Ask HN: What's the point of training bigger language models?

2 pointseclectic292 comments

What's the point of training bigger language models when you can't even serve (inference) them economically? What am I missing here?

> 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.

The number of parameters correlates with the believability of the result being from an intelligent human being.

Easy to inference/small model = poor results