Getting Started with AI (Without the Hype)

What people use it for

It helps to split the picture in two: what you can try yourself today, and where industries are already putting AI to work.

Everyday and creative tasks

Write and debug code
Draft copy: blog posts, ads, social posts
Generate images, audio, and video
Build 3D assets
Sketch UI/UX layouts
Crunch data and untangle messy problems

Zoom out to industry, and the map gets wider. The table below is a snapshot, not a full catalog—plenty of these are pilots or niche deployments, not something you will touch on day one:

Industry	Common uses (examples)
Healthcare	Medical imaging, clinical decision support, drug discovery
Life sciences	Protein folding, genomic analysis, drug target design
Finance	Fraud detection, credit scoring, quant trading
Manufacturing	Predictive maintenance, defect inspection, digital twins
Retail & e-commerce	Recommendations, demand forecasting, dynamic pricing
Education	Personalized learning, assignment grading, language practice
Agriculture	Precision farming, pest and disease detection, yield forecasting
Transport & logistics	Autonomous driving, route optimization, warehouse picking and delivery planning
Media & entertainment	Image and video generation, post-production automation, game NPCs and procedural content
Software & security	Coding assistance, threat detection, DevOps automation
Legal & professional services	Contract review, case research, due diligence documents
Energy & public sector	Grid load forecasting, equipment inspection, disaster early warning

That is real utility, not just investor theater. If you are just getting started, you do not need to open with self-driving cars or drug discovery—a conversational tool is enough.

Start with ChatGPT

For a first taste of AI, ChatGPT is still the lowest-friction on-ramp. Talk to it like a person, ask questions, ask it to spin up an image—no elaborate setup required.

Never tried it? Open ChatGPT and paste one of these:

Today is 2026-03-15. What is the weather like in central Taiwan?

Today is 2026-03-15. My zodiac sign is Aries. What is my lucky color today?

Dropping today's date into the prompt is a small habit that surprisingly often helps.

On ChatGPT, Gemini, Claude, and similar tools, your chats may feed model training—defaults vary, and you do not always need a paid plan to opt out. Before you get going, check privacy settings for chat history and whether conversations are used for training.

What an LLM actually is

You cannot talk about AI for long without running into LLMs. AI is the big umbrella; large language models are the piece most of us touch every day.

Vendors train them on massive text corpora so the model internalizes how language works. Under the hood it is math: given what you have typed so far, it scores what token might come next, then picks one—sometimes the safest bet, sometimes with a dash of randomness.

Attention(Q, K, V) = softmax(QK^T / sqrt(d_k)) V

Think of a trained LLM as the language brain. It reads context and writes back in natural language. It can answer questions. It still has no hands—it will not run errands on its own.

Products vs. models

In casual conversation, people ask which "model" you prefer. Here we mean language models—LLMs—not every kind of AI model.

The app is not the model

Say you love ChatGPT. Easy to call that your favorite "model." Not quite. Open ChatGPT, Gemini, or Claude and you will find a model picker inside. In ChatGPT that might be GPT-5.4, GPT-5.4 mini, or GPT-5.4 nano. ChatGPT is the product. GPT-5.4 is the model.

OpenAI and ChatGPT

OpenAI is the company. ChatGPT is the chat product—and that product ships with multiple models, including the GPT-5.4 family above.

Claude and Gemini

Claude comes from Anthropic. Common picks include Opus 4.6, Sonnet 4.5, and Haiku 4.5.

Gemini is Google's line. Options include Gemini 3.1 Pro, Gemini 3 Flash, and Nano Banana Pro.

Model names churn constantly. Treat the list above as a snapshot—check each vendor's docs before you commit.

Picking a model

Different training data, tuning, and infrastructure mean the same prompt can land differently depending on which LLM answers it.

Same prompt, different vibe

Real example: I asked ChatGPT and Gemini the same thing.

I was diagnosed with XX at the hospital. What should I do?

ChatGPT went reassuring—common condition, watch diet and exercise, keep up with checkups. Gemini went colder and more procedural: here is what could happen, here is what to do next, here is what to watch for. Same question, very different bedside manner.

The facts rarely diverge wildly. I do not stick with one LLM for everything. I throw the same question at a few, compare the answers, and triangulate—same as asking friends with different personalities.

Only want one to start? ChatGPT, Gemini, and Claude all have free tiers. Spin them up and see which one clicks.

Match the model to the job

That does not make the picker meaningless. More data and compute usually means a sharper model and a higher bill. In practice:

Repetitive, low-stakes work: a cheaper model like Haiku 4.5
Heavy analysis or reasoning: a pricier model like GPT-5.4

We will dig into model choice and usage patterns in later posts.

Closing thoughts

While writing this, it was already obvious that AI had changed how people look things up. Chatbot first, authoritative sources second. Sometimes that works. Sometimes it really does not.

Underneath, it is still next-token math driven by training data and your prompt. It can sound dead certain while being wrong—the 2024 BCCRT 149 episode is the kind of story that sticks.

People call AI an amplifier for a reason. Wins arrive faster. So do mistakes. Worth using—and worth staying skeptical while you learn to steer it.

For the vocabulary rabbit hole, see our AI glossary.