Why you can deep-six the DeepSeek hype
A fast, cheap open-source LLM-based AI chatbot has emerged from China. Here's what's bugging me about all the hype.
Tech stocks dropped today after Wall Street noticed a Chinese startup called DeepSeek, which makes a cheap AI chatbot everyone fears is also good.
Called R1, the one-year-old startup's large language model was developed with only $5.5 million, according to the company, which is a fraction of the cost of ChatGPT, which cost billions.
Investors sold US AI stocks out of fear that the big-budget American AI industry is vulnerable to Chinese innovation. Nasdaq dropped by 2.3%.
Nobody knows how good DeepSeek is for sustained use. But early enthusiasm is unbridled.
Tech news reports say breathlessly that R1 downloads exceed ChatGPT downloads.
Investor Marc Andreessen called DeepSeek "one of the most amazing and impressive breakthroughs I've ever seen… "DeepSeek R1 is AI's Sputnik moment."
Of course, Andreessen is often wrong.
Stratechery's Ben Thompson pours cold water on DeepSeek: "This has been overstated. R1 is competitive with [ChatGPT] o1, although there do seem to be some holes in its capability that point towards some amount of distillation from o1-Pro. OpenAI, meanwhile, has demonstrated o3, a far more powerful reasoning model."
The other problem is that (like TikTok), DeepSeek R1 contains Chinese government censorship within its training model.
In my previous issue of Machine Society, I predicted biased AI chatbots and specifically said, "A Chinese Communist Party-friendly set of chatbot services may gain popularity globally, much like TikTok." (This was in the paid-subscriber section of the newsletter.)
And finally, the downloads metric is meaningless. Of course, a new chatbot in the news that people just heard about this week will have more downloads this week than ChatGPT, which has been available for months and has already been downloaded by everyone interested.
The hysterical reaction to DeepSeek is foolish and, at best, premature.
Expect this hype to be gradually deflated over the coming months.
More From Mike
Apple is the latest company to get pwned by AI
Robots get their ‘ChatGPT moment’
Meta puts the ‘Dead Internet Theory’ into practice
How to calculate your AI-powered cybersecurity’s ROI
Where’s Mike? El Salvador!
(Why I’m always traveling.)
Love this viewpoint Mike.
Why do mainstream media hype up Chinese products so much? For all we know it can be a spyware like Tiktok. You already need email verification to sign in and cannot use "semi-trusted" Google login from US. And I agree with Sam Altman. New ventures are always riskier. I mean it's like someone came up with a new search engine from China and mainstream media is all abuzz with it. It's not as if DeepSeek will give better streamlined answers than Chat-GPT. Chat-GPT's main problem is memory and hallucination.