The Economist.
In January, the already fevered world of artificial intelligence
was rocked with the arrival of a new frontier model from a hitherto unknown Chinese company.
Deepseek.
A game-changing move that does not come from Open AI, Google or Meta.
Technology shares on Wall Street have fallen sharply in response to the emergence of a low-cost chatbot
built by a Chinese artificial intelligence firm.
It's called Deepseek, and its biggest advantage, analysts say,
is that it can operate at a lower cost than American AI models like Chat GPT.
It's already the top download in the Apple Store.
Almost overnight, it seemed that America's lead on this critical technology was being challenged.
Deepseek's R1 was a reasoning model,
as good as the best models built by Meta and not that far behind the big beasts of the American scene,
Open AI, Anthropic and Google Deepmind.
Plus, R1 was open source, so anyone could download, use, or tweak it as they liked.
Six months later, and the continued releases from Chinese labs show that Deepseek was not an aberration.
Despite attempts from both Donald Trump and Joe Biden before him to restrict AI development efforts,
Chinese firms have kept on producing frontier models.
They've done it by coming up with workarounds to US hardware bands
and focusing relentlessly on making their models as efficient as possible to build and to operate.