4 ms·

I agree, also my thought, but that is a principally different case then requiring consent from the author. That suggests that OpenAI would have downloaded pirat

I agree, also my thought, but that is a principally different case then requiring consent from the author. That suggests that OpenAI would have downloaded pirated material, or hacked paywalls, how did they get this material? BTW I noticed that GPT-4 is good at writing legal letters of the sort that is widely available online. But a subpoena, ('dagvaarding', the Dutch version I have researched) it completely fails to create. Also there are not many subpoenas available online, and the court (in the Netherlands) only publishes the verdicts, not the other documents. Lawyers OTOH have a lot more of this available in their libraries. So, my impression is that there is still a lot of material out there that is not in the corpus.

Jensson 3y ago
> how did they get this material? "Hello, we are a non-profit that wants to make AI models to benefit humanity. Can you give us access to your data to help us in our work?"
- jdiff 3y ago
  This is not what they did. They fed off enormous databases of pirated material.
  - Jensson 3y ago
    Yeah, but they wont get non-public data that way. I'd bet they did get access to a lot of non-public data just by asking and stating they do it for a non-profit mission.