We should keep things in perspective. The industry is barely three years old at this time. It's too early for believing there ought to be a larger gap between the industry leaders and the peasants.
As much as we talk about there being no moat, the list of countries that are actually competitive in large, generative models can be counted on one hand. It's basically just the US and China, and then maybe countries like the UK, Japan, and South Korea with a 2 years lag. The rest of the world isn't even in the running.
The US has already identified what it thinks and where it wants the bottle neck for AI to be - high-end machine learning infrastructure. This is why US companies have generally been pushing in the direction of larger models with escalating training and inference costs. The US figures it can afford it with its financial dominance, and it has access to the best chips, GPUs, etc. It's also why the US has attempted to keep specifically these things out of China's hands.
Make no mistake, this is a moat, even if it doesn't feel like one because China's been keeping pace. If you're a country like Iran, without access to the compute infrastructure (equivalent of high-end equipment) of countries that have them, chances are, you won't be training anything on the cutting edge and will continue to fall behind. Given that the 4th Industrial Revolution is predicated on AI automation, this means you're in for a rough time.
Long story short, we're building the moat for AI right this moment. That's why you feel there is no moat, when in reality, if the countries with the compute infrastructure banned you from using them, you'll fall right down into it. As of now, it's still pretty shallow, but imagine the size and the depth of the moat in two decades, when the winners of the AI race have had that much time to refine their infrastructure and ecosystem.
Fact is, without China and the handful of Western companies doing open source model releases, the vast majority of the world would only be able to access high-end generative models via service APIs - exactly where the US wants them.
But even with open source model releases, there is still a moat. Observe carefully and you'll realize that no one, not even China, is open sourcing how they're training the models. Open weights does not mean open training infrastructure. Being one of a few countries with the compute infrastructure and platform expertise to be able to train foundation models at the bleeding edge will become a huge moat.
It's not one that the rest of the world will be able to cross when they need to, and that is why the US is going all-out on this - the winners of the AI race will dominate the 4th Industrial Revolution.
As much as we talk about there being no moat, the list of countries that are actually competitive in large, generative models can be counted on one hand. It's basically just the US and China, and then maybe countries like the UK, Japan, and South Korea with a 2 years lag. The rest of the world isn't even in the running.
The US has already identified what it thinks and where it wants the bottle neck for AI to be - high-end machine learning infrastructure. This is why US companies have generally been pushing in the direction of larger models with escalating training and inference costs. The US figures it can afford it with its financial dominance, and it has access to the best chips, GPUs, etc. It's also why the US has attempted to keep specifically these things out of China's hands.
Make no mistake, this is a moat, even if it doesn't feel like one because China's been keeping pace. If you're a country like Iran, without access to the compute infrastructure (equivalent of high-end equipment) of countries that have them, chances are, you won't be training anything on the cutting edge and will continue to fall behind. Given that the 4th Industrial Revolution is predicated on AI automation, this means you're in for a rough time.
Long story short, we're building the moat for AI right this moment. That's why you feel there is no moat, when in reality, if the countries with the compute infrastructure banned you from using them, you'll fall right down into it. As of now, it's still pretty shallow, but imagine the size and the depth of the moat in two decades, when the winners of the AI race have had that much time to refine their infrastructure and ecosystem.
Fact is, without China and the handful of Western companies doing open source model releases, the vast majority of the world would only be able to access high-end generative models via service APIs - exactly where the US wants them.
But even with open source model releases, there is still a moat. Observe carefully and you'll realize that no one, not even China, is open sourcing how they're training the models. Open weights does not mean open training infrastructure. Being one of a few countries with the compute infrastructure and platform expertise to be able to train foundation models at the bleeding edge will become a huge moat.
It's not one that the rest of the world will be able to cross when they need to, and that is why the US is going all-out on this - the winners of the AI race will dominate the 4th Industrial Revolution.
