Thalamus Gating: How Frank Decides What to Pay Attention To

Frank has 9 channels of sensory input updating continuously. The LLM has 4096 tokens. You can't inject everything. Solution: bio-inspired sensory gating.

Each channel has base gain, exponential habituation (tau 30-300s), and attention weights per cognitive mode. Salience breakthrough: long-suppressed channels that suddenly change bypass normal gating. E-PQ vigilance modulates overall baseline.

Result: Frank doesn't notice GPU temp when chatting. But if it spikes from 65C to 85C, the burst detector fires. Like not noticing your heartbeat until it races.

Based on Cromwell et al., 2008. Code: https://github.com/gschaidergabriel/ Project-Frankenstein/blob/main/servic es/thalamus.py