ResearchStudy finds optimal AI agent team size is 16; larger groups polarize and underperformResearchers from NTT's Physics of AI Lab and Harvard's Center for Brain Science tested multi-agent consensus on a task called the Flag Game and report an optimal team size of 16.2 min read
ResearchAnthropic used Claude Mythos to surface theoretical flaws in HAWK and a reduced AES variantAnthropic researchers used the Claude Mythos model to explore and identify mathematical weaknesses in the HAWK post‑quantum signature scheme and in a deliberately weakened AES variant; Anthropic says these findings have no practical impact on current computer systems.2 min read
ResearchEight case studies on the transformation of scientific computing and the role of human oversightResearchers examine eight industrial and academic case studies to explore what a current transformation means for scientific computing; the study stresses that human oversight, responsible…1 min read
ResearchAI identifies subgroup of rectal cancer patients who benefit from added irinotecanResearchers at University College London used a bespoke artificial intelligence to reanalyse data from a previous clinical trial and found a subgroup of 414 rectal cancer patients whose diagnostic biopsy images predict benefit from adding irinotecan to standard chemoradiotherapy.2 min read
ResearchInsilico's Abu Dhabi-linked AI team discovers a potential non-opioid pain drug, moves toward human trialsInsilico Medicine said researchers working from its Abu Dhabi facility used locally affordable AI compute to discover a potential non-opioid painkiller, its 31st preclinical candidate and the second found in the emirate.3 min read
ResearchGrok 4.5 stands out in value-for-money among cybersecurity AI modelsAccording to the latest benchmarks, Grok 4.5 proved to be the best value-for-money cybersecurity artificial intelligence model, being 10 times cheaper than Sol, 5.7 times cheaper than Opus 5 and 2.2…1 min read
ResearchSurveyed Experts Assign Notable Short-Term Probabilities to Several Catastrophic AI RisksA joint study by MIT and the University of Queensland asked 272 international experts to assess 24 AI-related risks for the 2025–2030 period.2 min read
ResearchMirrorCode benchmark measures AI ability to reimplement large programs from black‑box accessEpoch and METR published MirrorCode, a benchmark that evaluates how well AI systems can reconstruct sizeable software projects using only command‑line interaction with the original program.3 min read
ResearchScaling large models boosts real-world robotics performance, Anthropic and Sunday reportAnthropic and Sunday published results suggesting that larger, general-purpose AI models can materially improve the capabilities and generalization of physical robots.3 min read
ResearchAI-designed gene editor outperforms natural enzyme with far fewer off-target cutsA team led by Nobel laureate Jennifer Doudna and the startup Profluent used AI to design a gene-editing enzyme not found in nature, releasing it as OpenCRISPR-1.2 min read
ResearchHow AI Is Shifting Who Does What at WorkOpenAI Economic Research analysed over 800,000 ChatGPT messages from U.S.3 min read
ResearchUniversity of Pittsburgh project uses Meta vision models to advance assistive roboticsThe Human Engineering Research Laboratories (HERL) at the University of Pittsburgh, together with ATDev and a national consortium, are leading RAMMP — a $41.5 million ARPA-H–funded effort to create robotic mobility and manipulation platforms for people with disabilities.5 min read
