News for hackers. Stuff that matters.

A comment on OpenAI says it has evidence China's DeepSeek used its model to train competitor. Parent comment

Re:OpenAI says it has evidence China's DeepSeek used its mod... by Voloskaya on Wednesday January 29, 2025 @03:01PM (#42870355)
> First of all, training off of data generated by another AI is generally a bad idea because you'll end up with a strictly less accurate model (usually).

That is not true at all.

We have known how to solve this for at least 2 years now.

All the latest state of the art models depend heavily on training on synthetic data.

[ Parent | on HN ]
Re:OpenAI says it has evidence China's DeepSeek used its mod... by bjourne on Wednesday January 29, 2025 @07:18PM (#42873203)
https://www.nature.com/articles/s41586-024-07566-y
[ Parent | on HN ]
Re:OpenAI says it has evidence China's DeepSeek used its mod... by Voloskaya on Wednesday January 29, 2025 @08:01PM (#42873528)
Key point from your linked paper:

> We find that indiscriminate use of model-generated content in training causes irreversible defects in the resulting models

No one is training on indiscriminate synthetic data. It's very much discriminated, but still synthetic.

[ Parent | on HN ]