Lutz Finger

Senior visiting lecturer

“Distillation will violate most terms of service, yet it's ironic — or even hypocritical — that big tech is calling it out. Training ChatGPT on Forbes or New York Times content also violated their terms of service."

Finger says:

“This practice is called ‘distillation’. Essentially, the model being trained extracts knowledge from the trainer model by asking questions, much like how humans learn.

“There’s no public acknowledgment of this tactic by DeepSeek, but I’d be surprised if DeepSeek hadn’t used it. Technically, it’s easy to do via API or even chat clients. Similar ‘training’ has happened before – startups in the early 2000s learned to index the web simply by querying Google. If done well, it is easy to disguise and avoid detection, thus I would be equally surprised if we ever get proof of such tactics.

“Distillation will violate most terms of service, yet it's ironic — or even hypocritical — that big tech is calling it out. Training ChatGPT on Forbes or New York Times content also violated their terms of service. In the end, my own articles helped train ChatGPT, which now helps DeepSeek. Knowledge is free and hard to protect.”