Science
# science
# physics
# math
How Alex Lupsasca Learned to Trust AI for Real Physics
Physicist Alex Lupsasca approached AI the way he approaches big claims in science: by stress-testing it. In early 2025, ChatGPT proved useful for academic housekeeping and for turning rough, multilingual drafts into clear scientific prose, but he did not view it as a tool for producing verifiable predictions.
Mathematician Ernest Ryu, one of more than 1 million weekly ChatGPT users working on advanced science and math topics cited in a new OpenAI report, began testing whether language models could translate real-world scenarios into rigorous optimization problems and learned early systems still missed key constraints. After reasoning models arrived, he used ChatGPT as a fast, iterative collaborator to solve an open problem related to Nesterov acceleration across three late-night sessions, verify the proof to publication standards, and later bring that same math-first mindset to OpenAI’s synthetic data team.
# math
# science
UC Berkeley linguist Gašper Beguš frames sperm whales as an “alien intelligence” in the ocean and argues that decoding their communication could reshape human ideas about culture, mind, and moral status.
# science
Junevity is a new biotech co-founded by entrepreneur Rob Cahill, UCSF scientists Janine Sengstack and Hao Li, and veteran CEO John Hoekman, with a mission to reprogram aging cells using small interfering RNA (siRNA). Cahill’s path from software founder to bioinformatics and lab work was shaped by family experiences with aging and Parkinson’s, pushing him toward preventive, disease-modifying therapies. From day one, Junevity made AI part of its operating system: every employee gets ChatGPT, and OpenAI models augment the company’s genomics and machine-learning pipelines. The team estimates its first development candidate, a liver-targeted siRNA program for Type 2 diabetes and obesity, can reach key milestones at 2–6x lower cost and 2–3x faster timelines than typical industry norms.
# startup
# biotech
At Scripps Research, PhD candidate Marco Uytiepo studies how neural circuits store experience by analyzing immense electron-microscopy datasets and reconstructing brain wiring across millions of synapses. Trained primarily as an experimental scientist, he taught himself to code and now uses ChatGPT to accelerate the unglamorous work that makes modern neuroscience possible: building data pipelines, debugging scripts, and automating 3D reconstruction workflows across tools like MATLAB, Python, and Blender. Work that previously required days of iteration can be reduced to rapid cycles that produce usable code and clearer visualizations, supporting faster hypothesis testing and more experiments within a single project.
# research
# science
# healthcare