Part 10/13:
They highlight that upcoming models—such as future iterations of hierarchical reasoning systems and advanced architectures like Gemini—will likely perform feats that humans cannot realistically replicate, owing to their speed and capacity to handle complexity. Examples include solutions to scientific problems that would require intractable amounts of expert effort for humans, such as predicting protein folding with AlphaFold, which simulated 500 billion years’ worth of research in months.