OPENAI'S NEW ALIGNMENT METHOD

Summary from the TLDR AI newsletter

OpenAI has developed a new alignment technique for reasoning models that enables them to think about grounded behavior goals, like those found in safety guidelines, and respond accordingly. It is a step forward in alignment and away from baking in alignment and preference at the same time.