Everytime @Alibaba_Qwen release new model everybody starts hating, "it thinking too much", "it did this and that", ..at the end everybody using these models until next release. Boys and girls, SHUT THE FUCK UP
@MiaAI_lab@ashxhart@coldniko@bijanbowen
@noe_achache I meant for situations where all the LLM got is basically url without description, but bigger / frontier models can find out from the url what it does (plus some trying)...thats something I wouldnt let for example qwen3.8-27b let to touch. MCP-server, its not really "server" btw
@sattyyouneed The worst part is, when you trying to explain its just api wrapper to somebody, who just started using AI and they contradict you and call you stupid, you can learn a lot about their personality..ππ
@nayuengin@gringopolitico How did you build it? Why sorry to use codex-cli? I'm building whole k8s platform on it for last 4 months...just wonder, what was the problem?
@yasakani_dev@StewRoseInnes Not my bussines, all I did I took your "quality" model and find out it cant even properly discover and use tools that nvfp4 has no problem with. If you want to publish your models, test them before release. Sorry..
@jayweeldreyer@RayaneRachid_ Thats old (in AI terms, everything is developing so fast) thing, I guess if you have long term memory and its your personal assistant its ok. But you cant have it in company with local models. Thats suicide...terminology should differ between:
-assistants (this)
-harness
-chat