The China Mail - DeepSeek dims shine of AI stars

Since late 2022, just a handful of AI assistants -- such as OpenAI's ChatGPT, Anthropic's Claude, and Google's Gemini -- have reigned supreme, becoming ever more capable thanks to multi-billion-dollar investments in engineers, data centers, and high-performance AI chips.

But then DeepSeek upended the sector with its R1 model, which it said cost just $6 million or so, powered by less-advanced chips.

While specialists suspect DeepSeek may have cost more than its creators claim, its debut fueled talk that GenAI assistants are becoming just a regular commodity, thanks to innovation and market forces.

"The first company to train models must expend lots of resources to get there," said CFRA senior equity analyst Angelo Zino.

"The second mover can get there cheaper and more quickly."

At a HumanX AI conference in Las Vegas this week, Hugging Face co-founder Thomas Wolf said it is getting less expensive to launch GenAI models -- and less important which one people use.

"I feel like we are moving to this multi-model world, which is a good thing," Wolf said, pointing to the muted reception given to the most recent version of ChatGPT.

- Stay flexible -

At the conference, OpenAI chief product officer Kevin Weil pushed back against the notion that all models are created equal.

"That's actually not true," Weil said.

"The days of us having a 12-month lead are probably gone, but I think we have a three- to six-month lead, and that is really valuable."

Weil said OpenAI plans to fight to keep that narrowing edge over its competitors.

With 400 million users, San Francisco-based OpenAI has the advantage of being able to use data from massive traffic to continually improve its models, Weil explained.

"OpenAI has the Google advantage of being the thing that's in everybody's minds," said Alpha Edison equity firm research director Fen Zhao.

Jeff Seibert, chief of the accounting and AI start-up Digits, agreed that OpenAI will stay ahead of the pack but added that he expects the gap to eventually close.

"For advanced use cases, yes, there will be a lot of advantages," he said of OpenAI's position.

"But for a lot of stuff, it won't matter as much."

Seibert advises entrepreneurs to design their technology to allow them to swap out GenAI models, affording them flexibility in a quickly changing industry.

- Cash burn -

Improved use of chips and new optimization techniques have driven down the cost of designing the large language models (LLMs) that power ChatGPT, Gemini and their rivals.

An open-source approach taken by some LLMs is credited with helping accelerate innovation by making the software free for anyone to tinker with and improve.

The valuation of closed-model startups such as Anthropic and OpenAI has likely peaked as their "first-mover advantage dissipates," according to Zino.

Japanese investment colossus SoftBank pumped $40 billion into OpenAI in February in a deal that valued the startup at $300 billion -- almost double what it was last year.

“If you're burning a billion dollars a month, which I think OpenAI is, you have to keep raising money," said Jai Das of private equity firm Sapphire Ventures.

"I have a hard time seeing how they get to a point where revenues are higher than the amount of cash they burn."

Anthropic raised $3.5 billion in early March, valuing the champion of responsible AI at $61.5 billion.

T.Wu--ThChM