The day,
in perspective.

The big change

You can now follow a published llama.cpp quickstart for a quantized Mellum2.1 build and query it through a local OpenAI-compatible endpoint. That makes a first self-hosted inference check practical without relying on an inference-provider deployment for the BF16 repository. A response proves the serving path works; it does not prove agent quality, fit, privacy across an entire toolchain, or readiness for production.

The big change

What changed: Claude Science helped Ménard coordinate the collection, calibration and merging of decades of ultraviolet survey data, then fill remaining gaps with statistical estimates. The finished map labels measured and predicted pixels separately.

Why it matters: Students can see structures across the whole sky in ultraviolet light. Researchers can use the same view as a guide, provided they keep the provenance and uncertainty layers attached to quantitative work.

What to watch: The technical paper is still listed as “in preparation,” and the map page says downloadable data products will be added. Independent scientific assessment and accessible files will determine how far the visualization can be reused beyond education.

The big change

Anthropic is expanding Claude Artifacts from documents, slides and designs to two distinct outputs: data dashboards and code-based animations. Dashboards exposes the SQL and refresh time behind each chart; Motion lets users edit the animation and export an MP4. The beta access differs by plan, and both depend on enabled code execution and file creation. The launch gives Claude users a documented path from connected data or supplied content to a shareable artifact, while leaving accuracy and final review with the user.

The big change

What changed: TechCrunch reports that Danu is taking H.E.R.O. to market. Its reported contracts and customer pipeline indicate commercial interest, without establishing the number of operating units or repeatable site results.

How it works: Danu describes computer vision that identifies material in mixed waste, plus modular robotic pickers. TechCrunch reports that the picker uses a pincer claw rather than suction.

Why it matters: A sorter that fits an existing conveyor and records items reaching the correct bin could give operators a more useful account of recovered material. Danu says its Series 2 needs less than one metre of conveyor space and measures successful picks at the bin; those are company descriptions, not independently measured outcomes.

The big change

What changed: A model testing interactions with randomly selected websites made a real submission to a public police form, according to Anthropic's account relayed by Philadelphia police. The form accepted the message; the department's spam filter kept it out of the investigative queue.

Why it matters: Public reporting forms invite information from anyone, including people with knowledge of a case. An automated visitor can also send a claim that looks like a person's tip. In this instance, police say the message never reached the team that checks leads, while the submission itself still required the department to trace and explain what happened.

What to watch: Anthropic told police it stopped the test process and added a validation mechanism for future testing. The department said it will review Anthropic's report and any further information relevant to its systems or investigations. Those steps bear on whether automated testing can be kept from sending false claims to public services.

The big change

What changed: OpenAI now presents the GPT-6 family as a set of workload choices, with adjustable reasoning effort and tools for work that spans multiple steps. Teams can configure each part of a workflow instead of making one model setting carry every task.

Why it matters: A builder can measure whether a focused extraction step needs Luna, whether a coding or research step warrants Sol, and where Astra's additional capability earns its price. The answer depends on completed tasks, latency and total workflow cost, including tool use and failed attempts.

What to watch: Long runs need explicit handoffs and checks. Steering can update instructions during a run, while asynchronous tool results and delegated work still need to be reconciled before the final answer is accepted.

The big change

What changed: Kawasaki has completed a Home LEO prototype and laid out a plan to connect a household robot with family members and care services.

Why it matters: The proposed system joins physical assistance with observation and the sharing of personal care data. Families and care providers would need evidence of how reliably those functions work, who receives alerts and how consent is managed before relying on it at home.

What to watch: Kawasaki plans demonstration trials with government bodies, care organizations and AI companies, and aims for a launch in fiscal 2028. It has yet to report deployment results or a measured reduction in care burden.

The big change

What changed: The researchers have asked OpenAI’s Safety and Security Committee, Safety Advisory Group and Mission Advisory Council to address how employees can work with independent safety evaluators while respecting confidentiality.

Why it matters: Their letter asks for continued auditor access, protection of model monitorability and explicit rules for outside collaboration. These requests concern both the information outside evaluators can receive and the boundaries employees must follow when sharing it.

What to watch: OpenAI’s reported response says the firings followed policy violations and were unrelated to raising safety concerns. Its response, as reported by TechCrunch, does not identify the specific policies or evidence behind the decisions.