LLMilestonesA milestone timeline of AI and large language models

Act IV · Multimodal, open source catches up

PaperAnthropic

Scaling Monosemanticity

Extracted millions of interpretable features from a production model, Claude 3 Sonnet, such as one for the Golden Gate Bridge.

Why it matters

Interpretability reached a real frontier model for the first time, showing that a model's concepts can be found and even steered.

Built on

Claude 3

Source: transformer-circuits.pub ↗

See it on the 3D timeline →