OpenAI and Microsoft are big mad that Chinese AI startup DeepSeek has stolen their market share and, possibly, portions of their code. It’s a deeply funny claim from the company that made ChatGPT, a program it once admitted couldn’t exist without free access to all the copyrighted data in the world.
OpenAI told The Financial Times that it had evidence DeepSeek used what it called “distillation” to build its wildly successful chatbot. I’ll let David Sacks, a storied member of the PayPal Mafia and current White House AI and Crypto czar, explain how distillation works.
“There’s a technique in AI called distillation,” Sacks told Fox News on Tuesday. “It’s when one model learns from another model. What happens is the student model asks the parent model a lot of questions just like a human would learn. But AI can do this asking millions of questions and they can essentially mimic the reasoning process that they learn from the parent model. And they can kind of suck the knowledge out of the parent model.”
Got it. Distillation is when one AI sucks off another AI. So it’s a fancy word for copying.
OpenAI is alleging that DeepSeek got a bunch of outputs from ChatGPT and used those outputs to train DeepSeek, thereby standing up an LLM at a fraction of the cost of the billions used to train ChatGPT. OpenAI wouldn’t give Financial Times any evidence to back up the claim and also said that it’s impossible for any company to make a direct “copy” of ChatGPT.
“There’s substantial evidence that what DeepSeek did here is they distilled the knowledge out of OpenAI’s models,” Sacks said on Fox News, without providing said evidence. “And I don’t think OpenAI is very happy about this. I think one of the things you’re going to see out of this is our leading AI companies taking steps to prevent distillation.”
Bloomberg had a few more details. An OpenAI insider told the outlet that Microsoft had noticed weird activity around ChatGPT’s API and flagged the company. The API calls may have been linked to DeepSeek’s “distillation” of ChatGPT. If the allegations are true, it’s possible that DeepSeek made a bunch of calls to ChatGPT’s API, generated a bunch of data, and then used that to train its own LLM.
They noted this might violate ChatGPT’s terms of service. Heaven forfend.
DeepSeek dethroning ChatGPT on Apple’s AppStore coincided with hundreds of billions of dollars in market value around AI companies being wiped out from the stock market earlier this week. NVIDIA, which manufactures the chips used to train the chatbots, lost more than $500 billion.
Sacks assured Fox News viewers that America was still number one and that the U.S. just has to keep building the big beautiful datacenters that Trump wants. “There are still great advantages to having an enormous number of chips. And this is an area where America could continue to lead, is a build-out of this infrastructure and having the most advanced chips,” Sacks said. “So I think it’s a little bit of an overreaction to say that America does not need AI data centers anymore. I think we need to build out these big AI data centers. I think President Trump has expressed support for them and I think we need to make it easier to build those data centers, to get them permitted and licensed, and to get power generated, to get electricity for those data centers.”
Always on message, Trump’s AI czar emphasized that the reason China was catching up was twofold. One was burdensome regulations imposed on AI companies by Biden. The other was that the AI companies had gotten too damn woke.
“I think that our AI companies got a little distracted…I think that maybe they got a little bit complacent. They didn’t realize how close these Chinese companies were to them. They wasted a lot of time on things like DEI. You saw there was woke AI,” Sacks said. “The models were basically producing things like black George Washington.”
As of this writing, DeepSeek is still number one on the App Store.
+ There are no comments
Add yours