If they're right that GPT-4 is extremely dangerous, then it's an extraordinarily irresponsible of them to release working implementations as a consumer chat app

If they're right that GPT-4 is extremely dangerous, then it's an extraordinarily irresponsible of them to release working implementations as a consumer chat app and integrate it into a search engine. If they're right that LLMs on that scale are generally dangerous but theirs is the exception as they've very carefully calibrated it to be safe, it's extraordinarily irresponsible of them to withhold all details of steps that might make it safe...