TechCrunch
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.
Given 10 benchmarks for specific misaligned behaviors, the automated systems were able to improve performance on every single one without degrading overall performance.
Originally published by TechCrunch. Read the full story at the source →
