Chinese ChatGPT Rival Spooks Tech Markets

Hype about a Chinese owned AI tool has sent major tech stocks into a slump. The DeepSeek tool is much cheaper to run, which could hurt profitability among tech providers.

DeepSeek is a Large Language Model (LLM), a form of AI of which ChatGPT is arguably the best known. While specific models vary, it largely works by writing word-by-word, predicting each time which would be the most likely to follow.

It's been described as "autocomplete on steroids." More advanced models will break down a request before writing, planning the overall message and organization in advance.

Drastically Cheaper than ChatGPT

The big difference is that the operators claim it's considerably cheaper to run, with some users reporting that it's between 2% and 5% of the equivalent cost of using ChatGPT. The performance is not as good as the most expensive models on the market, but could be "good enough" for many users.

Part of that cost is that DeepSeek's makers reportedly spent far less to train its model, around $5.5 million compared to tens of billions of dollars for rivals. (Source: reuters.com)

That's partly because they were able to build on existing technologies and models. Meanwhile, some analysts believe DeepSeek may even have been trained on text written by other LLMs as well as real world documents. That's more efficient, though there's a big debate about whether that will compromise quality.

Chinese Censorship

Media attention to DeepSeek proved bad news for the stock prices of both rival companies and businesses such as Nvidia that produce specialist chips for running AI tools. Demand for its products could decline if people are using less power-hungry models.

Ironically, it's also claimed DeepSeek used around 2,000 Nvidia chips to train its model, despite a US embargo that should stop Chinese companies getting hold of such chips. (Source: theregister.com)

Inevitably privacy and security issues have been reawakened by this news, similar to concerns over Chinese authorities potentially accessing data gathered by TikTok. Meanwhile, DeepSeek appears to have built-in restrictions that prevent it writing about controversial Chinese topics such as the Tiananmen Square protest and massacre.

Also worth nothing is that every keystroke entered into DeepSeek is stored on Chinese servers. That is extremely problematic, given China's record on mass surveillance.

What's Your Opinion?

Have you tried DeepSeek? Should governments treat AI tools in the same way as other tech when it comes to security issues? Is a dramatic reduction in the cost of such tools a good thing?