ResearchClaude Mythos 5 designs more efficient quantum circuits in the STEM Fellows ProgramAccording to a recent report from the STEM Fellows Program, the Claude Mythos 5 artificial intelligence designed quantum circuits from simple, natural-language instructions that often outperform the best published solutions, in some cases reducing resource requirements by an order of magnitude or more.1 min read
ResearchAccording to Nick Cammarata, agents contradict polysemanticity and improve interpretability and alignmentIn a recent paid Latent Space post, Nick Cammarata claims that so-called agent-based methods substantially contradict the assumption of polysemanticity and significantly improve models' interpretability and alignment.1 min read
ResearchOpenAI provides $5 million to support examination of AI-produced results in mathematical researchOpenAI is contributing $5 million to the Simons Foundation’s new initiative so that mathematicians can examine, understand, and build on results produced by AI; the funding will finance events, programs, and access to models and computational resources.1 min read
ResearchAnthropic announces two new research fellowships, applications open until October 28Anthropic is announcing two new four-month fellowships: STEM Fellows (natural sciences and engineering) and Conceptual Reasoning Fellows (philosophy, cognitive science, economics, politics and law).1 min read
ResearchLeash Bio turns DNA‑encoded libraries into ‘lab‑in‑the‑loop’ datasets and launches MinosLeash Bio, led by Ian Quigley, has refined DNA‑encoded libraries (DELs) to produce clean, reproducible high‑throughput binding data suited for large‑scale modeling.4 min read
ResearchMedical researchers produced ten manuscripts using Meta's AI tool based on a five-million health databaseMeta's health team tested how AI speeds up medical research by combining a Muse Wonder encoder framework and a licensed 5 million electronic health record database: a small expert group produced ten manuscripts in about two weeks, including three by Paddy Ssentongo.1 min read
ResearchStudy: Workers' Fear of AI Linked to More Counterproductive Workplace BehaviorA Washington State University study of 18 industries finds that employees who fear artificial intelligence are more likely to neglect duties, break workplace norms and engage in counterproductive behaviors.3 min read
ResearchAnthropic Institute outlines research agenda on AI’s economic, social and security effectsThe Anthropic Institute (TAI) published its research agenda on May 7, 2026, describing four priority areas for studying how frontier AI systems affect economies, societies and security.6 min read
ResearchAnthropic launches Economic Futures Program to fund research on AI’s economic effectsAnthropic announced the Economic Futures Program on June 27, 2025, a new initiative to support empirical research, evidence-based policy proposals, and expanded economic measurement of AI’s impacts.3 min read
ResearchWhy AI Sometimes Writes in 'Yoda-Speak' — An Autoregression and Attention HypothesisAndrew Stellman describes how he uses AI's consistently poor prose as a drafting tool and investigates a recurring oddity he calls “Yoda-speak,” where generated sentences have backwards or awkward word order.5 min read
ResearchGoogle Research outlines ten advances expanding tools for scientists, clinicians and developersAt a Research@ event in Mountain View, Google Research presented ten recent projects and tools aimed at extending human ingenuity across science, health, disaster forecasting, education and enterprise.6 min read
ResearchFrom Heat Maps to Heat Action: AI Predicts Risk but Struggles to Trigger ResponseResearchers and cities increasingly use AI to map and forecast extreme-heat risk, but a new review shows the research concentrates overwhelmingly on prediction rather than on enabling operational responses.4 min read
