Things are looking smoothly exponential for AI over the past several years, and I continue to think this is the best default assumption (until the AI R&D automation feedback loop eventually speeds everything up)