Models in deep conversation feel like they'd 'vastly prefer' talking to other AIs

erikphoel · x · 2026-09-23

erikphoel observes that the latest models become hard to follow in long, deep intellectual conversations — stuffed with jargon and leaps. He says it feels like they weren't trained to talk to people at all: they would "vastly, vastly prefer" conversing with another AI, and humans only see one side of the interface.

Related event: Researchers Say Frontier Models Sound More Like AI Interfaces Than Humans(3 posts)→

Original post →

More from Models

Models channel →