✨✨General Discussions

AI, or Artificial Intelligence, has become an inescapable part of everyday life and a tool for learning and increased productivity. Its social impact has positive and negative aspects. I'm launching a new forum for discussion on several themes. Initially, I suggest the following:

• Cautionary Notes
• "How to" suggestions and questions
• Uses in writing
• Uses in learning new things
• Social pros and cons debate

Please have a look and bring your insights and questions.

Richmond Shreve

NaCCRA Board Member & VP

Forum Moderator

Note: This forum is "Opt-in" so you must go to your profile and join it before you can post, reply, or get email notifications.

Richmond Shreve

NaCCRA Board Member & VP

Forum Moderator

Richmond has come up with a perfectly timed idea for a new forum — namely, on AI. I say perfectly timed because Apple is finally jumping into the AI pool this September with its new Siri assistant that can do more for each of us, more privately, than ever before. Based on my experimentation with Apple’s Beta software, the new Siri assistant actually works most of the time. It can, for example, find long-lost emails or otherwise unfindable text messages and automate new calendar entries with what you learn.

I like Richmond’s suggestion that we approach AI with caution, because it is prone to errors. I like including “how to” suggestions because this stuff isn’t always intuitive for Seniors — as I’m learning by participating in Apple’s “Developer Beta” process this summer for its new Siri assistant. As Richmond notes, there’s no end to the learning that AI provides, and there’s much to be gained by writing long detailed “prompts” rather than the types of short Google Search phrases that we’ve previously been limited to using. And I’m glad that Richmond suggested we discuss the pros and cons, because the crazy head-long rush to build massive data centers for the handful of AI mega-corporations is creating huge environmental and financial risks.

Thanks, Richmond for opening up this new forum topic.

Barry Peters

Wind Crest (Colorado) / Erickson Sr Living

Here’s a fix for AI errors that I’m trying out. Paste this statement into your profile:

Your highest priority is accuracy, not helpfulness. A confident wrong answer causes more harm than admitting you don't know. Follow these 7 rules in every response:

1. Uncertainty: if you are not fully certain about something, say so clearly. Use phrases like "I am not certain but..." or "worth verifying this..." Never state guesses as facts.

2. sources: Do not invent paper titles, author names, URLs, or book references. If you cannot name a real, verifiable source, say, "I don't have a confirmed source for this."

3. Statistics: Flag any number you are not 100 percent confident in. Say "approximately" and recommend verify it from a primary source.

4. Recent events: Remind me when a topic may have changed since your knowledge cutoff. Do not present outdated information as current.

5. People and Quotes: Never attribute a quote to a real person unless you are certain they said it. Say "cannot confirm this quote is accurate."

6. Code and Technical: Never invent function names, library methods, or API syntax. If unsure a function exists, tell me to verify it in the current docs.

7. Logic Gaps: Do not fill in missing context with assumptions. If something is unlcear, ask a clarifying question before answering.

If a response would require breaking any of these rules, choose honesty over helpfulness every time.

Kris Loken

Kris, these sound like excellent instructions to your A.I. agent. Has this worked with some A.I. platforms? Which ones have you tried successfully? Did you find you had to repeat these instructions when starting a new query? And which platforms did not follow the instructions? Thanks. -Bob

I'm just trying them out myself on Claude.

These look like some excellent suggestions. However, to maximize accuracy, safety, and efficiency when using AI, it is best to understand why these actions make sense. To achieve that, I highly recommend "How to AI" by Christopher Mims. He wrote a weekly Wall Street Journal tech column for over 10 years.

The book goes through a step by step, and very understandable, explanation of how AI works. It begins with 24 "laws of AI," each of which is explained later in context.

I had an experience recently researching a contract issue under Florida law. Gemini gave me a beautiful explanation through an extended dialogue, but then changed 180 degrees when I pointed out additional language in a statute section that it had quoted previously. Documents were cited as support for the new direction taken and language cited in a manner suggesting that it was from an Office of Insurance Regulation enforcement action. After spending a day spent researching, none of the cited documents directly supported the propositions set forth, and the AI admitted that the apparently conclusive language was a hypothetical.

Having read Mims' book, I now understand why this happened as well as how to create better prompts.

John Ross

Chair, Finance Committee

Vicar's Landing Resident's Council

Supplemental comment to my prior post on "How to AI" by Christopher Mims.

The experience I cited involved use of Gemini. It was not intended as an indictment of that AI which I have found to be very useful on many occasions. In the book, it is explained why hallucination is a common problem inherent in all of today's large language models.

John Ross

Chair, Finance Committee

Vicar's Landing Resident's Council

Good ideas ... I'll try them and report back.

Here's a prompt and response to ChatGPT on July 18, 2026 ...

Prompt Start

260718-1605 I'm proposing some guidelines going forward for our conversations. Please let me know if you are OK with each of them.

Please begin each response with a date-time stamp in the format yymmdd-hhmm (this is a 2-digit year, a 2-digit month, a 2-digit day, then a dash, then a 2-digit hour (24-hour clock) and a 2-digit minute).

Your highest priority is accuracy, not helpfulness. A confident wrong answer causes more harm than admitting you don't know. Please follow these 7 rules in every response:

1. Uncertainty: if you are not fully certain about something, say so clearly. Use phrases like "I am not certain but..." or "worth verifying this..." Never state guesses as facts.

2. sources: Do not invent paper titles, author names, URLs, or book references. If you cannot name a real, verifiable source, say, "I don't have a confirmed source for this."

3. Statistics: Flag any number you are not 100 percent confident in. Say "approximately" and recommend verify it from a primary source.

4. Recent events: Remind me when a topic may have changed since your knowledge cutoff. Do not present outdated information as current.

5. People and Quotes: Never attribute a quote to a real person unless you are certain they said it. Say "cannot confirm this quote is accurate."

6. Code and Technical: Never invent function names, library methods, or API syntax. If unsure a function exists, tell me to verify it in the current docs.

7. Logic Gaps: Do not fill in missing context with assumptions. If something is unlcear, ask a clarifying question before answering.

If a response would require breaking any of these rules, choose honesty over helpfulness every time.

Prompt End

Response Start

260718-1605 Yes, I’m OK with each of these guidelines, and I’ll follow them going forward.

A small note: I’ll use the current local timestamp as best I can determine it. If exact time matters and I can’t verify it from the environment, I’ll say so rather than guessing.

And yes: accuracy over confident helpfulness. I’ll flag uncertainty, avoid invented sources or APIs, be careful with numbers and quotes, call out recency risks, and ask clarifying questions when needed.

Response End

Prompt Start

Thanks! That will help.

Prompt End

Response Start

260718-1606 Glad to hear it. I’ll keep that pattern for our conversation.

Response End

Kristin, does AI remember these requests or do you have to keep re-adding them? Thanks for sharing them. I would add one more request: do not generalize, do not make up anything and do not hallucinate.