Comment by locknitpicker
2 hours ago
> Which means that the US lead is all but gone.
You can see in this discussions the goalposts being actively moved to desperately create a notion that the lead is still there.
You even see people in this thread referring to imaginary wonder models that they expect US companies to have under development that are supposedly leaps and bounds more impressive, albeit the only reference there is to their existence is the commenters hopes and dreams.
I think people need to realize that there is no longer a cutting edge company that is expected to dominate, and among the multitude of AI companies in the field competition is so tight that no one would even be surprised if tomorrow some other company far away from the US was able to crank out the top performing model, or something like a small local model that outperforms mid/high tier frontier models. The local model alone would be a death sentence to US companies, who are still burning through 10x their revenue.
If true, it’s quite a bit more impressive than many people seem to think it is.
It’s incredible how little GPU compute they have to work with compared to the US frontier labs.
People thought the models are getting more intelligent via better architectures but all that really happened is that tool calling became a first class feature that every model is expected to support and RL training with coding harnesses became the standard.
Two years ago, tool calling was always considered an extra feature on top that some models where good at while other models completely ignored tool calling.
Meanwhile the discourse about how intelligent these models have become and how this is a major step or shift in model intelligence would make you believe that these models suddenly have a more powerful architecture when in practice model optimization has lead to faster but less expressive models that are still more intelligent, meaning that the model architectures didn't really experience meaningful shifts, it's that the training methods had a massive shift away from supervised training which inherently constrains the models.
The AI labs don't have a monopoly on the outcomes of RL training because RL is significantly less dependent on private data. Data helps massively for bootstrapping, but beyond post training, it's up to the model itself and every AI lab has a model so it is much harder to differentiate between your competitor if the RL training environments are the same (I specifically mean coding harnesses and Linux command line tools) because the environment here is the source of the "synthetic" data and everyone has equal access to it.