> What's the point of training bigger language models when you can't even serve (inference) them economically? They're built to impress people and push the bou

> What's the point of training bigger language models when you can't even serve (inference) them economically? They're built to impress people and push the boundaries of can be done with ML. Economies, accessibility, scale - those things come later.