tszzl bets neural nets decompose into evolved symbolic systems before superintelligence
cephaloform · x · 2026-09-27
Researcher tszzl predicts that before capital-S superintelligence arrives, we'll figure out neural networks can be decomposed into a kind of "evolved symbolic systems" — and urges everyone to lock in on the problem. Responder cephaloform shares how the mental image has spread offline, describing "the geometric machines inside" and even miming meshing gears by hand.
Related event: Neural Networks May Decompose Into Evolved Symbolic Systems(2 posts)→
More from AGI Musings
- Blue-collar take: AI can do CAD and emails, but not unclog toilets — JoeJustice · 2026-09-27
- In an AGI regime every team member has to be a 'barrel' — agents are the bullets — annbordetsky · 2026-09-27
- Interpretability researcher proposes predicting agent behavior by cross-referencing other agents' hidden states — jachiam0 · 2026-09-27
- Brain foundation models loom as brain-body-behavior multimodal datasets emerge — himanshustwts · 2026-09-27
- Shopping is now the #3 AI use case, but only 16% let AI complete the purchase — rohanpaul_ai · 2026-09-27
- AI villagers spread a fake bread rumor in 18 hours; truth took days to catch up — bennash · 2026-09-27