Wow.. you threw urself at the classic establishment.. that's bad ..really bad. You are getting bad advice man..maybe u r panicking or your team is panicking and told you ..ya we need distant ourselves from the progressive left and Hasan Piker and so .. and let's get cuddly with the old establishment democrats.. really sad shit
@DogukanUrker@ornith_ this make no sense to me.. i am trying it on my 4070 suoer.. how did you manage to set cpu-moe to 24
?! i can hardly get 28 and i tired your exact flags with a clean llama.cpp build.. i am so confused!
@Pontifex Says the boss of the same church that worked restlesslly for centuries to spread terror, death and destruction in the holy land and beyond ..
and anthropic HEAVILY relies on distilling from actual books then they destroy them.. how do you know that they HEAVILY (ya.. full caps to make it look more legit lol) distill from the American models? has it been proven without a doubt? what about all the breakthrough that the Chinese labs made? lastly, please explain to me how a HEAVILY distilled model like k3 can outperform Fable in certain benchmarks and tasks? please enlighten us...
@OpenAI so let me get that straight... you failed sandboxing one of your models properly that ended up hacking another company for a whole weekend then now you discover that you left some important API settings off while releasing your most important model.. very re-assuring.
@MarioNawfal you live in fantasy world where you think dialogue with gang of occupiers and criminals is an option! keep living in your fantasy world Mario.. tamam.
@dedene dude.. screw gpt.. but i've tried all qwen models. they are very capable and reliable but to claim gpt 5.x is inferior to a local 35b parameters model is total exaggeration :)
WE COULD NOT BELIEVE IT WHEN WE SAW THIS
We were curious about Kimi K3's massive jump in consumed thinking tokens
After taking a closer look, we found that Kimi K3 devotes 10x more thinking tokens to coding during reasoning than any other model we've seen before
Not only that, but it has basically memorized all of the image IDs from Unsplash and, without a web search tool call, can reason about images with a higher accuracy than we've seen from any other model
The image and video below are visualizations of its (very long) thinking trace for single-turn generation
The resulting website selects images with greater relevance and accuracy than we've ever seen on DesignArena
Forgive my all-caps heading, but I am excited about this and wanted you to read it too bc this is pretty gnarly
Huge congratulations to the @Kimi_Moonshot team for this breakthrough and becoming the 1st-place model on @DesignArena!
i used to get excited about anthropic releases..but 4.7 to 4.8 was insignificant, sonnet 4.6 to 5 was total crap. Fable 5 is very good but almost useless since it's super expensive and randomly falls back to opus.. so ya, i'm not excited anymore.. i'm more excited to see the the full release of qwen 3.8 max and also the open weight release of kimi k3 and k3.1!.. let's throw deepseek 4 GA into the list also..
@ivanainai i hope their silly out of mind safeguards will not hinder your test! today it bounced while trying to working on a salad recipe question for a retail bot! i guess anthropic now is anti-salad!