Q&A: New AI training method lets systems better adjust to users’ values
Reports and Proceedings
Updates every hour. Last Updated: 18-Jun-2025 09:10 ET (18-Jun-2025 13:10 GMT/UTC)
University of Washington researchers created a method for training AI systems — both for large language models like ChatGPT and for robots — that can better reflect users’ diverse values. It predicts users’ preferences as they interact with it, then tailors its outputs accordingly.
Researchers at Stanford Engineering have developed a lightweight, comfortable knit sleeve that uses pressure-based haptics to simulate touch, opening up new possibilities for wearable devices.
Reported recently in a new study published in Nature, a team of researchers, led by JILA and NIST Fellow and University of Colorado Boulder Physics professor Jun Ye, in collaboration with Professor Eric Hudson’s team at UCLA’s Department of Physics and Astronomy, have found a way to make nuclear clocks a thousand times less radioactive and more cost-effective, thanks to a method creating thin films of thorium tetrafluoride (ThF4).