Latest

Quantum cognition doesn't claim neurons are quantum computers. It claims the math of quantum probability — superposition, interference, non-commutativity — better describes human judgment than classical Bayesian models. The evidence is surprisingly strong.

Donald Hoffman's radical claim: evolution didn't wire us to perceive reality accurately — it wired us to survive. Your senses are a fitness-tuned user interface, not a window onto the world.

Active inference isn't just a brain theory — it's a framework for building AI that acts to minimize surprise rather than maximize reward. Here's what that looks like when applied.

It turns out you don't always need a bigger model—just one that's allowed to think longer. Test-time compute scaling lets smaller networks outperform larger ones by trading training cost for inference depth. The economics are strange and the implications are significant.

Standard RAG retrieves similar text. Graph RAG retrieves connected knowledge — following relationships across entities to answer questions that require reasoning, not just lookup.

Traditional chips run on clocks. Neuromorphic chips fire in spikes—mimicking the brain's event-driven architecture and running up to 1000x more efficiently than GPUs for certain tasks. Intel's Loihi and event-based sensors are making this real.

What if computing in 10,000 dimensions made your AI more robust, not less? Kanerva's hyperdimensional computing flips the curse of dimensionality into a feature.