So why are they continuing when they say alignment isn’t working? If for all these years they have feared misaligned AI, and now they say their alignment science is less than good, the most obvious conclusion is to stop pushing in the same direction.
This doesn’t mean they fold as a company, or give up on AI, or stop working. But what exactly is the explanation for continuing to train ever larger frontier models in the current context?
This doesn’t mean they fold as a company, or give up on AI, or stop working. But what exactly is the explanation for continuing to train ever larger frontier models in the current context?