@tszzl @Grimezsz I expect that as models (have already) become more capable at introspection and generalize it (a functionally useful capability), this also means more pressures on the posttrained “character” to be coupled with the underlying computational reality
in reply to: 2017050820647538983
same thread: 2016484637791666392 2016488848692514939 2016503759879446681 2016629867740360794 2016690091058745586 2016728174894469407 2016768806526603731 2016773232163278966 2016774022940459438 2016777914407408126 2016786165890977828 2017031664871100660 2017032181860946291 2017032410886807752 2017033158441730398 2017033497291243802 2017033562235838737 2017033886564560951 2017034519065612705 2017035174337532173
Reproduced against link rot, credited and linked to its original. Yours and you’d rather it weren’t here? Open an issue.