While large language models' impressive capabilities in text generation have a potential to revolutionize workflows in many domains (e.g., software engineering), it is still unclear how they can radically change the scientific workflow and bring us to the new era of science with an exponential speed-up in scientific discovery. In fact, current large language models still fail on 65% of college-level scientific problems and show that understanding math and physics requires something beyond simple next-word prediction. At our lab, we depart from the current gold rush in AI that centers around generating ‘human-like’ contents and passing the Turing test. Instead, our aim is to build an AI system specifically designed for science that can find meaningful patterns in a large amount of scientific data faster than humans, formulate logical hypotheses, and communicate them effectively to humans to accelerate scientific discovery. Especially, combined with UCSB’s strength in physics and material science, our lab focuses on developing AI for quantum materials.