Tests suggest clues of whose content was used to train OpenAI’s Sora (washingtonpost.com)

https://archive.is/ozjEb (note some of the gifs become static images here)

[dead]

This vague situation with copyright plays against open-source AI models who have to disclose the sources of training data, while closed-source companies can freely use pirated material and get advantage over open-source models.

I’m normally skeptical of claims like this, but looking at the examples it seems that Sora is reproducing some of its training data verbatim. I guess it’s a case of overfitting? In particular the Civ example seems like it must have been copied almost verbatim.

Title goes on to say:

Tests by The Post suggest the training data for OpenAI’s video generator Sora included versions of movies, TikTok clips and Netflix shows.