all about?
I compared image generation between chatgpt, grok and deepseek with a very simple command: generate an image of a rainy day in Tokyo. Grok immediately generated a panel of 4 images each with a different scene of a rainy Tokyo street. ChaptGT took a few seconds to generate one. But DeepSeek failed to generate any image after producing a useless text of gibberish on its thought process of how it could fetch information to generate such an image. But no image was generated in the end. You don't have to just take my word for it. Just try youself.
Then I found out through ChatGPT that DeepSeek currently has no image generation cababilities at all. So what is the big deal on it being cheap and low cost ?!. To my knowledge, image (and video) generation with a simple command is the most challenging task that would have required tremendous amount of data and training to make sure the image (or video clips) is not only vivid and ethetically appealing, but also accurate with regard to all physical details such as light sources causing differnt shades of brightness/darkness, common laws of physics, biology and all other sciences, etc. No wonder it's cheap
So what's all hype about DeepSeek being a low cost AI model, not to mention that it uses the so-called data distillation process (which is to me another way of saying pillage and theft of pysudo data set generated from other AI models)
In terms of the low cost in hardware, even Elon Musk doubts that DeepSeek uses low end H800 GPU chips from NVIDIA. It may have very well used as many as 50,000 H100 GPUs that the Chinese government secretly bypassed all the restrictions to obtain from third party vendors.
You just can't trust them on anything, period!
Let's wait and see what happens after all the hooplas quite down.