How DeepSeek’s AI stacks up against OpenAI’s model

livemint.com:

Subscribe to enjoy similar stories. It’s impossible to look at the Chinese artificial intelligence startup DeepSeek’s new AI model without comparing it against OpenAI, the dominant American rival. DeepSeek has touted its latest AI model, R1, is particularly good at problem solving, performing on par with OpenAI’s o1 reasoning model—but at a fraction of the cost per use.

A DeepSeek app currently tops iPhone download rankings for the U.S. But just as DeepSeek and OpenAI are very different companies, R1 and o1 are also different technologies. Here’s a look at five ways the technologies share similarities and are different.

Wall Street Journal owner News Corp has a content-licensing partnership with OpenAI. DeepSeek reduced the data processing needed to train models, using its own inventions as well as techniques adopted by similarly constrained Chinese AI companies, The Wall Street Journal earlier reported. In addition to cutting down on data processing—offering massive savings in time and computing costs—DeepSeek uses a technique called “mixture of experts." DeepSeek and some other AI developers do something akin to delegating questions to experts in specific fields.

Each expert needs less training, easing the demand on chips to do everything at once. “The techniques they implemented aren’t new but applying them at the scale they did, with the conviction they had was novel," said Luke Arrigoni, CEO of Loti AI, an AI-based internet privacy platform. The Chinese company’s approach requires less time and power before a question is asked of the AI model, but uses more time and power to answer.