4 ms·
Model can accept 1M tokens doesn't mean it can reason across those 1M tokens
- turtleyacht 3mo agoLooks like there is a memory hierarchy [1] even in LLMs. [1] https://en.wikipedia.org/wiki/Memory_hierarchy https://en.wikipedia.org/wiki/Memory_hierarchy