News for hackers. Stuff that matters.
Re:The Secret Sauce behind 100K context window in LLMs: all ... by cma on Sunday June 18, 2023 @12:46AM (#36377402)
> I noticed that the model often disregarded the instructions if placed beforehand. It's clear that the model doesn't allocate the same level of "attention" to all parts of the input across the entire context window.

This would be similar with humans if everything was given verbally.

There are no comments on this comment yet.