In partnership with
Good Morning Surrey,
Welcome to The Surrey AI.✨
Every Tuesday: global AI stories explained, with a Practical move and Surrey Viewpoint, plus news, tools, jobs and a poll.
No technical background needed. Understand what changed, why it matters and what remains uncertain.
For a while, using AI has felt rather like having a clever assistant who could give you plenty of advice but couldn't quite finish the paperwork.
You could ask it to draft something, then copy the result into another application, sort out the formatting and check whether it had quietly invented anything. Helpful, certainly. Seamless, not always.
That arrangement is starting to change.
AI companies are moving beyond offering answers and towards becoming part of how work gets done. But as these systems become more useful, some rather important questions are following them into the room.
Who decides what information AI can learn from? How much electricity will the computers behind it require? And when a company announces a sensible-sounding solution, how do we tell the difference between a promising plan and something that actually works?
Those questions connect this week's developments.
The interesting part isn't simply that AI is becoming more capable. It's that its everyday responsibilities, and the expectations placed upon it, are becoming considerably more serious.
The technology is growing up. The paperwork, appropriately enough, is getting thicker.
In today's edition
1.AI is becoming a practical document-editing workspace.
2.Website owners gain simpler way to their AI-training preferences.
3.Flexible power use proposed for AI's growing physical footprint.
PRODUCTIVITY
Google Gemini can now edit files as you work
Google's Gemini app can now create, preview and edit supported files inside a conversation, rather than simply producing text to copy elsewhere. Announced on 29 October, the change lets users review a document, highlight a section and ask Gemini to revise that part.
For example, someone preparing a covering letter can adjust an overenthusiastic paragraph without repeatedly switching applications. Students, volunteers and small businesses could use the same approach for notes, newsletters and paperwork.
What works where?
PDF previewing and editing work on web and mobile. Google Docs, Sheets, presentations, text files, Markdown and code formats are supported on the web only. Google says the feature is available wherever the Gemini app operates, but support differs by device and format. Editing inside Gemini is not necessarily the same as changing a live Google Workspace file.
Why this matters
The benefit is less time moving between tools. AI is shifting from suggesting changes to helping apply them. That also raises the stakes: a modified spreadsheet needs closer checking than a rewritten introduction. Complex layouts, tables, factual accuracy and exports still need review; a product launch does not establish perfect results across accounts. Check how sensitive information is handled before uploading it.
Practical move this week
Try the feature with a non-confidential document you know well. Ask Gemini to improve a single paragraph, compare it with the original and check facts, formatting and the export. If unavailable on your account, review Google's demonstration instead.
Surrey Viewpoint
For a Surrey charity drafting a funding application, a small business updating customer information or a student improving a personal statement, the test is whether the document gets finished accurately with less effort, without losing control.
Cities Are Running Out of Salt Before Winter Even Starts
Road salt is one of those things nobody thinks about until plow trucks come back empty, and this year that's happening before the first snowfall. Some cities haven't gotten a single bid for winter supply, while others are paying nearly double last year's price.
North America doesn't mine enough of its own, which is why it ships in 8 to 10 million tonnes from places like Egypt and Chile, two weeks away by boat.
Atlas Salt (TSXV: SALT | OTCQX: SALQF) is building a mine in Newfoundland near a deepwater port where ships can reach Boston in three days. With salt, the real cost is in shipping, and CEO Nolan Peterson says the difference is so big that overseas producers could give their salt away and still come out more expensive than Atlas. The project is shovel-ready and projected to generate roughly C$188 million a year in free cash flow.
DIGITAL RIGHTS
AI companies move towards shared website training controls
How the proposed preference works: a website operator publishes an AI-training instruction; automated systems read it; participating providers interpret and honour it under their commitments. It signals a preference, not a technical barrier or universal legal ban.
Who decides whether online material can train AI models? On 30 October, OpenAI said it intends to honour a common website signal for preferences about foundational generative AI training. Website owners currently may need different instructions for different providers; a shared approach could simplify that.
How would it work?
Websites commonly use a robots.txt file to guide automated visitors. The proposed common signal could tell participating AI providers, for example, that a publisher does not want certain material used for model training. The effort is linked to work at the Internet Engineering Task Force (IETF), which develops internet standards.
What hasn't changed yet?
OpenAI's commitment is voluntary, with implementation expected over coming months; existing crawler controls remain. This is not yet a final, universally implemented standard. The instruction cannot physically prevent copying, remove content from models already trained, or automatically create an enforceable legal right. It need not govern AI-powered search or answers in the same way as model training. Nor does a website operator necessarily own every item published there.
Why this matters
Publishers, artists, researchers and businesses create information that AI systems may use. A clearer preference could help, but its value depends on provider adoption, consistent compliance and precise scope. Licensing, creator payment and access to information remain open questions.
Practical move this week
If you run a website, review its existing crawler instructions and ask your platform which settings affect training versus search visibility. Don't assume that blocking training means disappearing from AI-generated answers. Other readers can ask who created valuable online material and how that work is supported.
Surrey Viewpoint
Independent Surrey publishers, photographers, consultants, local firms and charities would benefit from a simple choice only if providers actually honour it. Watch implementation, not expressions of support.
ENERGY & INFRASTRUCTURE
OpenAI proposes flexible power use for future data centres
Behind an AI response are data centres packed with power-hungry computers. As new facilities are planned, electricity networks must serve both ordinary users and substantial new computing demand.
On 2 November, OpenAI announced its Grid Protectors Program, intended to let future US computing campuses reduce or shift electricity demand when the grid is strained. This is called demand response: adjusting power use at busy times instead of always increasing supply.
The Georgia arrangement
OpenAI says its planned Effingham County campus has an agreement with Georgia Power covering up to 1,000 megawatts of flexible demand response. A megawatt measures the rate of power use. The figure is the maximum flexibility specified, not electricity already saved. OpenAI likens its scale to demand from around 700,000 households; that does not show those homes avoided outages or received cheaper bills. The campus is under development and described as a future pilot.
Could this help electricity networks?
Potentially. Reducing a large facility's demand during a peak could create breathing room. But how quickly can computing jobs be shifted, which tasks can wait, and will reductions happen reliably under real pressure? Who benefits financially? Using less electricity at a peak is not the same as lowering annual consumption. Claims about prices, grid reliability and environmental benefits still need operational evidence.
Why this matters
AI infrastructure needs electricity connections, cooling, construction and investment. Here, flexible demand is included in a specific infrastructure agreement rather than only suggested in principle. Measurable results have yet to be shown.
Practical move this week
Distinguish (1) a plan, (2) an agreement, and (3) measured performance. Watch for reports showing when electricity demand was actually reduced, by how much and with what effect on the grid.
Surrey Viewpoint
UK networks also face difficult decisions about connecting large electricity users while maintaining affordability and reliability. Georgia offers a question worth testing, not proof of savings or lower household bills: can computing demand reliably flex when the grid needs it?
In Other News
AI-assisted glasses begin taking orders in selected countries — Google's wired XREAL AURA extended-reality glasses opened for orders in the US, Canada and South Korea on 28 October and Japan on 2 November. They are not ordinary invisible-assistant spectacles; UK availability has not been announced.
Anthropic reports early progress in AI-assisted drug research — Anthropic and six research partners report preliminary laboratory results, including engineered immune-cell experiments. These are neither approved medicines nor evidence of successful human cancer treatment.
Meta works with Canadian sight-loss organisation on AI glasses — Meta and the Canadian National Institute for the Blind will consult Blind, Deafblind and low-vision people, demonstrate and donate glasses, and provide training. The partnership has not yet proven universal accessibility or better independence or employment outcomes.
China proposes new restrictions on AI technology exports — Draft restrictions reported by Reuters could affect overseas transfers of important AI models and technology, distinct from remotely using an AI service. They are proposals, not an enacted blanket ban.
AI Tools
Four tools for different jobs; not all are new, and access and subscriptions vary.
SynthID Detector — Checks supported images, audio and video for SynthID watermarks. English-language global availability was announced in October. Try: Test a file, but remember a negative result doesn't prove it is genuine.
NotebookLM — Ask questions about study notes or reports. Try: Request three key arguments with citations, then verify them. Check data policies before uploading confidential material.
Canva AI — Make event posters, presentations or small-business graphics. Try: Create a poster and proofread dates and wording; available features and limits vary.
Perplexity — Use source links to investigate a question. Try: Ask something partly familiar, then check whether the cited pages actually support its answer.
AI Jobs
Three London opportunities listed by OpenAI, all experienced or senior. Check live availability before applying.
Applied AI Architect, Education — OpenAI — London, hybrid (three office days). Help educational institutions adopt AI; substantial technical and customer experience required.
AI Deployment Manager — OpenAI — London, hybrid (three office days). Support AI adoption, training and deployment; at least six years' relevant experience expected.
Applied AI Delivery Lead — OpenAI — London, senior role. Lead complex customer AI implementations; not an entry-level technical position.
A quick poll before you go
How much of your daily to-do list would you trust AI to manage?
Which part of AI's development matters most to you right now?
Thanks for reading Surrey AI.
The common thread this week is that AI is becoming part of everyday systems, from the document on your screen to the websites you read and the electricity networks you rarely think about. The useful question is no longer simply what AI can do. It's what we can reasonably expect it to do, and how we'll know whether it has delivered. Until next Tuesday,
