Concerns about training AI on synthetic data

An interviewer sharing a discussion of self-improving AI says the guest considers long-standing concerns about training models on synthetic data overblown. The interviewer offers tentative agreement: Given some of the latest models, the guest “might be right.”