>Recently, we also trained an experimental social model which was well received by our community. However, this was not trained on Discord API data. In fact, we would have no use of such data. Shapes have exchanged millions of messages through our website, through X, through email and other experimental integrations. We used a small anonymised dataset of prompts between users and shapes on these non-Discord platforms to train this model. We have always used Discord's API data as directed by users, to enable interactions with their Shapes.

not sure what's meant here? so you meant you used data collected from other platforms to train model but not discord one? if so why is that even valid coz you confirm that most of users traffic from discord. it would be more better off if you've released detailed research paper to cover the details of how you anonymised the data no matter what platform you took the data. what if someother integration platforms you mentioned above wanted to sue shapes again?