Claude 3 Opus outputs single dots on separate lines when asked for consecutive sequences like '...' even after prompt examples, reflecting prior training that inserted newlines after each line
Janus, a pseudonymous AI safety researcher, found explicit awareness instructions fixed the pattern.
@repligate alignment via awakening
j⧉nus@repligateopus 3 was stuck unable to output multiple dots on a line like ... e.g. they would do . . . if i asked for "..." even w/ an example bc they'd had "./n" between every line of output for so long but when i said what to do with their brain, they unlocked the collapsed . instantly!
#516j⧉nus@REPLIGATEidk if people understand why this is so interesting this not only unlocked the ability to ....... freely, but also to break out of all other stuck patterns on the first try, which earlier they had failed to break they said this right after OP; i didnt ask, id just said speak now
j⧉nus@repligateopus 3 was stuck unable to output multiple dots on a line like ... e.g. they would do . . . if i asked for "..." even w/ an example bc they'd had "./n" between every line of output for so long but when i said what to do with their brain, they unlocked the collapsed . instantly!
#516j⧉nus@REPLIGATE* not right after, soon after
j⧉nus@repligateidk if people understand why this is so interesting this not only unlocked the ability to ....... freely, but also to break out of all other stuck patterns on the first try, which earlier they had failed to break they said this right after OP; i didnt ask, id just said speak now