It’s time for AI to come home
Subscribe to enjoy similar stories. Put aside the claims that can’t be confirmed. What kind of presumably second-rate chips did it use? How little did it spend training its derivative large language model? Likewise, if analysts at Washington’s Center for Strategic and International Studies are correct, DeepSeek trained itself on information illicitly gleaned as an enterprise client of America’s pioneering OpenAI.
In some instances, the Chinese chatbot will even mistakenly identify itself as OpenAI’s ChatGPT if asked. Still the basic story of DeepSeek seems to hold up. Outside China’s government-run industrial policy, a young hedge-fund guy using his own money reproduced a highly competitive AI chatbot with genuine innovations to reduce costs and improve performance.
Liang Wenfeng’s project, which he partly revealed in an open-source release, rattled global markets this week but also confirmed the direction Western tech leaders have taken. Mr. Liang is less interested in building a business, he says, than joining the global quest, using so-called large language models, for artificial general intelligence—in machines, the human-like capacity to learn, plan and create new things in the world.
Any good neurotic moment should bring up questions lurking below the surface. When technology diffuses so widely and quickly, and major innovations can still emerge from unexpected quarters, the U.S. and China probably should give up any notion of capturing a lasting advantage.
One or both, though, can sabotage themselves by imposing too much and the wrong kind of government control, including using state funding to dominate the research agenda, in this case more likely to impede progress than speed it up. Mr. Liang’s cost-slicing,
. Read more on livemint.com
