Ex-Apple dev built KnobNet to manually tune AI weights with real knobs
An engineer from Apple and Meta built a physical mixing console for neural networks, turning terrifying linear algebra into an arcade game.
Former Apple and Meta developer Tyler Suard created a physical controller called KnobNet that replaces standard lines of code with tactile potentiometers. The hardware setup connects directly to a lightweight neural network designed for basic digit recognition. Each physical knob functions like a analog volume slider, wired to control a single mathematical weight inside the model.
Twisting the knobs alters weight parameters in real time while an onboard monitoring station provides instant visual feedback. Built-in displays stream live input matrices, hidden layer neuron activations, and shifting probability outputs, making error spikes look less like a computer science thesis and more like a audio equalizer going haywire.
The system intentionally avoids scaling up to modern multi-billion parameter behemoths. Its sole objective remains teaching foundational backpropagation dynamics to humans through physical interaction instead of opaque Python scripts.
The transition from mysterious black-box algorithms to plastic audio dials proves that modern artificial intelligence is ultimately just millions of tiny volume knobs hidden behind marketing buzzwords.
Source: Medium
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.