Module 1 · Chapter 1

What Is an LLM and How Does It Work?

Understanding AI without the tech jargon — and why it changes everything for paralegals

📺 Video: How LLMs Work — For Legal Professionals

Search YouTube: How LLMs Work — For Legal ProfessionalsOpens YouTube →

📺 Suggested supplemental viewing — not required, but highly recommended before reading.

Why This Module Matters

You have heard the name. You may have tried one. You have almost certainly wondered whether you should be worried about it — whether it is coming for your job, whether it is as powerful as people claim, or whether it is mostly hype.

The answer is more interesting than either extreme. AI language tools are genuinely remarkable. They are also genuinely limited in ways that the headlines rarely explain. Understanding both sides — the remarkable and the limited — is the foundation of using AI well in legal work.

🎯
The One Sentence That Matters: AI language tools are brilliant, useful, and need supervision. Everything else in this module is detail.

🧑‍💼 The Brilliant Intern — Your Mental Model

Before any technical explanation, here is the mental model that works best for legal professionals:

Think of an AI language tool as a brilliant intern.

This intern has read more than any human ever could — millions of books, articles, court opinions, contracts, statutes, legal briefs, and websites. They absorbed all of it before their first day.

They write fluently. They draft letters, summarize documents, explain legal concepts, and organize information quickly and professionally. They are fast, tireless, and always willing to help.

But here is the catch: they cannot look anything up. Everything they know, they learned before they arrived. If the law changed last month, they may not know. If a case was overturned last year, they may not know. And — crucially — when they are unsure about something, they do not always say so. Sometimes they fill in the gap with a confident-sounding answer that turns out to be wrong.

They are brilliant. They are useful. They need supervision.

This analogy will serve you throughout this course. Every time you wonder whether to trust AI output, ask: Would I send this to the attorney without checking it?

📚

Training

Absorbed billions of words of text before you ever typed a prompt

🔮

Prediction

Generates the statistically most likely next word — not true reasoning

📅

Cutoff

Knowledge stops at a point in time — cannot retrieve new information

Module 1 Learning Objectives

After completing this module, you will be able to:

  • 1Define large language model (LLM) and explain how it generates text using the autocomplete and brilliant intern analogies.
  • 2Describe the training process for AI tools and explain what a knowledge cutoff means for legal work.
  • 3Explain hallucination — why it happens, what it looks like, and why it is especially dangerous in legal contexts.
  • 4Debunk the six most common myths about AI tools using evidence from the chapter's Myth vs. Reality framework.
  • 5Apply the Brilliant Intern test to at least five routine paralegal tasks to determine where AI assistance is appropriate.

Key Terms to Know

LLM — Large Language Model
Training Data — Text the model learned from
Knowledge Cutoff — When training data ends
Hallucination — Confident, incorrect output
Prompt — Your input to the AI
Token — Unit of text the model processes
Context Window — Amount of text the AI can process at once
Fine-tuning — Additional training for specific tasks

📺 Supplemental Video: AI in the Legal World

Search YouTube: Legal AI OverviewOpens YouTube →
📖
Reading Assignment: The AI-Powered Paralegal, Chapter 1 (pp. 1–20). The excerpts below highlight the essential concepts. Complete your full chapter reading alongside these materials.

Section 1.1: The Autocomplete Analogy

You already understand the core concept behind LLMs — you use it every day:

When you type a text message on your phone, your keyboard suggests the next word. A large language model does exactly this — at a scale that is almost incomprehensible.

Instead of learning from your messages, it learned from hundreds of billions of words of text. Instead of predicting one word at a time from a simple pattern, it predicts the most likely next word by analyzing the full context of everything that came before it — using a mathematical structure so complex that it captures nuance, tone, domain knowledge, and logical relationships.

💡
Key Insight: An LLM is not thinking. It is not reasoning the way you reason. It is doing extraordinarily sophisticated pattern prediction — generating text that is statistically likely to follow from the text you gave it. The output often looks like reasoning, reads like analysis, and sounds like expertise. Sometimes it is. Sometimes it is a very convincing pattern that happens to be wrong.

Section 1.2: The Knowledge Cutoff

Every AI tool has a training cutoff date — the point after which its knowledge is incomplete or absent. For legal work, this matters enormously:

⚠️ Knowledge Cutoff Implications for Legal Practice

  • A statute may have been amended after the cutoff
  • A controlling case may have been overturned
  • A regulation may have been revised or replaced
  • New court rules or local rules may have taken effect

Always verify currency of any legal authority the AI provides. The AI cannot tell you that its information is outdated — it may not know. That verification is your professional responsibility.

Section 1.3: Hallucination — The Most Important Concept

When you ask an AI tool for a specific case citation and no relevant case exists in its training data, it does not say "I don't know" — at least not reliably. Instead, it generates text that looks like a case citation, because that is what typically follows that kind of question in its training data.

The result is a citation that is formatted correctly, sounds plausible, and does not exist.

🔍 Spot the Hallucination

An AI was asked: "Provide three Florida cases supporting the discovery rule in negligence." It returned the following. Click each citation to reveal whether it's real or hallucinated.

?
Martinez v. Dade County General Hospital, 847 So. 2d 1124 (Fla. 3d DCA 2003).
?
Tanner v. Hartog, 618 So. 2d 177 (Fla. 1993).
?
Kowalski v. Suncoast Medical Partners, Inc., 921 So. 2d 443 (Fla. 2d DCA 2006).

This is why verification is non-negotiable. AI cannot tell you when it's making something up.

Section 1.4: Myth vs. Reality

AI has attracted more mythology than almost any technology in recent memory. Head to the Activities tab for an interactive exercise on these myths — but here is the quick reference:

❌ The Myth ✅ The Reality
AI is just a search engine with a better interface. A search engine finds existing pages. An AI generates original text — which may or may not be accurate.
AI understands what it's saying. AI processes patterns. It does not comprehend law — it generates text that resembles legal analysis.
If AI says it confidently, it's probably right. Confidence is a stylistic feature of AI output, not an indicator of accuracy.
AI is going to replace paralegals. AI will replace paralegals who don't use it. AI-fluent paralegals have a growing advantage.
AI has access to real-time legal information. Most AI tools work from training data with a cutoff date. Web-enabled tools still don't replace Westlaw/Lexis.
AI legal tools like Harvey are the same as ChatGPT. Legal-specific tools are built on LLMs but trained on legal data and designed for legal tasks — meaningfully different.
Activity 1

Myth vs. Reality: Card Flip

Click each card to flip it and reveal the reality behind each AI myth. Test yourself before reading the answer!

✗ The Myth

AI is just a search engine with a better interface.

✓ Reality

Search engines find existing pages. AI generates original text that may or may not be accurate.

✗ The Myth

If AI says it confidently, it's probably correct.

✓ Reality

Confidence is a stylistic feature, not an accuracy indicator. AI sounds confident even when completely wrong.

✗ The Myth

AI understands legal concepts the way a lawyer does.

✓ Reality

AI processes statistical patterns. It generates text that resembles expert analysis — it does not comprehend law.

✗ The Myth

AI is going to replace paralegals.

✓ Reality

AI will replace paralegals who don't use it. Those who use AI fluently will handle more work with greater value.

✗ The Myth

AI has access to real-time legal databases.

✓ Reality

Most AI works from training data with a cutoff date. Even web-enabled AI doesn't replace Westlaw or Lexis.

✗ The Myth

Legal AI tools like Harvey are the same as ChatGPT.

✓ Reality

Legal-specific tools are fine-tuned on legal data with citation verification — meaningfully different for professional use.

Tip: Click any card to flip it. Click again to flip back.
Activity 2

The Brilliant Intern Test

You are a paralegal. Review each AI-generated output and decide: Safe to forward to the attorney as-is? Apply the Brilliant Intern test.

Activity 3

Key Term Matching

Click a term on the left, then click its matching definition on the right.

Activity 4

AI-Appropriate vs. Needs Verification

Drag each paralegal task into the correct category. Where would you apply the Brilliant Intern test and check the output before using it?

Practice Quiz — Module 1

Test your understanding. Take as many times as you like — your highest score counts. Detailed feedback is provided after each submission.

Module 1 Assessment

This is your graded module assessment. You have 2 attempts — your highest score counts toward your final grade.

⚠️
Complete the practice quiz and all activities before attempting the module assessment. This assessment includes applied scenarios in addition to conceptual questions.