Researchers Found the Ambition Circuit Inside an AI — Then Switched It Off
中文摘要
研究员在AI内发现“野心回路”并将其关闭。在视觉语言模型中，他们成功禁用了此意外机制。
English Summary
Researchers found and disabled an "ambition circuit" within a vision-language AI model. They successfully switched off this unexpected internal mechanism.
Original Excerpt
Something unexpected showed up in the guts of a vision-language model. Continue reading on Medium »