A neural network made of just 29 logic gates can play Mario, pointing to ultra-low-power AI
udmrzn · x · 2026-09-09
A Japanese team (Tebasakilab, Unexplored IT) showed a differentiable logic gate network (DLGN) playing Mario with only 29 AND/OR/XOR gates.
- Since logic gates aren't differentiable, training uses "soft gates": all 16 gate types approximated (AND as multiplication, OR as a+b−ab), blended with learnable scores, then each node is hardwired to its top-scoring gate
- Inference is pure 0/1 switching: nanosecond latency, extremely low power, no GPUs — suited to FPGA, drone controllers, appliances, and offline edge devices ("physical AI")
- Limits: soft-to-hard conversion loses accuracy; it won't replace LLMs
- The author predicts 2026 will see DLGN variants of Attention/ViT/CNN/RNN, with papers already appearing and Google exploring differentiable logic cellular automata
Related event: Japanese team builds Mario-playing AI with just 29 logic gates(2 posts)→
More from Embodied
- AnyRank ranks robot policies reliably with far fewer real-robot trials — pappasg69 · 2026-10-27
- Podcast: Is Tesla Building a Robot Army? The Optimus Legion Theory — MarwaEldiwiny · 2026-10-27
- Full Cybercab teardown video live: Tesla team walks through the engineering — elonmusk · 2026-10-27
- Open tactile skins like Johns Hopkins' OpenTac ease robot touch design, durability is next bottleneck — _krishna_murthy · 2026-10-27
- Farming robotics doc series debuts: five machines, one operator, wine grape growers signed — Scobleizer · 2026-10-27
- Agent Buddy: A Desk Gadget That Shows What Your Background AI Coding Agents Are Doing — DanWahlin · 2026-10-27