
When you "hit a wall" in something you are trying to learn, it's typically just a massive debt of unlearned prerequisites that are finally being called due.
31
436
23
5,818
85,109
October 19, 2026, Jensen Huang: Jensen says he is all in on Open Models and is sticking his “neck out” because there are “movements” trying to undermine open models in favour of closed models, namely Anthropic and OpenAI. It would make you wonder?
128
1,188
100
4,745
206,552
VLP Joaco retweeted
This is so fucking performative and dumb.
The Denny's where Jensen Huang founded Nvidia in 1993 has removed their sign honoring his achievement.
60
1,118
7
29,008
417,491
I know what you need to do right now. Go download the weights of these 4 open models and put them on your disk. Right now, even if you cannot run the big ones yet. Nobody knows where any of this goes next. Qwen3.8-27B Qwen3.8 Flash-Next GLM-5.3-Flash DeepSeek V4.1 Flash 1017 Grab the abliterated quants too while every link still works. RIGHT NOW!
I am one of the people behind the 6,500 uncensored models the press now wants off Hugging Face. Transformer News counted over 6,500 abliterated, uncensored and heretic models on Hugging Face in July. Their story is that these models could help cyber attacks. Hugging Face removed 1 model marketed for offensive cyber, and the reporter asks why the rest are still up. A few of those are mine. I upload abliterated Qwen and GLM quants. My Qwen3.8 Flash-Next one got about 358,000 downloads in the last 30 days. Abliteration removes the refusal direction. That is all. The model does not learn hacking from it. It stops lecturing you when you ask about a chemistry exam, a pentest you are paid for, or a dark scene in a novel. The base models are already on millions of drives. Removing my repo removes nothing. The real problem is somewhere else. Hugging Face is where almost everyone gets weights. Nvidia agreed to buy it, Nvidia just lost the CEO who promised to keep it open, and now the press asks why uncensored models are still there. Download what you need. TODAY!!!
104
179
11
2,096
262,351
VLP Joaco retweeted
Local AI. Local compute. Open weights. Open source. Global collaboration.
681
1,503
217
16,910
391,698
there were signs
253
694
122
7,124
1,007,803
VLP Joaco retweeted
Some of my smartest friends at Huawei in Shenzhen have gone quiet recently… quite a few appear to be in Hangzhou (DeepSeek)… they can only WeChat their spouses and parents. From what I’ve heard, V5 is already in pretraining: ~2T parameter ultra-sparse MoE on Huawei Ascend 950/950DT, no NVIDIA (!!!). If it converges, it’ll be the first frontier model trained entirely on Chinese silicon.
99
213
66
2,657
841,829
Esta jugada es clara, you will own nothing and be happy.
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter. AI will transform every industry, power every company, and be built by every country. Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty. The world needs both frontier closed models and frontier open models. images.nvidia.com/pdf/Open-W…
18
nothing suspicious about the timing of this announcement, no sir, today is a perfectly regular day for an announcement such as this to drop.
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.
13
30
554
13,467
VLP Joaco 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.
1,813
6,408
2,192
41,157
6,645,296
Siguen pegándole a NVIDA de todos lados, ahora con el contrabando a china. Parece que salto la ficha que Jensen estaba operando para ambos lados.
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.
17
Por estas cosas se escribió Ahora soy Peor de Bad Bunny
a mí hermana le escribió hoy un chaboncito q gusta de ella y q ya se le declaro dos veces no doy más estallé
30
VLP Joaco retweeted
Jensen right now
88% OF 232 NVIDIA $NVDA EMPLOYEES SURVEYED WANT JENSEN HUANG TO STAY CEO
58
3,805
31
91,271
733,603
50.000 personas coreando tu nombre. Que locura este tipo
13 years ago, Messi and Ronaldo traded braces in El Clásico. We were spoiled. 23′ — Cristiano Ronaldo ⚽ 31′ — Lionel Messi ⚽ 61′ — Lionel Messi ⚽ 66′ — Cristiano Ronaldo ⚽
77