now, i would like to put the blame largely on article sites which just really want some bombaatic headlines, but calling text-predictors output sensational is… quite something.
It’s crazy because it’s not just click hungry no name sites, but major well regarded papers are reporting similar things, or at least, treating the people who say such things as people worth “hearing out”.
There are a lot of publications and writers who i used to see as competent but have completely burnt up their credibility by going along with this.
are u addressing the mythos thing in particular or LM capability in general or umm… “safety” stuff?
the mythos stuff is entirely overblown and that has been proven in many actual usecases by now. its nothing new.
on general capability increase, it… really only matters if u care about it or wanna use those models. if not, its marketing hype. if u do, even in LM spaces the best saying is still “try it urself”
My issue is how reporting on improvement in models suggests that they will start being able to do things they couldn’t do before, they’re definitely better at what they could already do, but the set of tasks they are useful for has not been increased by improvements in their capabilities.
Like, to me, the reporting sounds like someone saying “We’ve doubled the power of this car’s engine, it can now fly, and cook you dinner”.
yea, its disappointing.
now, i would like to put the blame largely on article sites which just really want some bombaatic headlines, but calling text-predictors output sensational is… quite something.
It’s crazy because it’s not just click hungry no name sites, but major well regarded papers are reporting similar things, or at least, treating the people who say such things as people worth “hearing out”.
There are a lot of publications and writers who i used to see as competent but have completely burnt up their credibility by going along with this.
are u addressing the mythos thing in particular or LM capability in general or umm… “safety” stuff?
the mythos stuff is entirely overblown and that has been proven in many actual usecases by now. its nothing new.
on general capability increase, it… really only matters if u care about it or wanna use those models. if not, its marketing hype. if u do, even in LM spaces the best saying is still “try it urself”
im just yapping now, sorry >~<
My issue is how reporting on improvement in models suggests that they will start being able to do things they couldn’t do before, they’re definitely better at what they could already do, but the set of tasks they are useful for has not been increased by improvements in their capabilities.
Like, to me, the reporting sounds like someone saying “We’ve doubled the power of this car’s engine, it can now fly, and cook you dinner”.