ICTS

We are at the early stages of a fundamental shift, where new classes of AI tools have deep expertise in a range of domains and can autonomously generate solutions that match or sometimes exceed human-designed approaches. The application of such tools to scientific endeavors holds immense potential towards driving progress and discoveries. This talk will cover two aspects of a flywheel that can accelerate scientific research. One is the evaluation of the capabilities of such LLM and AI  models in handling the complexities inherent in scientific tasks. The second looks at an approach where an AI system can be guided to create expert-level solutions to complex scientific problems semi-autonomously to significantly save time and effort while still resulting in performance that compares or exceeds manually engineered ones. I will also share some examples of results of the system to diverse applications ranging from single cell RNA seq batch integration to CO₂ monitoring and weather related modeling.