What is a Large Language Model (LLM)?

An LLM is an AI system trained on massive amounts of text that can understand and generate human language. Think of it as a universal assistant that has read billions of documents. Examples include ChatGPT (OpenAI), Claude (Anthropic), and Gemini (Google).

Tokens are the units AI processes text in — roughly corresponding to parts of words. One token is approximately 3/4 of a word. Most AI services charge based on tokens used. A 500-word email uses roughly 650-700 tokens.

What is an AI hallucination?

A hallucination is when AI generates information that sounds completely plausible and confident but is factually incorrect or entirely made up. It happens because AI predicts what should come next statistically, not because it knows facts. Always verify AI outputs, especially for health and fitness information.

RAG stands for Retrieval-Augmented Generation. It's a technique where AI accesses specific knowledge bases or documents to provide more accurate, grounded answers instead of relying solely on its training data. Think of it as giving AI a reference manual to fact-check against.

Chapter 03: AI Decoded — What AI Actually Is

The Core Concept

AI Is Software That Learned From Examples

Strip away all the hype, fear, and sci-fi imagery. At its core, AI is just software that was trained on large amounts of data to recognize patterns and generate useful outputs. That's it.

🏋️

Fitness Analogy

AI Is Like a Very Well-Read Personal Trainer

Imagine a personal trainer who has read every fitness textbook ever written, memorized every study, reviewed millions of workout logs, and studied every nutrition plan in existence. They haven't personally lifted a single weight — but they can synthesize everything they've read to generate a remarkably useful workout plan for your specific client.

That's what an LLM (Large Language Model) like ChatGPT does. It has "read" billions of documents and learned patterns in language so well that it can generate new, useful text based on your instructions. It doesn't "know" things the way you do — it predicts what should come next based on patterns.

This is why your expertise still matters. The AI generates the first draft; you apply the judgment, experience, and human understanding that no amount of training data can replicate.

How AI Technologies Stack — From Broad to Specific

Artificial Intelligence

The broadest category — any software that mimics intelligent behavior

↓

Machine Learning

AI that learns patterns from data instead of following rigid rules

↓

Deep Learning

ML using neural networks to find complex patterns in huge datasets

↓

Generative AI / LLMs

Creates new content (text, images, video) — this is ChatGPT, Claude, Gemini

What You Need to Know

8 AI Concepts — Translated for Fitness

You don't need to know all 500 terms in an AI glossary. You need these 8 concepts to make smart decisions about AI in your business.

Large Language Model (LLM)

Core

AI systems trained on billions of documents that understand and generate human language. They predict what text should come next — astonishingly well. ChatGPT, Claude, and Gemini are all LLMs.

Fitness Translation

Think of LLMs as a universal assistant that has studied every fitness business document, email, proposal, and marketing campaign ever written. Ask it to write a client re-engagement email and it draws on patterns from millions of examples.

Generative AI

Core

AI that creates new content — text, images, video, audio, code — based on what it learned from training data. It doesn't copy; it generates novel outputs following learned patterns.

Fitness Translation

Like a trainer who has seen thousands of program designs and can generate a new, unique workout plan that's never existed before — optimized for your specific client's goals, equipment, and constraints.

Tokens

Important

The units AI processes text in — roughly ¾ of a word. A 500-word email ≈ 650-700 tokens. Most AI services charge by tokens used. Longer inputs and outputs = more tokens = more cost.

Fitness Translation

Tokens are like reps. Just as a workout is measured in sets × reps, AI usage is measured in tokens. A quick email is a light set; a full proposal is a heavy volume day. Understanding this helps you manage costs.

Context Window

Important

The maximum amount of text (in tokens) an AI can consider at once. Bigger windows = AI can reference more information per conversation. GPT-4 has ~128K tokens; Claude 3.5 has ~200K tokens.

Fitness Translation

Think of it as the AI's working memory — like how many client files a trainer can keep in mind during a single consultation. A bigger context window means the AI can consider your entire product catalog, client history, and business rules simultaneously.

Hallucination

Critical

When AI confidently generates information that sounds plausible but is factually wrong or entirely fabricated. This happens because AI predicts what should come next statistically — it doesn't "know" facts.

Fitness Translation

Like a new trainer who sounds confident recommending exercises for a pregnant client but hasn't actually studied prenatal exercise guidelines. The recommendation sounds right but could be dangerous. Always verify AI outputs, especially for health information.

Prompt Engineering

Core

The skill of writing effective instructions for AI to get the results you want. Better prompts = dramatically better outputs. It's the single highest-leverage skill for any AI user. (Chapter 05 is entirely dedicated to this.)

Fitness Translation

Like the difference between telling a trainer "make me a workout" vs. "design a 4-week hypertrophy program for a 42-year-old with a rotator cuff history, 4x/week, 60 min max, compound-focused." Same AI, wildly different results.

RAG (Retrieval-Augmented Generation)

Advanced

A technique where AI accesses specific knowledge bases or documents to provide grounded, accurate answers instead of relying only on training data. Dramatically reduces hallucinations for specialized topics.

Fitness Translation

Instead of a trainer answering from memory, RAG gives them your specific gym's SOPs, equipment manual, and pricing sheet to reference. The answers are grounded in YOUR actual business data — not general knowledge. This is how GymSpotter AI works.

Agentic AI

Advanced

AI systems that can plan, reason, and complete multi-step tasks across different tools autonomously. Instead of answering one question, they accomplish goals — booking appointments, updating CRMs, sending follow-ups, generating reports.

Fitness Translation

The evolution from "AI answers questions" to "AI does work." Instead of asking AI to draft an email, an agent notices a member hasn't visited in 14 days, checks their goals, writes a personalized re-engagement message, schedules it, and logs the outreach — all autonomously.

Quick Reference

AI Glossary — The Rest of the Terms You'll Encounter

Bookmark this section. When you hit an unfamiliar term in later chapters or in the wild, come back here.

Machine Learning Core

Software that learns patterns from data to make predictions without being explicitly programmed. Fitness use: predicting which members will cancel based on visit frequency patterns.

Deep Learning Core

Advanced machine learning using neural networks (inspired by brain structure) to find complex patterns. Powers image recognition, speech-to-text, and most modern AI. Fitness use: analyzing exercise form from video.

Training Data Core

The information used to teach an AI system. Quality and diversity of training data determines how well the AI performs. Garbage in, garbage out. Fitness use: why AI might know more about bodybuilding than Pilates — depends on training data.

Fine-Tuning Advanced

Further training a general AI model on specialized data to make it better at specific tasks. Fitness use: training an AI specifically on your gym's brand voice, pricing, and services so it responds like a team member.

Multimodal AI Advanced

AI that processes multiple data types simultaneously — text, images, video, audio. Fitness use: upload a photo of a gym floor and ask AI to suggest equipment layout improvements while describing the reasoning in text.

API (Application Programming Interface) Technical

How different software systems talk to each other. Think of it as a universal translator between apps. Fitness use: your booking system talking to your CRM talking to your email marketing — all automatically.

Natural Language Processing (NLP) Core

AI's ability to understand and work with human language — reading emails, analyzing reviews, understanding voice commands. Fitness use: automatically categorizing member feedback by sentiment and topic.

Computer Vision Advanced

AI that understands images and video — identifying objects, people, movements, and patterns visually. Fitness use: AI form-checking during exercises, occupancy counting, equipment usage analytics from security cameras.

Chatbot vs. Agent Important

Chatbot: answers questions (reactive). Agent: accomplishes goals across multiple steps and tools (proactive). GymSpotter AI uses agents, not just chatbots. Key difference: chatbot tells you the answer; an agent does the work.

Automation vs. AI Important

Automation: follows rigid rules (if X then Y). AI: makes judgment calls based on patterns. You need both. Automation handles the predictable; AI handles the nuanced. Fitness use: automation sends birthday emails; AI decides which lapsed members to re-engage and how.

"My background in philosophy and cognitive science isn't just academic — it's operational. Understanding how people think and make decisions lets me build systems that actually get used. I've seen too many 'perfect' systems fail because they ignored human behavior."

— Robert Szopa, AI Architect

💡 Chapter Takeaways

AI is software that learned from examples — not magic, not sentient, not coming for your job
LLMs predict what should come next based on patterns — they don't "know" facts, so always verify outputs
Tokens are how AI measures text — understanding this helps you manage costs and write better prompts
Context window = working memory — bigger windows let AI consider more of your business data at once
Hallucinations are real — especially critical for health/fitness information; human review is non-negotiable
Prompt engineering is the highest-leverage skill — Chapter 05 dives deep into fitness-specific templates
RAG makes AI business-specific — this is how GymSpotter grounds answers in your actual data
Agentic AI is the future — from "AI answers" to "AI does work" autonomously across your tools
You don't need to understand the engineering — you need to understand the capabilities, limitations, and applications