ideasthesia

Sign in Subscribe

Ryan Collison

I write about cool science and share it here.

Neuromorphic Chips: Silicon Learns from Neurons

Neuromorphic Chips: Silicon Learns from Neurons

Your brain runs on 20 watts. A GPU training a language model runs on tens of thousands. Neuromorphic chips attack that gap by mimicking the brain's event-driven architecture — processing only what changes, only when it changes, using spikes instead of clock cycles.

Organoid Intelligence: Wetware as Compute

Organoid Intelligence: Wetware as Compute

Cortical organoids in a dish can learn, adapt, and solve problems — at energy costs that make silicon look wasteful by six orders of magnitude. Organoid intelligence isn't a metaphor anymore; it's a research program with working prototypes.

The Brain's Impossible Efficiency: 20 Watts

The Brain's Impossible Efficiency: 20 Watts

GPT-4 needs a small power plant. Your brain needs a dim lightbulb. The human brain's 20-watt efficiency isn't just impressive — it's still the most embarrassing gap in AI.

Landauer's Limit: The Physics of Erasing Information

Landauer's Limit: The Physics of Erasing Information

Landauer's Limit: The Physics of Erasing Information In 1961, a physicist at IBM named Rolf Landauer asked a question that seemed almost philosophical: What is the minimum energy required to perform a computation? The answer he found wasn't philosophical at all. It was thermodynamics. And it

The Compute Ceiling: Why AI Hit a Wall

The Compute Ceiling: Why AI Hit a Wall

Training costs double every six months while Moore's Law fades. The compute ceiling is real, and the next AI breakthroughs may require rethinking the entire architectural stack.

The Intelligence of Energy

The Intelligence of Energy

Intelligence isn't free. Every inference, every weight update, every spike costs joules — and the physical limits on energy efficiency may be the real ceiling on AI scaling. This series examines what happens when you take the energy cost of thinking seriously.

Synthesis: Attachment as Coherence - A Unified Theory of Relational Regulation

Synthesis: Attachment as Coherence - A Unified Theory of Relational Regulation

Attachment isn't just emotional bonding — it's two nervous systems synchronizing into a shared dynamical state. Polyvagal theory, predictive processing, and geometric models of affect all point to the same underlying structure.