Is this test based on web rgb? I suggest use larger color space.

The speedup is amazing!

LLMs are very good at text reading. LLMs read tokenized text, while human use eyes to view words. Another scenario is that ChatGPT is good at analyzing cpp template error messages, which are usually long and hard to understand for human.

I have been wondering the reason why online generative AI can serving so many requests. This really gives me an explanation.