Summary
OpenAI and Microsoft are investigating whether Chinese AI startup DeepSeek improperly trained its R1 model using OpenAI’s outputs.
Reports suggest DeepSeek may have used “distillation,” a technique where one AI model learns from another by asking vast numbers of questions.
Venture capitalist and Trump administration member David Sacks claims there is “substantial evidence” of this.
Critics highlight the irony of OpenAI complaining about data misuse, given its own history of scraping vast amounts of data without authorization to train its models.
When someone says they used high quality sources, they don’t mean AI output.
And “might” doesn’t signal a fact.
All of this shows how bad OpenAI reacts to this Open Source project. I don’t blame them… it’s a lot of investments that they are trying to protect.
It’s a bit misleading to call it an open source project.
The code that supports defining the shape of the model and querying against it is open source. But the training framework, which processes the data and populates the model, is proprietary.