All of this content was scripted, planned, prompted, generated, and cut by agents. These are concept series in various states of development. Full episodes and some feature length movies coming soon.
Beyond how crazy agentic filmmaking is - this is 100% the most fun thing I have ever worked on.
From the outside, these AI math releases remind me of when AI started beating chess grandmasters.
In chess, at least they can continue to compete with each other but that period of "fall of man" must have been horrible.
https://t.co/J2k02MicyW
Every GTA release people have said almost exactly this.
Could be changing POV, age, shootings, some other subliminal factor - or maybe it is the realism.
Editing with AI is totally solved if you...
Have the content already created or can create it... and it looks good... and audio that sounds good
Some of the issues / failure modes in the outputted video AI can identify so you just need to have a good eye/ear for the rest - you don't need to know exactly what's wrong just give it a direction and some description on what feels off.
@jessespafford It is so weird to need a consensus in the first place.
People should just feel how they feel about it now - and hold that line unless they learn something / have a change of heart that isn't just the herd catching up.
@simonw It's so odd to me the way weaker models and thinking is expressed.
Why is it not only worse but also lazier? Why does hydrogen low create a simpler image? Is that actually more correct in some ways? There's background described in the prompt.
Did anyone actually think otherwise? I appreciate the evidence but did we think they weren't getting them and using them somehow?
Next you're going to tell us they're distilling the models from claude or they've been scanning in ripped up books to improve the writing capabilities - or that they're reverse engineering all published software and hardware to train on. wait
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.
@AIWarper It's really good at some things but here were my first two results for a horse haha. There was some automation involved in the testing - my gpt and seedream 5.5 prompts didn't crossover well - better results with that tuned.