A clearer view ofModels, research and the ideas changing artificial intelligence.
Keeping a character, room or object consistent across scenes is a different challenge from producing one convincing clip. Google’s research explores ways to manage that continuity.
Read the analysisAn experimental add-on speeds up the text a vision-language model writes after processing an image. The reported gain depends on the task, hardware and which part of the response is being timed.
Read the analysisImages can help cities understand traffic and public space. Turning those images into policy requires knowing what was observed, what was estimated and whose experience is missing.
Read the analysis
Google’s new accessibility feature describes camera views and helps reframe them. Its gradual Android rollout comes with explicit limits on safety-critical use.
Read the explainer
How standard API rates, cache reuse and fast mode affect an Opus 5.5 bill, with reproducible examples and limits to Anthropic’s savings claims.
Read the explainer
Google says its planned server-side memory will let AI retain context across devices, with access protected by encryption keys held on users’ personal devices.
Read the news
Adobe’s listing describes optional web context, but the full launch post is unavailable. Here is what the remaining evidence supports.
Read the news
YouTube says creators can now get draft guidance and test video cuts in Studio, while it plans to expand its Shopping affiliate program to 35 countries by year-end.
Read the news
The agencies recommend restricting permissions, examining data access and remote connections, and maintaining recovery capabilities when using industrial control system integrators.
Read the explainer
Adobe says its plugin now connects conversational requests with interactive image adjustments and direct PDF editing, with distribution beginning globally.
Read the newsMIT News describes a fellowship for the 2026-27 academic year focused on exploring how artificial intelligence could support scientific discovery.
Read the newsGoogle Research describes interactive practice built around teacher-approved objectives, reports positive early feedback and says classroom studies will evaluate learning gains.
Read the news
Google’s research separates preparation from query-time retrieval. Its speed comparison measures one stage, not the time to answer every search.
Read the news
The authors describe a verification layer that links agent claims to individual tool sources, while reporting remaining difficulties with similar sources and imperfect blocking decisions.
Read the news
Meta announces an enterprise initiative spanning agents, APIs and coding tools, with Chirantan Desai appointed to lead the effort.
Read the news
Build an instruction you can inspect, reuse and check against a small set of facts. Start with a fictional packing list, then keep the same structure for another task.
Read the explainer
Google says reusable instructions are arriving in Gemini chat, with existing Gems set to migrate automatically when their support ends.
Read the news
PlayStation says patches accompanying its announcement add AI upscaling to both games, with claimed improvements to image detail and stability.
Read the news
The rollout puts cloud-based agents inside ChatGPT, with access, memory and approval controls that users will need to understand.
Read the newsParticipants identified synthetic clips less reliably in DF26 than in two comparison datasets. The study shows why visual inspection alone is a limited way to verify a video.
Read the analysisBefore choosing an AI tool by its ranking, check what the test measures, how it was run and whether its tasks resemble your own.
Read the explainer