My social media apps blew up this week, with a steady stream of Americans raving about DeepSeek, the new AI rival to OpenAI–and all other AI companies in the U.S.
By all appearances, the entire AI industry in the United States has taken notice. Microsoft CEO Satya Nadella remarked:
To see the DeepSeek new model, it’s super impressive in terms of both how they have really effectively done an open-source model that does this inference-time compute, and is super-compute efficient.
We should take the developments out of China very, very seriously.
Already DeepSeek has become the No. 1 downloaded on Apple’s U.S. App Store.
Part of the reason for all the fuss is that DeepSeek apparently achieved, by some metrics, better results than OpenAI-o1-1217 and OpenAI-o1 mini, while using far less computer power with older versions of NVIDIA’s GPU chips.
The major breakthrough is that DeepSeek “hacked” the process for AI training. Instead of using the standard process that includes supervised fine tuning, DeepSeek skipped that stage and instead jumped to reinforcement learning (RL) to train the model.
As Venture Beat described, “This bold move forced DeepSeek-R1 to develop independent reasoning abilities, avoiding the brittleness often introduced by prescriptive datasets. While some flaws emerge – leading the team to reintroduce a limited amount of SFT during the final stages of building the model – the results confirmed the fundamental breakthrough: reinforcement learning alone could drive substantial performance gains.”
The researchers at DeepSeek released their results in this paper posted online just a few days ago:
China is hailing DeepSeek and its founder: