βReal innovation works like evolution.
First, you collect stepping stones, then build on them over time. With the right evolutionary framework, language models can become powerful tools for scaling discovery. The real breakthrough will come not just from optimizing solutions, but from evolving both the problems and the solutions together, creating more diverse pathways and unlocking a truly open-ended process of innovation.β
Thoughts from Robert Lange, founding researcher at Sakana AI on The Machine Learning Street Talk podcast discussing ways how AI will make real scientific progress.