Scheduled tasks in ChatGPT ↗
ChatGPT’s scheduled tasks now have a dedicated sidebar page where users can view upcoming runs, then pause, resume, edit or delete them. Tasks can also run within broad windows, such as morning or evening, rather than requiring an exact time.
Monitoring tasks can check the web and connected apps, alerting users only when something meaningful changes. The update is rolling out across paid workplace and consumer plans - essentially, ChatGPT is becoming a tiny operations desk… or so it seems.
A near-autonomous AI chemist improves a challenging reaction in medicinal chemistry ↗
OpenAI connected GPT-5.4 to Molecule.one’s Maria chemistry system, allowing it to propose research ideas, design experiments and analyse results. Maria’s automated laboratory ran 10,080 reactions, though human chemists still selected proposals and supervised the physical work.
The system identified TEMPO as a surprisingly effective additive for a difficult Chan-Lam coupling. Yields improved across more than 80% of tested substrates, while human-run validation produced higher yields in 11 of 14 pairs - consequential science, not merely chatbot chemistry theatre.
New research shows how AMIE, our medical AI, could help manage health conditions ↗
Google’s AMIE has moved beyond one-off diagnosis experiments and into long-term disease management. It combines live patient conversations with deeper reasoning across clinical guidelines, medication information and lengthy medical histories.
In a blinded study involving patient actors, AMIE was compared with 21 primary-care doctors. It matched clinicians on overall management reasoning and scored higher on plan precision and guideline alignment - impressive, certainly, though Google is still studying clinical use rather than releasing it as a doctor-in-a-box.
Anthropic opens Seoul office and announces new partnerships across the Korean AI ecosystem ↗
Anthropic has opened a Seoul office and signed an agreement with Korea’s science ministry covering responsible public-sector AI, Korean-language safety evaluations and intelligence sharing around AI-enabled cyber threats.
Claude is also spreading through major Korean companies. NAVER has deployed Claude Code across its engineering organisation, while LG CNS, Samsung SDS, Nexon and Hanwha are expanding workplace and development use - Korea is becoming a rather crowded AI engine room.
US awards $500 million to Nvidia-backed SandboxAQ for finding new chipmaking materials ↗
SandboxAQ received a $500 million US government award to develop materials for domestic semiconductor manufacturing. Its physics-based AI will search for PFAS alternatives, better catalysts, rare-earth-free magnets and improved batteries for chipmaking equipment.
The Commerce Department will receive a minority, non-voting stake and royalties from successfully licensed formulas. It is a slightly unusual government-contract-meets-venture-investment sandwich, but the strategic aim is clear - fewer fragile foreign supply chains.
AI lab Odyssey valued at $1.45 billion in latest funding round ↗
Odyssey raised $310 million in a Series B round led by Natural Capital, reaching a $1.45 billion valuation. Amazon, AMD Ventures, Google Ventures, EQT and In-Q-Tel joined the round alongside several prominent individual investors.
The company is developing world models that simulate physical environments and support autonomous action, rather than another conversational assistant wearing a nicer hat. AWS becomes its preferred cloud provider, giving Odyssey access to Trainium chips and joint commercialisation support.
In US, EU mutual interest for Europe to use best AI models, von der Leyen says ↗
European Commission President Ursula von der Leyen argued that Europe retaining access to leading American AI models benefits both sides. She also backed stronger testing of powerful systems, comparing responsible AI deployment with established aviation-safety practices.
The comments followed restrictions that forced Anthropic to disable its most advanced models for foreign nationals. Model access is no longer merely a product decision - peculiarly, it now resembles export policy, cybersecurity policy and diplomacy folded into one suitcase.
FAQ
How do scheduled tasks in ChatGPT work?
Scheduled tasks can run at a specific time or within a broader window, such as the morning or evening. Users can manage upcoming runs from a dedicated sidebar page, where they can pause, resume, edit or delete tasks. Monitoring tasks may also check the web and connected apps, then alert users when a meaningful change occurs.
How did OpenAI’s AI chemist improve the Chan-Lam reaction?
OpenAI connected GPT-5.4 to Molecule.one’s Maria automated chemistry platform to propose ideas, design experiments and analyse results. The laboratory carried out 10,080 reactions under human supervision. The system identified TEMPO as an effective additive, improving yields across more than 80% of tested substrates, while human validation produced higher yields in 11 of 14 pairs.
Can Google AMIE manage long-term health conditions?
Google is researching AMIE as a system for supporting long-term disease management, rather than limiting it to one-off diagnosis. It combines patient conversations with reasoning drawn from clinical guidelines, medication information and medical histories. In a blinded study using patient actors, it matched primary-care doctors on overall management reasoning and performed strongly on plan precision and guideline alignment, but it is not being released as an autonomous doctor.
Why is Anthropic expanding its AI presence in South Korea?
Anthropic has opened a Seoul office and formed partnerships spanning public-sector AI, Korean-language safety testing and intelligence sharing on AI-enabled cyber threats. Claude is also being adopted by Korean companies including NAVER, LG CNS, Samsung SDS, Nexon and Hanwha. The expansion reflects growing demand for enterprise AI tools, coding systems and locally relevant safety evaluation.
Why is access to advanced AI models becoming a policy issue?
Access to advanced AI models increasingly intersects with export controls, cybersecurity and international diplomacy. European Commission President Ursula von der Leyen argued that Europe’s access to leading American models benefits both regions, while also supporting stronger safety testing. The discussion followed restrictions affecting Anthropic model access for foreign nationals, showing that AI availability is no longer merely a commercial product decision.