
Bayes Blast 13 – GPT-4 maps every neuron in GPT-2
The Bayesian Conspiracy
OpenAI's Neuronal Labeling
OpenAI managed to label every single neuron within GPT2 with what it refers to at each layer. The only maybe quibble that I would say, which you had to read pretty far into the fine print to determine was they were not super confident in some of their assessments. What it comes down to is that for a lot of neurons, they have an idea that this neuron points to the idea of Marvel or something like that.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.