With DeepSeek models running at significantly lower cost but with the fear that the Chinese government will have access to data processed via these models it’s likely that they won’t be used in the US at scale.
Then with the US government pushing allies aside (for now) the EU will likely start pressing on local models. I’m thinking as it gets cheaper to produce models more and more countries will have a local model leading to global application providers having to be flexible. Super interesting article on the issue.
What do you all think?
6 Comments
I’m predicting a world where every major country has their own core model. Right now it appears dominated by China and US but why not a German model, Brazilian Model, Japanese Models etc. with governments creating favorable rules for use of local models especially because of possible job displacement (easier to tax a local provider) to make up for lost taxes and military use.
The global software game may be in for a big shift.
How would china get the data? You can run deepseek models in a docker container or locally on a computer that’s no longer connected to Internet.
I think people in general, OP included, are completely misunderstanding what AI in its current form is, and what the “arm’s race” actually is.
The models themselves are really not anything special. What is special is the amount of hardware that companies are throwing at them.
All AI is right now is throwing absolutely as much compute as you can get your hands on at the wall and seeing what sticks, regardless of the cost/financial viability.
Why was no one scraping and indexing the entire internet before? Because it’s so expensive to do, it makes no sense. That’s still true outside of the runaway speculative bubble.
The “arms race” is the same as it was “pre-AI,” it’s all about who has the most compute. It was happening before, it’s happening now, just at an accelerated pace, because the market and governments have lost their minds.
Everyone can get whatever data they want, and everyone can steal/copy whatever model they want.
Same as everything else, wait for China to do it 5-10 times cheaper. AI doesn’t have great returns, it’s not as good of an investment as things like PCs, cell phones and internet for now. It’s super expensive for what you get and while you get some impressive results at first, actually getting it refined and not hallucinating or completely misinterpreting things is like 80-90% of the work left to get the complex automation you think of when you think about what AI could do.
And then you also don’t have great robotics to full exploit AI even if you could get wannabe AGI working better.
Soo… it’s all pretty far away to see big job losses and big production gains. Narrow Scope AI is where the fast gains and huge production increases are, but nobody talks about that because you can’t use it to make Terminators.
Basically media and like 80-90% of people have a fantastical view of AI, not unlike when Japan and factory robots were going to take all out jobs and the Computers were going to take all our jobs and internet was going kill jobs and so on and so forth.
I’m sure when Homo Habilis made the first tool, there was lots of grunting about how it was going to take everybody’s jobs. Like OMG you made an Axe, what will all the hand wood choppers do! And then the next time food supplies got lean… they blamed the axe!
The distrust of the US isn’t limited to AI. You will see a lot of fallout from the current administrations policy in the next years and decades.
I’ve never read a more out of touch article in my life! Listen, if you want to sink your company, lose all respect from others and yourself, destroy people’s lives in the process, by going tech-bro-all-in, that’s your perogative I guess. Meanwhile there’s massive public blowback regarding AI that doesn’t live in the reality distortion bubble of AI grifting.
This bubble can’t pop soon enough.