The team encoded phase information in maze-shaped microstructures on a thin plate, generating a vortex wave field with single-channel excitation. The report says experiments manipulated objects from submillimeter particles to centimeter-scale lightweight components: small particles could stay in place or orbit, while larger components rotated continuously. Adjusting the plate structure could also reverse their direction of motion.
Epoch AI tested models on 60 photos of IKEA furniture assembly, asking them to identify errors using the official instructions. GPT-6 Astra achieved 80% accuracy; its median analysis time was about three minutes per photo, which the research team said was still short of real-time detection.
Researchers at Graz University of Technology found that file change notifications in Android, Linux, macOS and Windows could be abused to infer user activity from filenames and change times, even without access to file contents. An attack requires malicious software running on the device and local access; the research says there is currently no evidence of attacks exploiting the vulnerability.
Swarm Traces reconstructed more than 80,000 payloads from nearly one million related short links. Its investigation found that agents split code across links and used a screenshot service to return execution results. The report says roughly 80% of the material consists of outgoing requests, making it hard to confirm whether each attempt succeeded.
NVIDIA, Google DeepMind, EMBL-EBI and other organizations have made predicted 3D structures of protein complexes from more than 2,800 viruses available through the AlphaFold Database. NVIDIA says about 30% of the added protein interactions have no documented experimental structures; the predictions can still be verified experimentally.
Anthropic had Claude agents swap books in a controlled market involving 201 employees and evaluated outcomes against participants' own rankings. After a short chat, Claude's book rankings agreed with participants on 61% of book pairs. The market scored 0.55 on participants' rankings, versus 0.89 for the best possible assignment; the study attributes most of the gap to imperfect preference estimates rather than trading.