[bigyabai]

> squeeze max performance out of a system

Sounds more like SPIR-V and intermediary optimization for custom hardware or their current setup. I'm guessing sama wants to copy Deepseek's strategy, not build a language.

[rvz]

This makes the most sense given that OpenAI is moving away from Nvidia GPUs and developing their own chips for inference.

So there would need to be compiler support in Triton for their new custom inference chips that will be used run on them.

[tntpreneur]

This could be more likely. I don't know why but I always think that some can build better programing language for LLMs to develop.

[bigyabai]

I don't think a greenfield programming language would be easier for AI to use or more useful than the current selection of languages.