@Ma_WuKong Xiaomi is a mediocre, insignificant company not worth mentioning. If it had any real capability, the US would have sanctioned it long ago. Your tweet looks like a deliberate attempt to humiliate it.
@jokieliu Japanese are like language models with low parameter counts and short context windows—not only lacking world knowledge, but also prone to hallucination after just two turns of dialogue. Just give up communicating with them.
@infraflakes When I was in school, everyone wrote everything in C++. Now everyone writes everything in Rust. What’s there to complain about? Chill out, dude—you’re obsessively hating on Rust.
@Fulminatus56587@BristolGalt If you knew even a little bit about tech, you'd know this couldn't have existed in 2017, before Transformer was proposed! You idiot.
@HowToPrompt__ That's total nonsense. That's ancient history from 2024. After DeepSeek-R1, the math capabilities of LLMs have advanced by leaps and bounds—they're now on par with an IMO competitor.
@HoborgAcolyte@bdsqlsz As I understand it, this new image model shares the video model's VAE for encoding but uses a separate VAE for decoding, so it still has the potential to produce high-quality images. The only concern is during image editing—the encoder may lose details.