Entropy neurons not fitting a specific sub-type.
What It Does
Entropy.General neurons carry uncertainty and noise signals that blend or transcend the specific sub-types. They activate on general epistemic hedges ('it is possible that', 'one could argue', 'this is debated'), mixed uncertainty (something that is simultaneously ambiguous and low-confidence), and the background entropy present in all natural language due to the inherent imprecision of words.
How It Behaves
Entropy.General is the largest Entropy sub-type by count and shows a broad layer distribution with slight early and late peaks. Despite being the largest Entropy sub-type, it is still tiny compared to Symbol or Number neurons — even the 'general' catch-all for uncertainty is rare. This rarity of Entropy neurons overall is the key finding: models represent uncertainty with a small and fragile system, which is part of why they are susceptible to confident confabulation when that system fails or is bypassed.
Research Example
In Gemma 9B, Entropy.General neurons activate on 'it is widely believed that' and 'most historians agree' — both epistemic hedges that signal the model is reporting a consensus view rather than a verified fact. Models that produce 'it is widely believed that' followed by a fabricated 'consensus' have Entropy.General neurons active during the hedge but not during the fabrication — the hedge is recognized as an uncertainty marker, but it doesn't prevent confabulation of the uncertain content.