Many users agree million-token contexts cause context rot and degraded performance in models like GPT-5.5 beyond roughly 100k-300k tokens, while some dismiss the warning itself as sensational or obvious.
Based on 13 visible X reactions from 27 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@jeffreyhuber @0xCarnagee 2015 Chroma predicting decoder-only transformers would have a rot problem before decoder-only transformers existed is either incredible foresight or you invented time travel before you invented vector search.
@0xCarnagee Exactly, the context wars are total nonsense. Anything over 100k context starts to degrade big time, doesn't matter what model. Opensource is more like 50k
@0xCarnagee This is a fact that people don't like to hear. 384k is absolute max attention span.
@0xCarnagee I have yet to use a model past 300k tokens that ddnt seem like a drooling zombie
@jeffreyhuber @0xCarnagee 2015 Chroma predicting decoder-only transformers would have a rot problem before decoder-only transformers existed is either incredible foresight or you invented time travel before you invented vector search.
@0xCarnagee Exactly, the context wars are total nonsense. Anything over 100k context starts to degrade big time, doesn't matter what model. Opensource is more like 50k
@0xCarnagee I have yet to use a model past 300k tokens that ddnt seem like a drooling zombie
Many users agree million-token contexts cause context rot and degraded performance in models like GPT-5.5 beyond roughly 100k-300k tokens, while some dismiss the warning itself as sensational or obvious.
Based on 13 visible X reactions from 27 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@0xCarnagee @willccbb 🎯