Welcome to Issue 5. Last time we told you never to give an AI agent your banking password. This time there’s a new tool that lets some agents sign in for you without ever seeing it. Also this week: two new models from OpenAI and Google, and a plain-English guide to the scoreboards that rank them.

Plus two Deep Dives on using AI with old photographs, two podcasts worth a listen, and a reminder about government imposter calls.

AI SIG announcements

First SIG meeting: Monday, November 9, 4:00 to 6:00 pm

The AI SIG has 661 members as of October 31. Our first meeting is Monday, November 9, from 4:00 to 6:00 pm. After that, we’ll meet on the first Wednesday of each month, 4:00 to 6:00 pm, starting Wednesday, December 2. All meetings are virtual on Google Meet until further notice. You don’t need to install any software or app: select the link in a web browser such as Chrome, Safari, Edge or Firefox, and you’re in. You don’t need a Google account to join, though you may be asked to type your name and wait to be admitted.

We hope online is temporary: we’re still working through the room-scheduling maze and hope to offer in-person options soon.

Every AI SIG member receives this newsletter automatically. Computer Club members who haven’t joined yet can email SCHHCC.AI.SIG@gmail.com with the subject “Join.” (Computer Club members only.)

ChatGPT classes filled; more in December

Both November sessions of “ChatGPT: The Parts You’re Not Using (Yet)” filled within two hours of the announcement in Education Weekly, which suggests a lot of interest. I plan to offer more in December, so stay tuned and watch the SCHHCC Education Weekly announcements for sign-up details.

Wednesday Walk-In Help: AI questions welcome

I’ll be at the club’s Wednesday Walk-In Help sessions, 2:00 to 4:00 pm, for the next couple of weeks. It’s a chance to ask general AI questions or get help with specific tools such as ChatGPT, Claude, Gemini, Gemini Notebook and others. Depending on interest, I’ll continue and possibly expand this to other days.

Computer Club Hour: Fix It, Design It, Imagine It, Friday, November 13

Sun City Hilton Head Computer Club logo: a gold sunburst with a clock face inside a blue circle.

Computer Club Hour is a Sun City Hilton Head Computer Club event.

I’ll be presenting “Fix It, Design It, Imagine It: Using AI Image Tools” at the Computer Club Hour on Friday, November 13, at 10:00 am in Magnolia Hall. It will include demonstrations of image creation, photo restoration and enhancement, photo analysis, creating websites with photo galleries, and much more.

Two related articles are now in our Deep Dives, if you’d like some background first: Preserving the Past While Revitalizing It: AI and Photo Restoration and Image Compositing with AI: What It Can Do (and What It Can’t). You may also like Restoring Family Photos with AI: How Much Is Real? from Issue 3.

Have a question or an idea for a future AI SIG session? Send it to SCHHCC.AI.SIG@gmail.com.

AI news worth knowing

1Password lets AI agents sign in without seeing your password

An AI agent is a program that carries out tasks for you, such as shopping or booking, instead of just answering questions. To do that on a website, an agent often has to log in. On October 22, the password manager 1Password announced a way for that to happen without the agent ever holding your password.

Here is how it works, according to 1Password. When an agent reaches a login page, it asks for your credential. You review the request and approve it in 1Password, and 1Password fills in the login on the site. The agent never receives the password. You can see what will be shared and disconnect the integration at any time.

It’s available at no extra cost on Individual and Family accounts at 1Password.com (not the .ca or .eu versions). Agents that work with it today are Meta’s Muse, Grok Bot and Townie. Claude Duet and ChatGPT Work are listed as coming soon. Logins that rely on passwords saved in your browser or in notes don’t work with agents.

Why it matters: Our Issue 4 advice stands: don’t hand an agent your banking password. This is the safer pattern for the logins you do approve, but it only protects your password. The agent can still act inside the account once it’s in, so start with low-stakes tasks and read what you’re approving.

Sources: 1Password for Agents, 1Password, Oct. 22, 2026

Also this week

  • OpenAI’s GPT-6.1 Terra (rolling out since Oct. 26) is a faster, lower-cost model for everyday tasks, including operating software on a computer. It is in ChatGPT’s Work and in Codex, a coding tool, not in ordinary ChatGPT chats, and the free plan isn’t on the list. Coverage of OpenAI’s release notes says paid plans from Plus through Business, Enterprise and Edu are included, and access is rolling out gradually, so check your plan. Company-run pages weren’t reachable, so details come from news coverage. (New to model names? See our guide to AI model names.) (Online Tech Tips, Let’s Data Science)
  • Google’s Gemini 4 Carbon (Oct. 28) is aimed at cybersecurity and long coding projects. Google is giving it first to a small group of security defenders while it tests safeguards, so it isn’t in the Gemini app, and nothing says when it will be. Performance claims are Google’s own. (Khaleej Times)
  • ChatGPT can now connect to iCloud Mail and iCloud Calendar (Oct. 27), on every plan including free. Once connected, ChatGPT can search and summarize your iCloud email, draft or send messages, and add or change calendar events. Setup is a few steps: turn on two-factor authentication for your Apple Account, create an app-specific password at account.apple.com, and enter it under Settings, then Plugins, in ChatGPT. It reaches only @icloud.com, @me.com and @mac.com mail, not Gmail or Outlook accounts you’ve added to Apple’s Mail app. Because it can read and send your email, start with a task like summarizing, and if you stop using it, also delete that app-specific password in your Apple account. Details come from news coverage. (Digital Trends)
  • Gemini can now create and edit documents inside a chat (Oct. 29). Ask for a document, PDF, presentation or spreadsheet, preview it, then highlight a part and ask for a change, such as turning a table into a graph. Google’s own example turns photographed handwritten notes into a PDF study guide. Google says it’s available to all users where Gemini is offered, but coverage disagrees on which file types and devices can be edited, so try it on yours. (Google’s announcement, Digital Trends)
  • ChatGPT’s Chat and Work are expected to merge. In Issue 4’s recipe-card article, Ron guessed that the choice between the two would not last. OpenAI has now said it plans to combine them, so you would describe what you want and ChatGPT would decide how to do it. No date has been announced; one report says an OpenAI executive has talked of the end of the year. Usage limits for the combined version are unclear, and some users worry that quick chats will count against the same allowance as longer jobs. This comes from news coverage of OpenAI’s remarks, not an OpenAI announcement page. (Pasquale Pillitteri)

Deep Dive: Two articles on AI and old photographs

Two new Deep Dives look at what AI can and can’t do with old family photographs. Both use pictures from my own family archive, and the ideas in each will be covered, with short demonstrations, at my Computer Club Hour presentation on Friday, November 13, at 10:00 am in Magnolia Hall.

  • Preserving the Past While Revitalizing It: AI and Photo Restoration. Where does restoration end and alteration begin? This article offers a few principles for using AI on old photos, such as keeping the original untouched and being open about what was done. It then works through one print of my father, scanned two ways, with three versions that show how much the capture, the tool and the instructions matter. It prints the exact instructions I gave the AI.
  • Image Compositing with AI: What It Can Do (and What It Can’t). I gave two AI tools ten faded headshots of my grandfather from the early 1930s and asked for a single portrait, then asked them to age it to about 60 and compared the results with real photos from that age. The results look convincing, but they are an informed guess, not a recovered image.

Both come back to the same point: a sharper, cleaner picture is not necessarily a truer one.

Member Voices

Your turn. Have you used AI for something practical, like checking a login with a password manager, planning a trip, or organizing a club event? We’d love to hear what worked, and what didn’t. Email SCHHCC.AI.SIG@gmail.com with a subject starting “Share:” and tell us how you’d like to be credited: full name, first name and last initial, or anonymous. We may share it in a future issue.

Tips & tools

How to read an AI model ranking

Every few weeks a company announces that its new model is the best. How would you check? Two public scoreboards are worth knowing.

Arena (formerly LMArena) began as a University of California, Berkeley research project. You type a question and get two anonymous answers from two different models, then vote for the one you prefer, or call it a tie. The model names appear only after you vote. Arena turns those votes into a ranking.

Artificial Analysis doesn’t use votes. It runs models through a set of tests and combines the results into an “Intelligence Index,” then lists each model’s price and speed beside its score.

What a ranking can tell you: which models are in the top group right now, and how a new release stacks up against the old favorites. Both sites change often, so check the date.

What it can’t tell you:

  • Preference isn’t accuracy. Arena’s votes show which answer people liked, and a confident, polished answer can win even when it’s wrong.
  • Small gaps mean little. Models a point or two apart are close to tied.
  • The best overall model may not be best for you. A ranking says little about the tasks you do, or whether the model is on a free plan or in the app you use.
  • The makers are involved. Arena says it works with model makers to test models before release. That isn’t wrong, but it’s worth knowing.

Try it yourself: Take a question you actually care about, such as explaining a medical term or planning a trip, and ask two assistants. Judge which answer you’d trust, then check one fact. That’s the same test Arena runs, with you as the judge.

Sources: Arena leaderboard, Arena · Hello from LMArena, Arena · Artificial Analysis model leaderboard, Artificial Analysis

Worth a look

  • Machine Gods (NPR podcast). NPR describes it as “the official podcast of the singularity,” with news, analysis and interviews with the people building the AI future, and new episodes on Wednesdays and Fridays. Some of it may lean toward industry insiders, so sample an episode before committing.
  • AI for Humans (YouTube show and podcast, hosted by Kevin Pereira and Gavin Purcell). A weekly show, posted on Fridays, that covers the week’s AI news, new models, tools and demos. Its creators describe it as “no hype, no PhD required.” It’s the same channel as the AI-made documentary we recommended in Issue 4.
  • Hack Apocalypse Now (Paul Krugman, Substack, Oct. 28, 2026). An opinion column arguing that AI tools are already finding software weaknesses, including some that took government websites offline, and that a more closed, “gated” internet may be needed. It’s the author’s view, and it’s a Substack page with subscribe prompts, though the full text was readable without paying.

Stay safe: Government imposter calls

A good reminder.

The phone rings, or a text arrives. Someone says they’re from a government agency, claims you face criminal charges or have a prize waiting, and tells you to act now. However convincing the voice sounds, the script is the giveaway.

The FTC’s rule of thumb: a real agency will never demand money, make threats, tell you to transfer money, or promise you a prize.

  • Slow down. Pressure is the tell. Hang up.
  • Check it out. Look up the agency’s real number yourself and call that, not the one the caller gave you.
  • Stop before paying anyone. Especially by gift card, wire transfer or cryptocurrency.

The FTC also notes that search engines and chatbots now often surface its warnings, so asking one “Will the FTC text me?” can help. Chatbots can be wrong, though, so confirm on the agency’s own site. Report imposters at ReportFraud.ftc.gov.

Sources: Getting the word out about FTC imposter scams, Federal Trade Commission, Mar. 26, 2025

That’s it for Issue 5. Tell me what helped, what confused you, and what you’d like covered next at SCHHCC.AI.SIG@gmail.com.

Until next time,
Ron

Prepared with AI assistance.