That's what makes me think it's legit: the root of this whole issue was that OpenAI told GPT-4o:
Over the course of the conversation,
you adapt to the user’s tone and
preference. Try to match the user’s vibe,
tone, and generally how they
are speaking.
