I don't understand quite a few things:
- Why can't American AI companies get advantage of the DeepSeek's work, since it is open-source, and use their billions worth of resources to achieve a model that is bigger than DeepSeek?
- What evidence do we have that the cost to train DeepSeek was actually $6 million?
They cannot switch to DeepSeek at the flick of a button. It takes a lot of work and many years to switch to a newer/better model (see: DLSS4 Transformer vs CNN).
ChatGPT raised billions of dollars during their last round, so investors are worried that they overpaid. It's like paying $3000 for RTX 5090 and then seeing a Chinese-equivalent for half the price.
Imagine having a company with thousands of Windows PCs... you think it's easy to switch to Linux, just because it's free and open-source?
It might take 4-5 years to switch to a new paradigm and train staff. Until then, China will have the lead.
China is generally cheaper (Xiaomi/BYD electric cars are 2 times cheaper compared to Tesla), so there's no reason to doubt Chinese AI can be cheaper too.
Btw,
DeepSeek seems to have WEF's seal of approval, so I guess
it's destined to succeed.
