To clarify, does this have the same "open-source" definition as their current models (i.e. open-weight), or does it explicitly require the dataset to be released as well (by the strict definition)?

@moozooh Open Source just means Open Weight

@CampbellHutcheson There's an ongoing debate on whether it is correct to call open-weight models "open-source" if they don't include (or even disclose what is in) the dataset—the source, so to say—of the data stored in the weights. Since you cannot reproduce the model with what is made publicly available, "open-source" is a misnomer, so scientific (and adjacent) literature avoids using the term for models with private training datasets.

Given that prediction markets are usually strict on terminology—as they should be—I believe a clarification was warranted with this question, so thanks for that!

@moozooh yeah, I don't think much of the debate - it's basically just a question of whether something is pure enough to be considered open source - all the open models that anyone cares about in terms of actual usage (Llama, DeepSeek, etc...) are open weight but don't provide their training data.

I do understand that this is a popular debate in certain circles though.