AI Research Atlas

Golden Gate Claude

Anthropic · 23 May 2024

For 24 hours the public could chat with a Claude whose Golden Gate Bridge feature was turned up, a live demo that features steer behavior.

Companion to Scaling Monosemanticity. Amplifying one learned feature made Claude 3 Sonnet steer every answer toward the bridge, even a love story about a car crossing it. A vivid proof that interpretability features can steer behavior.

Date
Thursday, 23 May 2024
Lab
Anthropic
Kind
feature
Access
app only

Research demo, taken offline after about 24 hours.

Sources

  1. www.anthropic.com/news/golden-gate-claude

This record was checked against its sources on 6 October 2026. How we check

Related