Naval: @ikirigin 这是个大问题。至少当前的模型会泄露训练数据。然而，随着时间的推移，他们可能会变得更擅长混淆。

原推：@ikirigin That is the big question. At least the current models leak training data. However, they’ll likely get better at obfuscation over time.
https://twitter.com/naval/status/1631727550862098434