EP 5: HOW TO USE AI TOOLS WITHOUT GETTING BURNED BY THEM
DESCRIPTION
"Forget the hype cycle. This episode breaks AI into pieces you can actually use: the broad umbrella term, machine learning underneath it, and generative tools like ChatGPT, Claude, Gemini, and Copilot on top. These models predict tokens to produce fluent language, and fluent is not the same as true.
I get into the difference between chat and search, why treating a chatbot like a fact engine will burn you, and where these tools genuinely help: drafting, summarizing dense documents, untangling messy email threads, comparing options when you provide the specs.
The practical core is a five-point checklist: define the role and quality bar, add constraints, treat output as a draft rather than an authority, learn the red flags, and protect sensitive data. And for legal, medical, or financial decisions, get a qualified human involved. The tools are useful. Trusting the confident tone is the mistake."
TRANSCRIPT — EP 5
You've probably heard someone say AI is gonna take your job and someone else say AI can't even count fingers in a photo both of those reactions come from Well the same problem. We're using the word AI like it's one Single magic machine and guess what? It's not today. I'm gonna give you a simple mental model for what what AI is, what chat GPT and friends are, and how to get the benefits without falling for all the hype. Welcome back to the Tyler Woodward Project. I'm Tyler, a broadcast engineer by trade, a Linux
nerd by choice, and I enjoy demystifying tech that's supposedly quote unquote too complicated for people. Today we're answering a deceptively basic question. What is AI and what are chat GPT, Claude and Gemini? Here's the plan. First, we'll define AI in a way that matches how it shows up in your life. Then I'll pull the covers off of these chat tools. What they're good at and where they get weird. Finally, I'll give you a practical checklist for using them safely and effectively. One quick note, I'm staying
platform neutral. The brand names change fast, but the underlying ideas are pretty stable. So let's get into it. When people say AI, they might mean three different things. If you don't separate them, everything sounds contradictory to each other. First, AI is kind of being used as this umbrella term. This includes lots of techniques that help computers do things we associate with intelligence, recognizing speech, finding patterns, making predictions, generating text and so on
and so forth. Second is machine learning. That is when you train a system on examples instead of hand coding every single rule. Spam filters, photo tagging, and recommendation systems are all common examples of this. Number three. And this is what's dominating headlines right now. Generative AI. That's when a system generates new content like Text images audio or code based on patterns it learned during its training now tools like chat GPT Claude Gemini Co -pilot,
they're all chat style assistance built on something called and an LLM it stands for a large language model an LLM is trained on a huge amount of text so it can predict what comes next in a sequence a useful way to think about it is it's like autocomplete but scaled up massively and then tuned to follow instructions and and hold a conversation basically and it's important to say this out loud these tools don't know things the way a person knows things they don't have senses they don't have lived experience, they don't automatically check
reality before they answer. What they are, are great at producing plausible language. Often correct, sometimes wrong, occasionally wrong in a way that still sounds overly confident. You'll hear a term, hallucination. In this context, it means the model generated an answer that sounds legitimate, but it's not grounded in real facts. It's not lying the way we would think of lying in a human sense. It's generating the most likely next words. Here's why this matters. If you treat
a chat bot like a search engine, You're going to get burned. Search engines try to point you to sources. Chatbots try to produce a response. The difference changes how you should trust the output. One level deeper, LLMs operate on tokens, which are chunks of text. Think pieces of words. When you ask a question, the model predicts the next token, then the next. then the next, until it builds an answer. If you're thinking, so it's guessing, yeah, yeah, sort of, but it's guessing with statistical models that, you know, it learned
a ton about how language usually works. This reminds me of audio noise reduction and broadcasting when it's dialed in, it's magic, but If you push too hard, you get watery artifacts that still sound smooth. Well, AI techs can be similar. Polished output can hide problems. So you need a quick reality check. So where do chat GPT, Claude and Gemini and all these other chat bots fit in? Broadly, they're different products built on different model families. with different tuning and safety rules and sometimes different add
-ons like reading files, analyzing images, or using web tools. If you've ever tried two assistants with the same prompt and getting different answers, that's normal. Different training, different tuning, different guardrails. The big beginner takeaway, don't ask which one is the smartest. ask which one fits my task and my risk level. Let's connect this to real life with a scenario where these tools shine and fail at the same time. You've got a pile of information and you want clarity fast. Maybe you're writing a cover letter. Maybe you're trying to understand a medical
bill. Maybe you're comparing laptops or maybe you're, I don't know, staring at a long email thread and you just need actionable items. Chab bots are great at turning a messy prompt into structure. If you paste a job description in your resume and you say, draft a cover letter with a confident, friendly tone, well, you'll usually get something usable in a few seconds. And that matters because it lowers the activation energy, if you will. You're not stuck staring at a blank page. You can react to a draft instead
of inventing one from nothing. But here's the failure mode. The chatbot will happily fill in gaps with confident nonsense. If you say compare these two laptops but you don't provide exact model numbers and specifications the assistant may quietly invent some details for you or it might assume you mean a popular model and give you a comparison that sounds right but doesn't match what you're actually looking for so here's a safety a safety pattern. Make the AI stay grounded in what you provide and make it show its work
in a way that you can verify. Instead of asking which laptop is better, which is kind of vague, ask, here are the specs I'm looking at. Create a table of differences using only what I pasted. Then tell me which one fits video editing. better and why. Now you're forcing it to stay inside the fence and you're getting a decision based on constraints that you actually care about. Another practical example, summarize a long document. These tools are excellent at pulling out themes, turning paragraphs into bullet points and translating dense language into plain usable English. And
I'll make this personal for a second. For me, with ADHD, attention is a limited resource, all right? If I'm staring at a long article, a really overly dense PDF, or even a messy email thread, my brain can bounce off. It's just, it's not going to tune in. But if I can paste it into a chat bot and say, summarize this into a detailed list, break it into sections, and pull out actionable items and deadlines, that, for me, has been genuinely amazing. It's been, I dare say, life -altering. It's really helped. It turned something I'd avoid
normally into something I can actually start with. And starting, that's usually the hardest part, right? Now, all this goes without, you know, I've... I've got to at least give you some caution, all right? So here's where we're going to get into the caution part. If the document was legal, financial, or medical implications, treat the summary like helpful notes, not a final product. Verify important points in the original text. And if you're making a serious decision, Use a qualified professional. One more cultural
thing that's worth saying out loud. We're trained to trust confident language. These tools can generate confident language on demand. So we have to shift trust away from tone and toward verification. Let's turn this into, well, a quick beginner checklist you can use today. First, tell it what you want, who it should be, and what good looks like. Act like a tutor. Explain this at a beginner level. Give one example, then quiz me. Second, add constraints, so it can't
drift off. Use only the information I provide. If you're unsure, ask me clarify questions before answering. I do that sometimes, and it will indeed actually ask you stuff back to try to figure out what you're getting at, where the end goal is. Third, use it for drafts and structures, not final authority. It's fantastic for outlines, rewrites, and brainstorming. It's risky as a pure fact source, unless it can point you to verifiable references you can check. Fourth, learn the red flags. If it gives you very specific numbers, quotes, or according to a study, but
doesn't give you the source of what that study actually is, treat that as a cue to verify. If it can't provide a grounded answer, you can ask it what it would need to know or what sources you should consult. And fifth, protect sensitive information. Assume anything you paste could be stored depending on the service and settings. Don't paste passwords, private keys, or confidential work documents unless you explicitly know your workflow is approved and safe. Don't get yourself in trouble. That's the point. used well these tools can save time and reduce friction used
carelessly they can quietly inject errors into important decisions so the next time someone says AI you can ask do you mean machine learning in general or do you mean a generative chatbot an LLM that's great at language but not automatically great at truth visit tylerwoodward .me, follow at tylerwoodward .me on Instagram and threads, subscribe and like the show on your favorite podcast platform. I'll catch you next week.