Aim for insights over minor improvements

Currently, there are too many research papers showing minor "state-of-the-art" benchmark improvements for LLMs. These are not interesting.

Instead, focus on papers that yield new insights on models.

Linus @thesephist

References