antipodal-solution-loses-joint-activation
IN premise — summaries/2026/08/24/elhage-2022-toy-models-superposition-chunk-5.md
Created 2026-08-25T02:58:00+00:00
The antipodal solution W=[1,-1] for two binary features in one dimension stores features in superposition but specifically loses the ability to represent the joint-activation pattern [1,1].
Summary
When two on/off features are squeezed into a single shared dimension using the antipodal encoding, they overlap but the system can no longer detect the case where both features fire together. This matters because losing that conjunction means any reasoning or pattern-matching that depends on "both conditions met at once" will silently fail, even though each feature individually is still stored.