TabLOID

OpenAI announced a National Community Compact on November 5 for future campuses it owns and operates. The company says it will pay for the electricity and infrastructure needed to serve those sites, with financial assurances before major construction spending to protect other customers if plans change.

The commitments also include early community engagement and annual reporting on power, water, noise and hiring. OpenAI says it will avoid nondisclosure agreements with local officials, with narrow exceptions for legitimate security and sensitive business information.

The scope matters. This is a company commitment for its own campuses, not a rule covering every facility supplying OpenAI or the entire industry. A separate policy proposal asks governments to establish broader standards. Whether neighbors receive the promised protection will depend on the agreements and their delivery. The bill has been acknowledged; now comes the accounting. OpenAI's compact, November 5.

Google introduced Gemini Omni 1.2 Flash on November 5, extending native video generation to 15 seconds and allowing durations from three to 15 seconds. The model is generally available through its API and Gemini for business.

The editing pitch is conversational: change a background, adjust the lighting or weather, or revise an aspect ratio without starting the whole project again. Google also describes better consistency and reference adherence. Those are the maker's claims, not a guarantee that every face, movement or object will survive an edit correctly.

For a creator, the attraction is a little more room to develop a visual beat and a shorter route to trying variations. The obligation is familiar: inspect the output. A convincing generated scene remains a constructed scene, however casually the software lets you order another take. Google's announcement, November 5.

Planet launched its AI app on November 5, offering searches across more than ten years of satellite imagery through a map-based chat interface. Users can describe a kind of place, examine changes over time and inspect the imagery behind a result.

The company says search works with an AI interpretation of visual features in its monthly mosaics. It also describes research agents, external-file uploads and scheduled reporting. Launch access is free, with premium capabilities anticipated.

The useful distinction is between finding a lead and establishing a fact. Planet's own examples emphasize the discovery stage: locating places worth a closer look before a fuller investigation. A search result can point an analyst toward a changed landscape. It cannot, by itself, establish who caused the change or whether a law was broken. Open the images before writing the accusation. Planet's launch report, November 5.

Apple's HALO research project combines a language-guided planner, a motion generator and a fast movement tracker. Demonstrations include sitting, crouching, passing under a desk, pushing a box and following a person.

The project describes onboard orchestration and embedded control. That does not mean the entire system was developed on an ordinary desktop. Its published setup lists 250 H100 GPUs for dataset generation, 64 B200 GPUs for motion-generator training and four H100 GPUs for tracker training.

That distinction separates the resources needed to learn a skill from the resources used to perform it. HALO is a research demonstration of turning instructions into movement, not a finished household helper or proof of dependable behavior in every home. Getting under the desk is impressive. Working safely around everything else in the room remains a bigger assignment. HALO project and training setup, checked November 8.