Hacker News Reader

Boringness classifier! Pretty cool because this implies the large models already know what is useless and what isn't.

> NAMMs are trained separately from the LLM and are combined with the pre-trained model at inference time […]