It could be quite accidental and not pre-concieved though. Where some metric, "time spent in chats" drives certain features or parameters. Could be fully automa

It could be quite accidental and not pre-concieved though. Where some metric, "time spent in chats" drives certain features or parameters. Could be fully automated (i.e. continuous A/B tests) or through layers of "Excel managers". But it stands to reason that a company like OpenAI or Anthropic has metrics in place that drive their setup towards "more engagement" and away from "factually correct".