alt Hacker NewsThat’s an argument made about training the initial model. But the comment stated that DeepSeek stole its research from the US which is a much stronger allegation without any evidence to it.
Replies
elmomle • 12/08/2025
That's a fair point. I suspect that to one outside the field, their touting major breakthroughs while trying to conceal that their first model was a distillation may cause a sense of skepticism as to the quality of their research. From what I've gathered, their research actually has added meaningfully to understandings of optimal model scaling and faster training.
For starters ChatGPT was pretty much trained on "stolen" data. However I actually do support it. I think both cases - ChatGPT preying on world wide data and Deepseek using such data by partially "borrowing" it from ChatGPT are fair game.