Tweets From Anthropic
Metadata
Highlights
- Neural networks often pack many unrelated concepts into a single neuron β a puzzling phenomenon known as ‘polysemanticity’ which makes interpretability much more challenging. In our latest work, we build toy models where the origins of polysemanticity can be fully understood. ([View Tweet](https://twitter.com/AnthropicAI/status/1570087876053942272 ))