An Anthropic researcher just gave us a peek at self-improving AI
Given 10 benchmarks for specific misaligned behaviors, the automated systems were able to improve performance on every single one without degrading overall performance.
خواندن متن کامل در تککرانچبه زبان اصلی، در سایت ناشر باز میشودمتن اصلی (انگلیسی)
An Anthropic researcher just gave us a peek at self-improving AI
Given 10 benchmarks for specific misaligned behaviors, the automated systems were able to improve performance on every single one without degrading overall performance.