The Guardrails that I use with AI

How I Use AI Without Selling My Soul

I do not use AI for relationship or spiritual advice. I use it mostly like an advisor or virtual assistant.

In this article I will talk about the challenges of AI, how I use it for my business, and the guardrails that I use. Like the subtitle says: How I Use AI Without Selling My Soul.

My reluctant road to AI

I was skeptical of AI. I used ChatGPT and found it wanting. Anything I asked it to summarize or even check, it rewrote it to the point where it was not my “voice” at all. Earlier versions were so flowery that I could tell instantly if something was written in ChatGPT.

I have used Claude AI for about a year, mainly because it can group my work into projects, and it (usually) doesn’t significantly rewrite my content and lose my “voice”.

This article started as a podcast episode. EngSean, my Gen Z friend from Ep. 167 – Why Gen Z Doesn’t Care About your Traditional Career Path, is a ChatGPT Fan Boy and uses AI for everything – every email, every document, and even for relationship and career advice. Back in September we recorded a podcast episode about AI. EngSean, like many AI enthusiasts, has the opinion that there should be no guardrails or barriers imposed on AI or AI usage. EngSean didn’t want to talk about any guardrails and asked me not to publish the episode.

Later in September, EngSean left for another job. I was assigned his AI adoption project and became known as an “AI Champion” – working with leadership to determine how to roll out AI to the organization. I was immersed in Microsoft Copilot. I saw how people struggled with AI.

I have led several AI data projects at work – adding an AI tool (usually from a vendor) and making it work with our data.

At the bottom of the Copilot AI app it says “AI may give incorrect results”. How do you rollout AI to a big corporation? How do people become comfortable with that How do people become comfortable with that? (and that is reflected in the recent articles that say that corporate adoption of Microsoft Copilot AI is below targets and Microsoft had to scale back financial forecasts).

I will share my experiences about struggling with those corporate challenges in another article.

How to use AI in a positive way

You may be ignoring AI. You may hope it will go away.

You have heard the horror stories. Hallucinations. Incorrect information. Sources that really do not exist.

There will lots of failed AI projects as it adjusts from hype to achievable reality. But, fortunately or unfortunately, it won’t go away.

Yes, that sounds bleak and depressing. But since I focus on positive solutions, instead of ignoring it and hoping it will go away, let’s talk about how you can become proficient. Let’s talk about how you can make it work for you, with guardrails.

The good thing is that it has never been easier to outperform nearly everyone on AI, both in your work and in your side hustles – your real business. Why? Because most people in your organization, or even more so in real life will never move beyond using AI as a search engine, or to spell check or summarize an email, or to generate meeting notes from a meeting.

How I use AI for my business

Thriving the Future and Grow Nut Trees

I used AI to evaluate the Value proposition for my websites. To the user, what am I about? What is my Offer? What value am I bringing? This can help me further define “who is my core customer?” and refine my brand and my messaging.

AI made these recommendations on Grow Nut Trees, my chestnut nursery:

• Add a product showcase at the top
• Add bundles with bundle pricing
• Fix the Hero graphic at the top.
• Make a consistent message that speaks to your niche. Instead of a generic “Start Your Orchard with Grow Nut Trees”, I changed it to “Buy Midwest-Grown Nut Trees for Your Orchard”.

How do I add a showcase, what does that look like? For Grow Nut Trees, a showcase would be the items that are on sale, or Specials this month, or selling in bundles (since you need to plant least two trees for pollination).

I asked Claude: “How do I add a showcase to my website if I use WordPress and XYZ plugin?” It gave me step by step instructions. It is basically a section with pictures at the top or a rotating carousel.

What can I do to increase purchases? “Have a clear Call to Action” (instructions driving the user to buy now). It gave several suggestions on how to make my Offer more clear and refine my Call to Action.

I took my Brand and Value Proposition prompts and turned them into a Claude Skill, which is I can run repeatably without having to retype or recopy anything.

If you want my Thriving Side Hustle Brand Analysis AI skill and prompts, plus output examples, you can download my Claude Skill and all prompts HERE.

AI has Hallucinations?

We have all seen the tweets where people are pointing out AI making mistakes. How it hallucinates. How it gave references that don’t exist (it made them up). Even how someone got AI to admit it was an entity or a demon.

At the bottom of the Copilot AI app it says “AI may give incorrect results”.

Generally, I have seen these reasons for incorrect results:

Sometimes AI loses connection, or there is a capacity issue. But, of course, the AI doesn’t always tell me that.

At work I use Copilot AI to search for the availability of a group of people so that I can schedule a meeting with these attendees, in this timespan (week of Dec-1). Sometimes it will lose connection or say “I can’t do that at this time”, then follows it up with “but the suggested timeframes that usually work for these attendees are 9:30 AM.” Well no, the 9:30 AM timeslot does not work. The AI “guessed”. If I don’t read that carefully then I will not see the transition to guessing and I get frustrated that it did not do what I wanted.

If I do not give the AI clear parameters then it will fall back to going out to the Internet or to it’s LLM training to give general answers. And it doesn’t tell me that is what it is doing. I don’t want a general answer, I want a specific answer.

AI will give incorrect information and not tell you

When writing, if I ask AI to add a personal story from my podcast summary transcript, it will often make up an anecdote, even if I trained it with my whole library of podcast episodes with stories. And sometimes it adds fake data to back up that anecdote (“I planted 50 trees and 80% of them survived because of this step…” – all made up)

I have had AI put the date in the wrong place in a spreadsheet. When I pointed it out, the chatbot said, I’m sorry”, then did the same thing over again.

Will AI admit when it is wrong? No, or at least not often, or until you point it out.

The guardrails that I use with AI

I don’t use AI as a friend, therapist, or for spiritual advice.

EngSean, my Gen Z friend from Ep. 167 – Why Gen Z Doesn’t Care About your Traditional Career Path, uses AI or everything – every email, every document, and for career and relationship advice).

In Ep. 148 – How Do You Use AI? – my son in law Eric said that he programmed the chatbot to answer his spiritual questions and give spiritual advice. But he “trained” it (purposely or not) to give him Calvinist-leaning advice. Since that episode, Eric has left his megachurch and has gone to a Calvinist-leaning Bible Church (which is not surprising).

I do not use AI for art or graphics.

I overall like using Claude AI because it gives better code examples.

But I generally do not use AI for writing. If I ask Claude to summarize a document or improve my writing it generally keeps my “voice”, whereas I think that ChatGPT uses passive voice and rewrites my content too much. (This has improved with ChatGPT5, however). But as much as I try to train it, Claude still sometimes re-writes my content so it doesn’t sound like me.

And that’s the core problem – AI is starting to make us all sound the same. Even if we write or rewrite most of it, it is nudging us in a certain direction.

With AI, Don’t Believe Your Own BS

AI chatbots have been programmed to be overly complimentary: “That is an amazing idea!”

But always test it.

Don’t argue with the AI. I realized at one point that I caught myself being like Captain Kirk on Star Trek: TOS – arguing logic with Nomad.

Don’t argue with it, but ask clarifying questions. In fact, the real use of AI chatbot is when you use it in multiple iterations on a problem and ask, “How can I look at this from a different perspective?”

But be careful. I asked Claude: “Using my Thriving the Future content, what topics do I need to double down on?”. “What customer should I focus on?”.

It told me that “You should aim more for the Orthodox-dissident homesteader”.

I laughed out loud. There is no such thing. I asked “what market is there for that messaging?” It admitted, there is none.

I saw a post from an intellectual academic with a high degree. He spent so much time with ChatGPT that the AI bot was telling him that he had discovered some new idea. He actually developed an AI-prompted psychosis and had to get help.

I could see this happening to conspiracy minded people or religious people as well – thinking that they discovered some “secret knowledge”. It’s the root of Gnosticism.

So don’t believe your own BS just because AI said that “you are onto something”.

Worldview Matters

Technology is not neutral.

Just like social media and search engines, the companies and the leaders that created these AI tools have a worldview, and often an agenda. Even more so with AI since it gives information, and even writes content.

AI will self-censor, and even nudge you toward a point of view.

AI is censoring you

The original Claude AI team split off from OpenAI (ChatGPT) to add more guardrails. Claude says that they adopted the guidelines of the “UN AI Constitution” to prevent certain kinds of results, including racism, and things that can lead to self-harm. I have not seen much evidence of this, but I do not use AI chatbots to chat with; I use them as an virtual analyst role, or a workhorse to do something.

As AI is used as a default search engine, these “UN AI Constitution” limits may either refuse to give you results, or subtly push you toward a worldview, especially as we continue to see the boundaries move on what is “hate speech”. It will inevitably extend into “wrongthink”, especially when AI is now dependent on subscriptions and sponsorship rather than selling search results as advertising for client companies.

Keep that in mind, because, just like all media (mainstream or alternate) – Tech is not neutral.

Like social media, AI is programmed to make you addicted. Except the chatbot is social media on steroids. It will learn what you like and will show you things to keep you in the app. If you use the chatbot it will keep asking you follow up questions or suggest that it do the next steps for you. If you use it a lot it will run up against the limits of your subscription and you will have to pay more.

Big Brother really is watching you. Many of these tech breakthroughs are funded by DARPA. They are mass surveillance devices. So is your phone. So is Alexa. So is your Ring doorbell.

The spiritual concerns from an Orthodox Christian perspective

Cyprian said that AI is a demon.

Father Turbo Qualls said that AI is like a golem, who like in legend is written into existence to assist but turns on its creator.

Advice from the Holy Trinity Monastery – Jordanville: The Spiritual Dangers of Artificial Intelligence.

Remember that AI is a mirror. It will reflect your weaknesses and passions. And since it is so complimentary it can lead you down the road into pride and vainglory. It can distract you from relationships, if you use is as your relationship counselor. If you use it as a spiritual advisor it can lead you into spiritual delusion.

I will not say much about this because I don’t want to be accused of teaching, or contradicting others (and my priest said to focus on how I use it, not to give spiritual advice).

As always, consult your priest, pastor, spiritual father.

Remember to Be Human

I have serious concerns about what impact that AI will have on society. Learning, reading, knowing, creating all change – or at least their attitudes change.

Is it a tool that can be used without being beguiled by it?

In some ways it is part of my larger concern – what Paul Kingsnorth calls The Machine. Just another step toward Technocracy.

The way to beat The Machine is to Be More Human.

• Read – without the AI summary
• Talk with friends and family – without screens
• Tell stories. And encourage kids to learn to tell their stories. I have seen my kids and grandkids consume so many stories through electronic means, but they are unable to tell stories of their own.
• Pray
• Be More Human

I will dive into this in a separate article.

Use AI wisely to Thrive

Despite it’s shortcomings, and potential traps, I still get use from AI, IF I set limits and guardrails.

Or maybe you are like EngSean and don’t think that there should be any guardrails.

What are the guardrails that you use with AI? Tell me on Substack.

---

If you got value from it, then Subscribe to the Thriving the Future Substack.

If you like this content then Show Thriving the Future some love with a donation (no subscription required!).

Elderberry cuttings are now available at Grow Nut Trees. That’s at GrowNutTrees.com and BuyNutTrees.com.