Goodfire Debuts Autonomous Tool to Map Computational Roles of AI Neurons

Goodfire debuted an autonomous tool called the "Neuron Scientist" to map the computational roles of individual neurons within large language models. Researchers claim the vast majority of neurons in models with as many as trillions of parameters are "legible." Goodfire provided descriptions for 1,500 neurons in the Qwen3.6 35B-A3B model.