AI (artificial intelligence): Introduction and best-use practices

By: Tom Cloyd & Claude AI (validated by Tom Cloyd) - 13 min. read (Published: 2025-09-02; reviewed: 2025-10-04:1909 Pacific Time (USA))

How I use AI

AI is here, now, whether you see it or not. AI’s already evident power, likely to increase rapidly, is seriously changing our relationships to our environment and each other. Affordable personal computers - and smartphones - have been history-changing technologies, but most authorities expect AI to be much more impactful. None of us can afford to be ignorant of AI, what it is, what it is doing, and what it is likely to do in the future.

“Knowledge is power; when you know more you can do more.”

Using AI is a continual learning experience. I have been using AI (artificial intelligence) tools for quite some time. I have also been educating myself on the general topic of AI so that I can use it well. Due to the very rapid development of this emergent new tool, this is a constant and challenging task. I don’t expect this to change.

AI isn’t perfect, but it can be remarkably useful and save time far beyond what you might expect. When we ask AI to assist us it can give an incorrect or even a nonsense answer. But, then, so can a person! This problem is well known, and over time AI’s error rate has dropped significantly. Worked with intelligently and collaboratively, AI can dramatically assist us in organizing and summarizing information and solving problems. I have worked mainly with three of the major models (I discuss this more below), and I’ve consistently gotten excellent results when working with them as collaborators. The Anthropic “Claude” model, to which I have a paid subscription so that I can use its advanced features, has proven particularly useful, especially with any computer-related problem

I give AI clear instructions. I often use AI models to research a topic for me, as a supplement to regular Internet search. When I do this I ask that the model give me a detailed answer to a question, accompanied by cited sources that can be checked. This approach has proven consistently useful.

I consider sharing the results - but only after manually validating them. If I think the resulting research report is likely to be helpful to others, I may publish it on one of my two websites, but always with proper author attribution, and only after I have assured myself that the article can be trusted. That means that I have checked the main assertions of the article against the cited sources and confirmed that the assertions correctly represent the sources. This is something that any thinking person should do with any article, regardless of how it was written or by whom. Something isn’t to be trusted just because we understand what it’s saying! (…Or, because the statement is coming from an expert, or because we like the person from whom it comes.)

Sadly, source-checking is not common behavior, and often not even possible. Most written material we encounter does NOT have cited sources. For example, consider newspapers, or TV news broadcasts, or virtually all magazine articles or website pages. As presented, what is being asserted cannot be checked. Most people don’t concern themselves about this, and this is one reason we have such a problem in our times with misinformation. We can do better than this. With urgent matters, we must.

Page contents…

• How I use AI
• A very brief introduction to essential language ^
• AI models I currently prefer and recommend ^
• Understanding AI Language Models: A Beginner’s Guide ^
• What AI Language Models Really Are ^
• About real-time Internet access ^
• What They Can Do ^
• Important Cautions to Keep in Mind ^
• Getting Started: Best Practices ^
• 10-Prompt Starter Pack ^
• Appendix - image prompt ^
• Notes ^

A very brief introduction to essential language ^{^}

All domains of endeavor develop specialized language - words and phrases with rich meaning whose use makes efficient communication possible - but only for those knowing the language. Glance over these and return here if you need to, while reading.

Here are key terms you need to know¹:

AI - Artificial Intelligence. “Artificial”, because it is not organic. “Intelligence”, because it receives information, and from it produces new information. Refers to “computer systems that can perform tasks typically requiring human intelligence.”²

AI agent - This is a rather new term that you will be hearing about a great deal in the years ahead. An “AI agent” is an AI model that has added to it the kind of superpowers normally only sophisticated computer programs possess. As a result, it can retrieve information, make decisions based this information, using directions it’s been given, and produce sophisticated output. One of the fancy thing it might do is use multiple specialized AI models to work through a problem to solve a problem it’s been given. At present, the most likely role for many AI agents is to replace customer service personnel in responding in real time to customer problems. “Vibe” coding is another Ai agent function people are hearing about.

ChatBot - A computer program added to an LLM that is capable of conversing with a human. It is itself a form of AI.

Hallucination - AI output that is presented as fact but is really made-up by the AI.

LLM (“Large language model”) - “AI trained on vast amounts of text to understand and generate human-like language.²

Model - A particular computer program that embodies the AI. It imitates, to a certain degree, the network of neurons of which organic intelligence is made, but is modified and managed by other programs to make it more useful.

Prompt - A text request given to an AI model, to which it responds to produce output for the requester. Good results require good prompts.

AI models I currently prefer and recommend ^{^}

All of these models produce rich output of great potential value, especially when pushed into their “deep research” modes. For important questions, I use all three, request explicit sources, save the output to a file, and study the results carefully - both for the ideas and for the sources given.

This is significant work, but clearly faster than old-style prowling-the-stacks library methods, although this prowling is not completely avoided in many cases.

In summary, AI proves to be a fast and hard-working research assistant whose work in every case must be gleaned and carefully checked. (But that would be the case with any research assistant.)

I access the following models most often, in this order:

• Claude - I have consistently found reports from this LLM to be richly valuable. Always ask in advance for sources, if you use the “deep research” option.
• ChatGPT - One of the nice features of this LLM is that it often responds to an initial prompt with 3 questions whose clear function is to increase the specificity of the user’s request. This not only reduces the work it has to do but also increases the likelihood of user satisfaction with the outcome report.
• Google Gemini - This LLM’s deep research reports are invariably thorough and well-sourced.

Understanding AI Language Models: A Beginner’s Guide ^{^}

In the next section is a report produced from a report generated by Claude, which I compared with those produced from each of the other two models listed above. This was the prompt I used with each model to get a response:

Please write an introduction to AI LLMs such as Anthropic’s Claude, OpenAI’s ChatGPT, Google’s Gemini, and Perplexity which would help someone who only knows about search tools like Google or DuckDuckGo. Address a) what AI LLMs really are, b) what they can do, and c) cautions to be observed in thinking about them and using them, and d) how best to get started using and learning about them. This narrative should use analogies appropriate to a high school graduate who occasionally uses a computer, but regularly uses their phone.

At the end, give me a compact, copy-paste-ready 10-prompt pack I can use on my phone or computer with ChatGPT, Claude, Gemini, Perplexity, or similar. Each prompt should be followed by a one-line tweak I can add (tone / length / details) and what to expect back.

I studied each of the reports and used the parts that, based on my own experience, seem most useful. The model from which a part comes is clearly noted.

(Immediately below is Claude’s report, produced in response to this prompt.)

What AI Language Models Really Are ^{^}

Think of AI language models like Claude, ChatGPT, Gemini, and Perplexity as incredibly sophisticated predictive text systems—similar to the autocomplete feature on your phone’s keyboard, but exponentially more advanced. While your phone might suggest the next word in a text message, these AI systems can predict and generate entire conversations, essays, code, and creative content.

Unlike Google or DuckDuckGo, which search through existing web pages to find information, AI language models are more like having a conversation with an extremely well-read person who has absorbed millions of books, articles, and documents. They don’t search the internet in real-time (though some can access current information); instead, they draw from patterns learned during their training to generate responses that sound natural and contextually appropriate.

About real-time Internet access ^{^}

(This section is an update, based on research done 2025-09-07 with Claude AI. ~TC)

In the past year, all the major AI models have acquired real-time Internet access capability. In some cases it is automatic, while in other you must specify it or ask for a specific version of the model. Note that all major AI models had free versions and paid versions which have extra features.

Here’s a brief summary table (do consult the notes, for details):

AI Model	Available on free tier?
Anthropic: Claude	yes3
Open AI: Chat GPT	yes, but limited4
Google: Gemini	yes5
Perplexity AI	yes6

If you are working with an AI model and want to know what it can do - like accessing the Internet in real-time - ask it!

What They Can Do ^{^}

AI language models excel at several key tasks:

Communication and Writing: They can help draft emails, essays, stories, or professional documents. Think of them as a writing assistant who never gets tired and can adapt to any style you need.

Explanation and Teaching: Like a patient tutor, they can break down complex topics into understandable explanations, provide examples, and answer follow-up questions.

Problem-Solving: They can help analyze situations, brainstorm solutions, and walk through step-by-step processes—similar to talking through a problem with a knowledgeable friend.

Creative Tasks: From generating story ideas to helping with brainstorming, they can serve as creative partners for various projects.

Code and Technical Help: They can write, debug, and explain computer code, making them valuable for technical tasks.

Important Cautions to Keep in Mind ^{^}

They Don’t “Know” Things Like Humans Do: AI models generate responses based on statistical patterns, not true understanding. They can confidently present incorrect information because they’re predicting what sounds right, not what is right.(See important note⁷ concerning this! ~TC)

Information Currency: Most AI models have knowledge cutoffs (Claude’s is January 2025), meaning they may not know about very recent events unless they can search the web. (For more on this, see here. ~TC)

Hallucination Risk: AI can generate convincing but entirely fabricated information—names, dates, studies, or quotes that seem real but don’t exist. Always verify important information, especially for professional or academic use.

Bias and Limitations: These models reflect patterns in their training data, which can include societal biases and misconceptions.

Privacy Considerations: Be cautious about sharing sensitive personal, financial, or proprietary information, as conversations may be stored or reviewed.

Getting Started: Best Practices ^{^}

Start Simple: Begin with straightforward requests like “Explain photosynthesis in simple terms” or “Help me write a polite email declining a meeting invitation.”

Be Specific: Instead of “Help me write,” try “Help me write a formal complaint letter to my landlord about a broken heater, maintaining a professional but firm tone.”

Iterate and Refine: If the first response isn’t quite right, provide feedback: “Make this more casual” or “Add more technical details.”

Fact-Check Important Information: Treat AI responses as starting points, especially for factual claims, statistics, or professional advice.

Experiment with Different Models: Each AI has slightly different strengths—try the same question with different services to see which works best for your needs.

---

10-Prompt Starter Pack ^{^}

1. Explain [topic] to me like I’m a beginner

• Tweak: Add “using everyday analogies” or “in under 100 words”*
• Expect: Clear, jargon-free explanations with helpful comparisons*

2. Help me write a [type] email to [recipient] about [situation]

• Tweak: Add “professional tone” or “casual but polite”*
• Expect: Complete draft email with appropriate greeting, body, and closing*

3. What are the pros and cons of [decision/choice]?

• Tweak: Add “focus on financial aspects” or “consider family impact”*
• Expect: Balanced list with reasoning for each point*

4. Create a step-by-step plan for [goal/project]

• Tweak: Add “with specific deadlines” or “for a complete beginner”*
• Expect: Detailed action items in logical order*

5. Summarize this text and highlight the key points: [paste text]

• Tweak: Add “in bullet points” or “focus on actionable items”*
• Expect: Concise overview with main ideas emphasized*

6. Help me brainstorm ideas for [creative project/problem]

• Tweak: Add “outside-the-box thinking” or “practical solutions only”*
• Expect: List of diverse, relevant suggestions*

7. What questions should I ask when [situation like buying a car, choosing a doctor]?

• Tweak: Add “prioritize most important” or “include red flags to watch for”*
• Expect: Comprehensive list of relevant questions to consider*

8. Rewrite this text to be more [professional/casual/persuasive]: [paste text]

• Tweak: Add “keep same length” or “make it more concise”*
• Expect: Revised version matching your specified tone*

9. Compare and contrast [option A] vs [option B]

• Tweak: Add “focus on cost differences” or “which is better for beginners”*
• Expect: Side-by-side analysis with clear distinctions*

10. Check this for clarity and suggest improvements: [paste your writing]

• Tweak: Add “focus on grammar only” or “suggest better word choices”*
• Expect: Detailed feedback with specific suggestions for enhancement*

---

Remember: These AI tools work best when you treat them as collaborative partners rather than infallible authorities. Start with simple requests, verify important information, and don’t hesitate to ask follow-up questions or request clarifications.

Appendix - image prompt ^{^}

This is the prompt I gave to Google Gemini to produce the image at the top of this article:

Using pencil sketch style, give me an image of a 30-ish middle-class American woman at a laptop computer. Our viewpoint is 45-degrees off-center from looking directly at her. We are looking somewhat over the top edge of the laptop, so that we can see all of her face.

The lighting is diffuse, as if from a nearby window. The background is only vaguely sketched in, and is of a living room. Leaning over her shoulder and helpfully pointing out something on the laptop screen is an android (human-form) robot representing AI. Perhaps it is wearing a T-shirt with the “AI” initials on it.

Notes ^{^}

 

☀   ☀   ☀