OpenAI Says DeepSeek Used Its Work Without Permission to Create an AI That's Stealing Its Job

Yahoo

29-01-2025

After spending years indiscriminately ripping off other people's work — and getting sued for copyright infringement left and right — OpenAI is sobbing that the buzzy Chinese AI startup DeepSeek did the same thing on its AI that it built on all that pilfered content.
As the Financial Times reports, OpenAI is accusing the app of using its proprietary models to train its own ChatGPT competitor. The company claimed that it had "some evidence" of DeepSeek using the output of OpenAI's models to train its own, a technique called "distillation," that it says may have breached its terms of service.
The news comes after DeepSeek flipped the AI industry on its head, wiping out over $1 trillion worth of market capitalization in a single day with an advanced model that's much less resource-intensive than anything cooked up by Silicon Valley. As the main purveyor of AI computing infrastructure, chipmaker Nvidia broke its own record for the biggest single-day loss of any company in history.
Now that DeepSeek has come for OpenAI, the ChatGPT maker is suddenly crying foul, whining that the upstart lab did exactly what it did to human creatives: vacuumed up its work and used it to build a competing product.
"The issue is when you [take it out of the platform and] are doing it to create your own model for your own purposes," a source close to OpenAI told the FT.
It's hard not to see the development as the peak of hypocrisy. OpenAI has a long track record of hoovering up data with an astonishing disregard for giving credit or fairly compensating rights holders.
That's not an exaggeration. OpenAI CEO Sam Altman has previously admitted that it would be impossible to "train today's leading AI models without using copyrighted materials."
Put simply, it's roughly the equivalent of a school bully complaining to the teacher that his stolen lunch was stolen from him by another bully.
"Funny on OpenAI, who totally didn't gobble up our data without asking, is now pointing fingers at DeepSeek for doing the exact same thing," one Bluesky user wrote.
"I'm so sorry I can't stop laughing," AI critic Ed Zitron wrote in a scathing post. "OpenAI, the company built on stealing literally the entire internet, is crying because DeepSeek may have trained on the outputs from ChatGPT."
"They're crying their eyes out," he added. "What a bunch of hypocritical little babies."
It's also worth reiterating that despite its name, OpenAI is a closed-source and for-profit company — while DeepSeek's AI models are open-source.
The Sam Altman-led company refused to officially comment or provide further details, the FT reports.
And in reality, whether DeepSeek actually infringed on its intellectual property remains debatable.
On Tuesday, president Donald Trump's "crypto czar" told Fox News that distillation is "when one model learns from another model [and] kind of sucks the knowledge out of the parent model."
"And there's substantial evidence that what DeepSeek did here is they distilled the knowledge out of OpenAI models, and I don't think OpenAI is very happy about this," Sacks explained.
If independently confirmed, what DeepSeek pulled off with its latest model, dubbed R1, could signify a major advancement in the development of cheaper-to-run, but still as powerful, AI models.
The company used just over 2,000 Nvidia graphics cards, purchased shortly before the US banned exports of these cards to China.
Some experts believe that DeepSeek likely didn't do anything wrong.
"It is a very common practice for start-ups and academics to use outputs from human-aligned commercial LLMs, like ChatGPT, to train another model," University of California, Berkeley PhD candidate Ritwik Gupta told the FT. "That means you get this human feedback step for free. It is not surprising to me that DeepSeek supposedly would be doing the same."
In a statement, OpenAI said that it engages in "countermeasures to protect our IP, including a careful process for which frontier capabilities to include in released models" — despite having the word "open" in its name.
The company added that it's "critically important that we are working closely with the US government to best protect the most capable models from efforts by adversaries and competitors to take US technology."
Whether DeepSeek actually represents a threat to national security remains to be seen, and the White House has since revealed that it's currently evaluating the possible risks.
But given Trump's ongoing efforts to get Chinese-owned video platform TikTok unbanned in the country — Congress banned the app last year — it'd be a notable double standard if he'd start coming after DeepSeek.
How OpenAI's accusations will play out remains to be seen, but the optics aren't great, no matter which way you spin it.
"There's no way OpenAI can make this argument without looking very, very silly," tech journalist Mike Masnick wrote in a post on Bluesky.
Meanwhile, Zitron didn't pull any punches.
"Are you crying because your plagiarism machine that made stuff by copying everybody's stuff was used to train another machine that made stuff by copying stuff?" he said. "Are you going to cry? Cowards, losers, pathetic."
More on DeepSeek: Former Intel CEO Says After Seeing DeepSeek, He's Done With OpenAI