March 4, 2026
Spikes, snark, and dopamine drama
Reverse-Engineering the Wetware: Spiking Networks and the End of Matrix Math
Brain-like AI vs Useful AI: commenters throw shade
TLDR: The article says the brain learns with spikes and dopamine, not calculus, and suggests this could reshape AI. Commenters fire back: some love the brain–AI crossover, others say usefulness beats biology, and many mock neuromorphic hype while warning we still don’t know the right network designs.
An engineer dives into how the brain really works—think spiky electrical pulses, predictive coding (your brain “guesses” what you’ll see), and dopamine acting like a reward signal—then asks what it means for AI. Cue the comments: the crowd splits fast. One camp rolls eyes at the “wetware revolution,” dropping the classic meme that neuromorphic chips are always 5 years away. Another camp loves the crossover moment where dopamine matches Temporal Difference learning from reinforcement learning, the brain-meets-computer plot twist. Skeptics clap back hard: “Usefulness over biology,” says the utilitarian crew, insisting that copying brains isn’t the point—shipping tools is. A third group nitpicks the science, noting Stanford research that STDP (those spike timing rules) isn’t a simple cause-before-effect story and real synapses get weird. Meanwhile, a meta-joke breaks out when someone asks why they’re reading an LLM summary of a piece about ditching LLM math—oh, the irony. For curious readers, Predictive Coding, STDP, and Temporal Difference are the big ideas. The spiciest thread? Whether brain-like chips matter if we don’t even know the right network layout. The vibe: hype vs homework, spikes vs spreadsheets, dopamine vs deliverables.
Key Points
- •The article contrasts matrix-math, backpropagation-based AI with the brain’s computation, motivated by practical AI engineering experience.
- •Human perception is framed as predictive coding: top-down predictions compare with bottom-up input, propagating only prediction errors.
- •Backpropagation is argued to be biologically implausible due to unidirectional synapses and the credit assignment problem.
- •Local learning via spike-timing–dependent plasticity (STDP) adjusts synapses based on precise pre/post spike timing.
- •Dopamine is introduced as a third factor linking local plasticity to goals, with a reference to temporal-difference learning, though details are truncated.