Stanford researchers led by James Zou built a system called Paper2Agent that automatically transforms a scientific paper's text, code and data into an AI agent acting as a stand-in for its corresponding author. The tool deposits a paper's materials onto an MCP server, has AI agents build tools that apply the paper's methods to new data, and lets scientists query the resulting agent in plain language via any large language model. In one test, it built an agent for the AlphaGenome paper in about 45 minutes for $14, and that agent answered genetics questions with near-perfect accuracy.
nature.com
· 2026-09-16
Researchers led by Sergiu Pașca at Stanford transplanted lab-grown human brain tissue into newborn mice missing part of their brains, and the tissue expanded to fill much of that space while forming functional connections with the host nervous system. Published in Nature on September 16, the study represents the most extensive fusion of human neural tissue into a living animal achieved to date, though behavioral tests showed the mice did not gain enhanced cognitive abilities.
nature.com
· 2026-09-16
Stanford researchers, working with the Terminal-Bench team and scientists across disciplines, launched Terminal-Bench-Science, a benchmark testing AI agents on real scientific research workflows. The initial version includes 70 tasks spanning life, physical, Earth, mathematical, and engineering sciences, with Claude Opus 5 running Claude Code topping the leaderboard at a 30% resolution rate.
terminal-bench-science.ai
· 2026-08-28
More than 600 teams, mostly from Chinese universities and robotics firms, brought humanoid machines to Beijing's second World Humanoid Robot Games from August 22 to 26. Alongside sprint, jump and combat contests, organizers added scenario-based challenges like shelving books and making beds, scoring autonomous performances higher than remotely operated ones. Robots set records in speed and jumping, but researchers noted that everyday chores remain far tougher for machines than flashy athletic feats.
nature.com
· 2026-08-27
MIT Technology Review spoke with cognitive scientists comparing how babies and large language models acquire language, finding that infants learn to speak proficiently after hearing only 10 to 30 million words, while LLMs require vastly more data to approach similar fluency. Stanford's Michael C. Frank noted that training a model like GPT-2 on the same word count a toddler hears produces a 'nonsense generator,' not coherent speech.
futurism.com
· 2026-08-25
GoKawiil