Chinese AI gets better — and cheaper
source: axios.com (contributed by Bill Amshey) | image: pixabay.com
Chinese AI makers have learned to build powerful models that perform almost as well as the best ones in the U.S. — for less money and with much less demand for energy, Axios’ Scott Rosenberg and Alison Snyder report.
- V3, an open-source model developed by Chinese firm DeepSeek, performs about as well on various benchmark tests as OpenAI and Anthropic’s most advanced models.
- DeepSeek says it cost just $5.6 million to train V3 — compared to the hundreds of millions of dollars American companies have spent to build and train their models.
🤖 Between the lines: The Biden administration has done a lot to advance AI in the U.S. and keep those advancements out of the Chinese government’s hands.
- It has invested heavily in domestic manufacturing for powerful chips and new energy sources. And it has imposed tight export controls to prevent those chips from reaching China, including through third countries.
- That seems to have worked in the short term, while spurring China to compete just as aggressively to develop its own tools.