Sakana AI's Beyond Imitation paper introduces a contradiction benchmark and a multi-layered review system that caught 73.43% of core-claim errors with four reviews, versus 14.81% for the best baseline, but exact matches on retracted papers were 16.11%.
A ByteDance paper reports that DeepSeek V4 models can change answers when small token shifts place key information in weak slots created by chunked KV-cache compression. The flaw is described as structural and affects other models using similar compression.
A Global Threads study examines how the exploitation of enslaved Africans in the Caribbean was connected to the 1819 Peterloo massacre, tracing Manchester elites, the Portico Library and cotton workers' struggles.
WorldArena 2.0 Challenge at IROS 2026 unveiled final rankings on Sept. 16. Three tracks tested video quality, online RL environments and real-robot manipulation. Visincept, Zhongguancun Academy and Xiaomi Robotics led their tracks.
At IROS 2026 in Pittsburgh, Columbia University's Yunzhu Li presented FlexiTac, an open-source tactile sensor platform being commercialized by Analog Devices, while University of Queensland's Jen Jen Chung argued in a keynote that more data does not guarantee better robot decisions. Conference papers and workshops pointed to physics-aware foundation models, hierarchical control, tactile manipulation, and cheaper data collection as the field's main engineering directions.
Archaeologists have cataloged 597 wooden Neanderthal artifacts preserved as hollow molds at Abric Romaní in Spain, dating from 62,000 to 40,000 years ago, according to a Science Advances study.
A USGS study in Science describes an unusually detailed record of lava fountaining at Kilauea and compares it with ideas about what drives molten rock hundreds of meters skyward.
A Harvard study using Jellyfish analytics from over 700 software firms found AI coding tools produce more code but not more software, as longer human review absorbs the time saved.
A University of Southern Denmark ethics committee and Guinea-Bissau officials approved a trial that would withhold hepatitis B vaccine from some newborns, while the U.S. CDC reportedly set aside $1.6 million to fund it, prompting resignations and criticism from health experts.
OpenAI released nearly 400 AI-generated mathematical results across more than 700 manuscripts, according to The Verge. Mathematicians described the drop as overwhelming and said understanding it could take years, while fewer than half the manuscripts have been formally verified.
Nikon has disqualified the original first-place video in its Small World in Motion contest after the entrant acknowledged using AI-assisted processing, and named Nguyen Nam Nhat of Vietnam the new winner.
At MIT, a TechCrunch AI reporter shook hands with a $50,000 humanoid robot named Gabe, then attended Sherry Turkle’s “Artificial Intimacy” book launch. Researchers say anthropomorphism can support human flourishing but also create risks when chatbots are trusted over people.
After OpenAI released more than 700 mathematics manuscripts, the Human Mathematics Association urged mathematicians to stop working with the company, setting off a dispute over how AI-generated results are verified and credited.
New research presented in Boston suggests GLP-1 drugs may reduce biological age by two to three years in overweight and diabetic users. Use is rising, side effects range from gastrointestinal issues to hair and nail changes, and WHO says it does not recommend the drugs for children under 10.
A Lancet study found Google's diagnostic AI chatbot AMIE safely assessed 98 patients before emergency visits, with no conversations interrupted and 90% diagnostic agreement, according to QbitAI.
Margaret Hamilton, whose Apollo guidance software helped land astronauts on the moon and who helped establish software engineering as a discipline, has died at 90, MIT News reports.
Lei Feng Net has published the fifth, sixth and seventh layers of the Carbon-Silicon Daotong Cross-domain Migration Governance Code, covering meta-regulation, civilization-level risk control and final self-audit.