Smith et al.: present automated-lab limits and proposed latent-capability tiers
Posts here