haha i guess this is a old long problem no and the root of major rule based religions? we might go down the same road haha but that is clearly not the way i think we should go; Values i believe comes from raising in humans (parents, environment e.t.c.)
People love saying this kind of stuff and then just stopping there. Why not explain what youโve tried? What has helped? What hasnโt helped? What are the current top hypotheses? What are the blockers?
that creates a bias against familiar things and stuff; we can't ask AI to reason and do the best case cuz in certain cases it might not be in humanity's interest;
So I believe the only way is making AI and humans grow together hence some form of coevolution is needed
the guardrails in the recent class of models makes it really unsuable in diverse aspects of life(it is very enterprise focused); I feel really sad of the reality where open models with substantial capability is nearing zero.
@leedubo my point was not the price but wanting abliterated models that could satiate my needs. I am discontent seeing current capable of open models and their rollout rate
@v10se@elshayib_ benchmarks is always practical use and there oss is kinda shit; i would argue sonnet 5.5 would be better than most (claude free plan pretty generous) but I hope opensource wins cuz i am tired of all the guardrails and trying to be good not opinionated models
@tszzl@RhysSullivan bro please it is not, it drains your usage like shit and tries to open 100 tabs and once and complete none; sometimes fucks up basic things like uploading files from finder. Meanwhile opus 5.5 is a beast at medium
@stisov16@ycombinator@bcherny Sadly they ban it in my region to the most extreme extent; sure they had ways to stop distillation without banning access rah