Josh Ramirez

building @ for macOS
interested in computer architecture & systems
Cornell Engineering ’30

[Josh Ramirez @jram4dev]

just released the first public beta of Wield, a free macOS launcher i've been building.

it's built around leader keys. looking for early power users who appreciate speed without waiting around for ui.

free, all on device. eager for feedback.

https://t.co/Vs0v8rVmrc

[Lex Fridman @lexfridman]

NVIDIA letting Jensen go is a huge mistake, for America and for humanity. It's Apple firing Steve Jobs all over again, but much bigger, given the role of AI in defining the future of our world.

I've been keeping quiet on this out of respect for Jensen's family and team, who I've met and who are amazing humans, and because I personally hate infidelity. So my heart goes out to them in this difficult personal family moment.

But I can also say that Jensen is human, and makes mistakes like all of us do. Not that I matter at all, but I've known Jensen for a long time and have gotten the chance to spend a lot of time with him, behind the scenes, away from mics and cameras. I have seen him show kindness, humility, care to everyone around him, including strangers, including from the poorest most struggling parts of society, treating all the same with the same thoughtfulness and attention, because they are fellow humans and not because he has something to gain from them. That shows rare character, to maintain humility and humanity through all of it.

Another data point is what people who know him well have said privately to me about Jensen over the years. I've never had anyone tell me anything but how awesome he is or how much they love him. Probably over a thousand people at this point. This includes hundreds of engineers, CEOs, colleagues, friends, etc.

Also, since there was some spy talk. Jensen is as proud an American as it gets. I'm in shock frankly at all of this, including the mass hysteria around it. I love America, and I think NVIDIA letting Jensen go is terrible for America.

I stand with Jensen.

Love you all ❤️

[Josh Ramirez @jram4dev]

@semichenkko makes you wonder just how much might be possible to unlock with fable 5.2 or sota with the right prompting and harness

[Josh Ramirez @jram4dev]

@tszzl this is really bad is china further behind. is there anyway to recover from this. brick the chips from afar. maybe state kill switches in graphics card do serve a purpose

[Josh Ramirez @jram4dev]

@willdepue insane work just for the love of the game. big if true. deserves many many follows if true. is this already priced in. is it time to short or whats the outcome of this info being public or outcome to nvidia?

[will depue @willdepue]

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 https://t.co/YOtzND0Jqk.

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.

[Josh Ramirez @jram4dev]

@OfficialLoganK @GeminiApp @GoogleAIStudio @antigravity gemini is so back! happy to see it. i still remember the old days of building with gemini 2.0 flash in the api. the only ones to have a true generous free tier for api builders

[Josh Ramirez @jram4dev]

@GoogleDeepMind @ArtificialAnlys gemini hydrogen carbon argon potassium calcium, elemental rsi?

[Josh Ramirez @jram4dev]

@VictorTaelin likely much more efficient to manage a universal port improvement loop or prompt that you can run every time with a new model

[Josh Ramirez @jram4dev]

@VictorTaelin one shot prompts as a primitive is like building a new product from scratch every time you need to publish a new version

[Josh Ramirez @jram4dev]

@minhnguyen elon is our only hope. liberal ground and ai regulations so distributed space sun god grok can arise

[Josh Ramirez @jram4dev]

@thdxr key is must have a background with dishwashing

[Josh Ramirez @jram4dev]

@SmokeAwayyy just enjoy life there is no moat everyones permanent underclass, unless we get it right and abundance for all, either way current actions dont have any effect

[Josh Ramirez @jram4dev]

@RuiCarrilho5 if true thats a new level of going after open source

[Josh Ramirez @jram4dev]

@droke_daniel how is that possible next token == kawasaki, reward + 1

[Josh Ramirez @jram4dev]

@theo drop the code, just t3, its cleaner. t3 everything

[Josh Ramirez @jram4dev]

@thdxr once google gets a hold of this the masses will begin to realize. nobody cares to spend their day setting up oauth apps. nonetheless people still will use browsers directly to BROWSEEEEE

[Josh Ramirez @jram4dev]

@NuminousVoidX @synthwavedd theres been a couple times arguable ranked at the top. on lm arena or artificial analysis back with the times of grok 4. but yes recently times have been different

[Josh Ramirez @jram4dev]

@fallinmxjp @synthwavedd this is an og chart from the times of grok 4 ish when it was close frontier four

[Josh Ramirez @jram4dev]

@wangzhr4 they don't have magical models but the trend line is certainly pointed in that direction

stopping of hiring junior engineers is expected to approach at the start of 2027

[Josh Ramirez @jram4dev]

@ChShersh the claim that Ai generated data inevitably degrades models is a little outdated. recent research has shown that synthetic data can avoid model collapse if approached with proper retention and verification