File: mind_blown.gif (2.92 MB, 500x333)

2.92 MB GIF

The Chinese AI companies open sourcing their LLMs is an absolutely genius move for undermining the big American AI companies.
Releasing the source allows companies and individuals to host their own models locally for only the cost of hardware and electricity. The big drawback of course is that the open source models like Deepseek and Kimi K2 aren't quite as good as the big players like Claude and ChatGPT, but they provide a good enough option for a lot of uses. They also continue to distill better models' weights all the time. This means that the open source models will eventually catch up if the closed source models aren't continuously improving.
Anthropic and OpenAI both plan to eventually make money by hooking enough customers on their service and then increasing costs until they can make money. At current prices, inference isn't profitable for these companies, but it's close enough that it could possibly become profitable. Training new models is a huge money sink and is likely a very long way from being profitable, if it ever is.
The existence of open source models forces the big players to continue training new models at a huge loss in order to be able to charge for their service while the competition is free. If free models like Deepseek get close enough in quality, nobody will pay for LLM access.
All the Chinese companies have to do is continue applying this small amount of pressure at a relatively low cost, and either the American companies will be forced into charging so much that customers will balk, or they'll run out of investor money and go broke.
I don't get why this isn't the first thing that comes up in every conversation about AI as a business.

Thanks for coming to my TED talk.