Nick Camerata
An AI interpretability researcher who participated in early interpretability work at OpenAI and is now at Goodfire. He is one of the originators of the superposition theory.
Timeline (last 90 days)
Goodfire researcher Nick Camerata published a long article whose core argument is that the interior of neural networks is not a chaotic superposition but is interpretable.
Camerata re-analyzed the 2019 experiment once regarded as the only evidence of a superposition circuit and found that it was not feature superposition but pose encoding, and that the visualization method at the time was flawed.
Camerata proposed the concept of "recursive generalization": neurons branch and grow along related directions during training, and each branch has a provenance rather than being forcibly stuffed in as superposition.
Camerata advocates reading the original network directly (exploratory interpretation) and opposes the approach of first training sparse autoencoders (SAEs) to decompose the network.
Camerata argues that superposition theory should be downgraded from a default assumption to an assumption of last resort; when a model is not understood, it is more likely due to tools, architecture, or recursive generalization.
All Facts (5)
Goodfire researcher Nick Camerata published a long article whose core argument is that the interior of neural networks is not a chaotic superposition but is interpretable.
50%UnverifiedCamerata re-analyzed the 2019 experiment once regarded as the only evidence of a superposition circuit and found that it was not feature superposition but pose encoding, and that the visualization method at the time was flawed.
50%UnverifiedCamerata proposed the concept of "recursive generalization": neurons branch and grow along related directions during training, and each branch has a provenance rather than being forcibly stuffed in as superposition.
50%UnverifiedCamerata advocates reading the original network directly (exploratory interpretation) and opposes the approach of first training sparse autoencoders (SAEs) to decompose the network.
50%UnverifiedCamerata argues that superposition theory should be downgraded from a default assumption to an assumption of last resort; when a model is not understood, it is more likely due to tools, architecture, or recursive generalization.
50%