2 articles
- YesterdayAnthropicEarliestAutomated researchers can reliably mitigate alignment failures
- YesterdayAnthropicEvaluating feature steering
← Back to the boardNews
Evaluating feature steering
Why it’s here: Anthropic's interpretability research on feature steering for mitigating social biases in Claude models is AI policy-adjacent technical work relevant to federal AI safety, standards, and evaluation debates, with no outside-scope or roundup issues.
Trace across the boardItem
YesterdayAnthropicEvaluating feature steeringAnthropic says Claude autonomously researched and trained fixes for ten categories of alignment failure, beating 28 human safety researchers and transferring to larger models. 31 linked
Coverage
1 outlets · first seen yesterdayConnected
1 links- NewsUnderstanding and addressing AI harmsBoth involve Anthropic, Claude 3.7 Sonnet
About
- Anthropic