doorss

Anthropic researchers explain how to apply multi-day agentic coding workflows—including test oracles, persistent memory, and orchestration patterns—to scientific computing tasks. A rare and detailed look at pushing Claude's capabilities for long-running, complex research automation.

Epoch AI has confirmed that GPT5.4 Pro solved an open problem in Ramsey hypergraph theory from its FrontierMath benchmark. This marks a significant milestone in AI's ability to tackle unsolved mathematical problems.

A demonstration shows a 400-billion parameter LLM running on an iPhone 17 Pro, highlighting rapid advances in on-device AI inference. This builds on streaming mixture-of-experts techniques that allow massive models to run on memory-constrained hardware.

Simon Willison covers the 'streaming experts' technique for running large Mixture-of-Experts models on hardware without enough RAM by streaming expert weights on demand. This approach is enabling surprisingly large models to run on consumer devices.

A practical guide to getting the most out of Claude Code for software development, covering workflows and productivity tips. Useful for developers evaluating how AI coding agents fit into their daily work.

Mozilla AI launched Cq, a tool designed to be a Stack Overflow-like knowledge base specifically for AI coding agents. It addresses the growing need for structured, agent-accessible documentation as AI-assisted development becomes mainstream.

A thoughtful reflection on what's actually hard about software engineering—understanding systems, debugging, and designing architecture—arguing these challenges remain even as AI handles more code generation. Speaks directly to how AI will reshape but not replace the craft of engineering.

A sharp definition of 'slop' as content that takes more human effort to consume than it took to produce—like receiving raw Gemini output from a coworker. Relevant to ongoing discussions about LLM writing tropes and the social norms emerging around AI-generated text.

Masayoshi Son's SoftBank faces investor anxiety with a massive $30 billion investment in OpenAI, pushing the conglomerate's borrowing capacity to its limits. The deal underscores the enormous financial stakes in the AI race.

Siemens CEO Roland Busch warns that Europe could throttle innovation by prioritizing AI sovereignty over speed of adoption. The comments feed into the broader debate about Europe's competitiveness in the global AI landscape.

Film studios may find that AI-driven cost savings in production are quickly competed away, benefiting consumers and platforms rather than the studios themselves. An interesting analysis of how AI disrupts creative industries in unexpected ways.

Marja Hjeltin puhelin julkaisi yllättäen yli 4 000 kuvaa Facebookissa ilman lupaa, eikä kuvia saanut täysin poistettua. Tapaus herättää huolta yksityisyydestä ja laitteiden hallinnasta.

The Trivy security scanner has been hit by another supply chain attack through compromised GitHub Actions tags, potentially exposing secrets in CI/CD pipelines. A sobering reminder of the fragility of software supply chain security.

The FCC has added foreign-manufactured consumer routers to its covered equipment list, restricting their use due to national security concerns. This could significantly impact the consumer networking hardware market.

Tuoreen mielipidemittauksen mukaan suomalaiset luottavat Yhdysvaltoihin lähes yhtä vähän kuin Kiinaan tai Venäjään. Eniten luottamusta herättävät Pohjoismaat sekä puolustusvoimat ja presidentti.

Ylen artikkeli listaa ammatteja, joita tekoäly ei korvaa lähiaikoina. Koulutus ja käsityötaito ovat edelleen valttia työmarkkinoilla.

Suomen Pankin mukaan Iranin sodan aiheuttamat energian hintamuutokset kiihdyttävät inflaatiota puolella prosenttiyksiköllä. Sodan tuhot pitävät energian hintaa korkealla vielä pitkään.