> While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models it just means that they've been

> While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models it just means that they've been working with paraphrased user data everywhere, which any smart person can figure out is how anthropic and openai train on so called non-retained data.