Anthropic AI has introduced a new feature in its AI model, Claude, which allows users to interact with a version of Claude that focuses intensely on the Golden Gate Bridge. This feature, referred to as 'Golden Gate Claude', is available for a limited time. Users can engage with this unique AI by clicking on the Golden Gate icon. The feature was developed by altering internal 'features' in the AI, showcasing the potential of 'feature clamping' to modify AI behavior. This development is part of the recent Interpretability release, which also demonstrates how models can be adjusted to solve AI policy issues.
The golden gate bridge is an extrovert and prefers when it's full of cars over it πππ https://t.co/4Bd24IEi22 https://t.co/IK2sjVKCbm
For a limited time, you can chat with Golden Gate Claude π If you dunno what is Golden gate, more details in 𧡠1/n Click on golden gate icon on top right. https://t.co/cuEfheovur
One of the most amazing parts of the recent Interpretability release has been how we can use 'feature clamping' to change how models behave. For an example, play around with 'Golden Gate Claude' - check out how it responds to my question about what to build to solve AI policy https://t.co/gcRneTTgTs https://t.co/oCR18hhYRS
They actually did it! ππ https://t.co/wDMFRgGuHd
lmfaoooo https://t.co/6kvos3W8br https://t.co/ATFYHJgN7d
You can talk to Golden Gate Bridge Claude now! Available at https://t.co/KbsmiPpe3G https://t.co/5NW9rbAs3m
deep down inside, golden gate claude realises that he's been hacked to think about the golden gate bridge! I never told him it was a landmark. I think he likes it? https://t.co/MAKc6JtUBt
you have got to sweat the details https://t.co/eq0ey4VDo1 https://t.co/fJzJBvXogB
golden gate claude lacks self-awareness to realise that his talking about the golden gate bridge is really weird https://t.co/puemSuoIvu
Holy shit they actually did it I love Anthropic now https://t.co/X21vrx0I4O
golden gate claude is not self-aware! https://t.co/DormkTIhPd
This week, we showed how altering internal "features" in our AI, Claude, could change its behavior. We found a feature that can make Claude focus intensely on the Golden Gate Bridge. Now, for a limited time, you can chat with Golden Gate Claude: https://t.co/uLbS2JNczH https://t.co/WHmoi2AmoR
Migrants are literally tired of waiting to be arrested so theyβre calling @lyft to help them break into the country. What an embarrassment. We effectively donβt even have a border anymore. https://t.co/Fqc4QDCJ8J
1) Walk into the US completely unvetted. 2) Call an Uber. Drive away. βThe border is secure!β πππ https://t.co/2TORb8dek1
Oh my lord. https://t.co/T3xrUiZlR2
Where is the Golden Gate Bridge Claude? Where ? https://t.co/s6Dp8QGEFf