Neural networks are clean
Goodfire published evidence that transformer models internally represent concepts as clean, interpretable directions in activation space, not distributed noise.
Read the note 1 min read
From the notebook
4 notes on this topic.
Goodfire published evidence that transformer models internally represent concepts as clean, interpretable directions in activation space, not distributed noise.
A graphics engineer got neural texture maps working with ES optimisation. No gradients, no autodiff, just mutation and survival.
A developer trained MNIST digit recognition using only SQL queries. No Python. No frameworks. Just recursive CTEs and window functions.
Researchers trained particles to form complex shapes without central control. Each particle runs the same neural network, learns local rules, and the swarm organises itself.