The failure sparked debate over self-hosting AI models.
heard that anthropic eased up the classifiers, so i thought i'd try fable again on a very simple web app. after spending $200+ on a single request (please do a code quality review of this app) it was classified as unsafe and failed i think i am officially done trying here
Many users criticized Anthropic's safety classifiers for inconsistently blocking high-value prompts and code reviews due to hubris and overreach, though a few praised the model and remained eager to try it.
it's very funny considering that their classifiers are basically a skill level check. if you're deemed too good, or the value of what you are asking is too high, they will withhold the tokens... only useless prompts for you sir!
this is basically why xjdr has his own gpu cluster
heard that anthropic eased up the classifiers, so i thought i'd try fable again on a very simple web app. after spending $200+ on a single request (please do a code quality review of this app) it was classified as unsafe and failed i think i am officially done trying here
you actually cannot give this up to another corporation. you need to control your own models, your own data. it is such an insane amount of power to let someone wield over you. a corporation whose stated goal is to cannibalize your business...
it's very funny considering that their classifiers are basically a skill level check. if you're deemed too good, or the value of what you are asking is too high, they will withhold the tokens... only useless prompts for you sir!
this is basically why xjdr has his own gpu cluster
@_xjdr What do u think about Kimi 🙁K2.7 coder
@_xjdr nah they never mentioned easing up on the classifier, instead they claimed they would make it more transparent when they degrade/refuse basically
@ChrisRinvesting im still an AI researcher. understanding frontier capabilities is important for that effort
I don't think that ant people have bad intentions, moreso that their hubris causes them to make awful missteps & miscalculations. It's a really important time right now to not create negative sentiment, because you won't be at the fronteir forever
@sun_hanchi Haven't tried it yet but am eager to
@egebasoner nope, simple web app / CRUD . nothing AI / ML / related at all. i intentionally gave it the most vanilla and simplistic repo on my machine
@_xjdr Why are you people giving them money
i mean.. who owns your compute? 🌕
@_xjdr Anything that could make it think you were doing ML work or something?
@_xjdr There are still people who are defending this btw
@_xjdr I used the same prompt on the same website 6 times yesterday Every other one would either pass, or switch to Opus.
No reason to it at all
@_xjdr Must've been the wind.
@_xjdr Simples explanation: they aren't safety classifiers, they are *subject* classifiers.
@_xjdr I asked for a definition of a word and it rerouted me
@_xjdr I guess. Maybe we can designate one person to this task to save on api cost
@_xjdr phew, its a good model sir what can i say, didnt run into problems, but have not touched any ml repo yet