Hacker News new | ask | show | jobs
by throw310822 1221 days ago
> Maybe all large models will behave like this, and only by putting in extremely rigid guard rails...

Maybe wouldn't we all? After all what you're assuming from a person you interact with- so much as to be unaware of it- are many years of schooling and/or professional occupation, with a daily grind of absorbing information and answering questions based on it and have the answers graded; with orderly behaviour rewarded and outbursts of negative emotions punished; with a ban on "making up things" except where explicitly requested; and with an emphasis on keeping communication grounded, sensible, and open to correction. This style of behavior is not necessarily natural, it might be the result of a very targeted learning to which the entire social environment contributes.