it still goes into a loop if you ask it about tiananmen square
[You must be logged in to view images. Log in or Register.]
huggingface the website i posted awhile back already has the uncensored Deepseek r1 they retrained without all the Chinese shit in it
https://huggingface.co/huihui-ai/Dee...abliterated-v2
only 50 people have downloaded it so far because its literally whatever
[You must be logged in to view images. Log in or Register.]
you can't actually "see" that 9% improvement over Mistral which is one of the ones i tried with my chatbot you'd have to be asking a specific question prolly math or coding and compare it's answer to the other models to even notice
it's like horsepower in a car, you can't really feel couple HP from some bolt on ebay part
i had mine dual boxing models for a bit anytime i would ask it something it would respond twice using each model, and it was literally the same answer just worded differently with Mistral and Google
so they just shot their shot and got 2% over their last model, 3.5 to 4 for ChatGPT was a small gain also, 5 will beat Deepseek and the next Gemini after 2.0 will as well.
[You must be logged in to view images. Log in or Register.]