Model can accept 1M tokens doesn't mean it can reason across those 1M tokens

1 comments

• turtleyacht 13 hours ago ago

  Looks like there is a memory hierarchy [1] even in LLMs.

  [1] https://en.wikipedia.org/wiki/Memory_hierarchy