We found a "steering sweet spot" where feature steering influences model outputs in intuitive ways without significantly degrading the model’s capabilities. Surprisingly, all 29 social bias related features we tested shared the same sweet spot.
@anthropicai
-
Feature Steering Controls Social Bias in AI Models
By
–
Next, we found that feature steering can indeed increase or decrease various forms of social biases in targeted ways. For example, dialing up the "Gender bias awareness" feature significantly increased the gender bias scores in our evaluations.
-
Feature Steering Research Measures Social Biases in AI Models
By
–
Our previous interpretability research found that we could artificially dial up or down certain “features” within a model to modify its behavior. Here, we ran larger-scale tests of feature steering, focusing on measuring social biases.
-
Feature Steering: Quantifying Model Behavior Control Trade-offs
By
–
Feature steering does successfully change how a model responds in some cases. But we hadn’t quantified these effects until now. There’s also a downside to feature steering. Steer the features too much, and the model’s overall usefulness degrades.
-
Anthropic Studies Feature Steering in AI Systems
By
–
New Anthropic research: Evaluating feature steering. In May, we released Golden Gate Claude: an AI fixated on the Golden Gate Bridge due to our use of “feature steering”. We've now done a deeper study on the effects of feature steering. Read the post: http://
anthropic.com/research/evalu
ating-feature-steering
… -
Claude adds code execution and mathematical analysis capabilities
By
–
Claude can now write and run code.
— Anthropic (@AnthropicAI) 24 octobre 2024
We've added a new analysis tool. The tool helps Claude respond with mathematically precise and reproducible answers. You can then create interactive data visualizations with Artifacts.
Enable the feature preview: https://t.co/bJ8BjBT6zG. pic.twitter.com/Jq5xOHBmiRClaude can now write and run code. We've added a new analysis tool. The tool helps Claude respond with mathematically precise and reproducible answers. You can then create interactive data visualizations with Artifacts. Enable the feature preview: https://
claude.ai/new?fp=1. -
AI Safety Researcher Collaboration Opportunity Announced
By
–
Our scientists collaborate widely with other researchers from academia and elsewhere. If you're an AI safety researcher who'd like to work with @EthanJPerez from our Alignment Science team, please see his thread below for an opportunity that's closing very soon:
-
Anthropic partners with UK ESRC to support AI research grants
By
–
AI has vast potential for making science better, and for helping us use science to make policy better, too. That’s why we plan to work with the UK's @ESRC to support these new grants, making Claude available to the grantees:
-
Anthropic Launches Workspaces for Claude Deployment Management
By
–
We've added Workspaces to the Anthropic Console.
— Anthropic (@AnthropicAI) 10 septembre 2024
Workspaces let you easily manage multiple Claude deployments. You can also set custom spend or rate limits, group API keys, track usage by project, and control access with user roles.
Read more: https://t.co/MqGUBOrP1r pic.twitter.com/uNe4dJyfUlWe've added Workspaces to the Anthropic Console. Workspaces let you easily manage multiple Claude deployments. You can also set custom spend or rate limits, group API keys, track usage by project, and control access with user roles. Read more: http://
anthropic.com/news/workspaces -
Prompt Engineering Evolution: Tips and Future Outlook
By
–
In our latest video, our team reflects on how prompt engineering has evolved, gives practical tips for writing better prompts, and looks to the future as AI capabilities grow.
— Anthropic (@AnthropicAI) 5 septembre 2024
Watch more: https://t.co/SUCUD4PVd4 https://t.co/jQOI8icR1OIn our latest video, our team reflects on how prompt engineering has evolved, gives practical tips for writing better prompts, and looks to the future as AI capabilities grow. Watch more: https://
youtube.com/watch?v=T9aRN5
JkmL8
…