3 ms·
I'm curious to know what proportion of the training data at frontier model shops is composed of their scanned book text vs the internet.
by cmcollier 2mo ago
I'm curious to know what proportion of the training data at frontier model shops is composed of their scanned book text vs the internet.
- devmando 2mo agothis is a good question, and I wonder if its even open to the public. My limited knowledge knows that Anthropic settled a lawsuit of I believe ~1.3billion for scanned textbooks if I'm not mistaken.
- adityaathalye 2mo agoOne would hope that a "public benefit corporation", such as OpenAI, would be transparent about the stuff that is being ingested by sausage factory. The literal legal entity is about benefiting the public after all. Since the leadership is obviously not going to say anything factually true or psychologically honest, I wonder if anyone has tried invoking some aspect of corporate or constitutional law, given their legal status (as nonprofit that controls a PBC) to force such disclosures.