>Officials in Beijing are concerned about the cyber capabilities of AI frontier models made by US companieshttps://www.businesstimes.com.sg/international/china-getting-more-anxious-about-mythos-ahead-trump-xi-meeting
>>109460586Why? Isn't kimi k3 the same thing but cheaper?
shit fake news post it to faggit
>>109460619lol
meanwhile Iran blows up more data centers and mythos is helpless against it
>>109460586Fake news, we know Deepseek is better because it doesn’t lie a lot. We know Kimi k3 can do anything that other American Ai failed to achieve.
>>109460619No, Kimi is only Fable level (Mythos level on everything except cybersecurity, AI research and biology), which makes it obvious that everyone is just distilling Anthropic.>But distillation takes months to doThere is no time requirement, only compute requirement, and also you don’t need to use only Fable to distill stuff. Distillation these days is more then just taking one existing expensive model and letting dumb blank model prompt it. On 80% of cases Fable is as good/barely better then Opus 4.5, only the 20% of possible tasks are what truly tests the model and what people use it for. You can train on Opus or even weaker model basic shit like making the LLM predict correct sentence structure or getting basic facts right, stuff that GPT3 was able to do, then distill more on more complex tasks like deducting code that compiles and understanding correct image embeddings. Only when you get to long horizon tasks and making CAD slop with three.js do you need to actually use Fable.
>>109461127schizo
>>109461127Go bribe some shabbos goys rather than shitting up the board, dario
>oh no, what if you enter text into the box, what if it outputs text? or images, or audio? and at marginally better quality than the current models in some domain while being worse in others? that would be terrible, i'm so anxious