Encodes positive confirmation and agreement signals.
What It Does
Boolean.Affirmation neurons activate when the model processes or produces positive confirmation signals. These include explicit agreement words ('yes', 'indeed', 'correct', 'exactly'), implicit affirmative constructions ('it is', 'they are', 'this confirms'), and semantic contexts where something is validated or accepted as true.
How It Behaves
These neurons tend to activate early in the processing pipeline, where token-level polarity is established before deeper semantic integration. They often co-activate with Identity neurons when a named entity is being positively attributed — for example, 'Einstein was a physicist' triggers both Affirmation and Identity.Person signals together.
Research Example
In Llama 3.1 8B, Boolean.Affirmation neurons fire strongly on prompts like 'The answer is yes' and 'Water does conduct electricity.' They remain quieter on hedged statements like 'Water may conduct electricity' — where Entropy.Ambiguity neurons take over instead.