What is an LLM?
This page introduces large language models, the technology at the center of this course. It explains why LLMs change how humans direct computers, what capabilities make them useful, and how business students can use them both to think through problems and to build practical solutions.
A working definition
A large language model (LLM) is a statistical model trained on large quantities of text and code to predict the next token in a sequence. A token is a word, part of a word, or symbol. By learning patterns across books, websites, code repositories, articles, and other written material, the model becomes able to generate useful continuations: answers, explanations, summaries, plans, tables, and code.
A useful analogy is that of a well-read generalist who has internalized patterns from many kinds of writing. Unlike a database, an LLM is not designed primarily to store and retrieve facts. It generates responses by using the input as context and producing text that fits the patterns it learned during training. For example, if asked to write a professional email, the model recognizes cues such as “professional,” “email,” and “reschedule” and generates a response that follows the conventions of business email. This explains both its power and its limits: it can produce fluent and useful answers, but it can also make mistakes, invent details, or sound confident when it is wrong.
For example, suppose you type:
Write a short professional email asking a client to reschedule a meeting.The LLM recognizes several patterns:
"Write" → produce text
"short" → keep it brief
"professional email" → use a polite business tone
"client" → be respectful and clear
"reschedule a meeting" → include apology, new time options, and a closingSo it generates something like:
Hi Jordan,
I'm sorry, but I need to reschedule our meeting originally planned for Thursday.
Would either Friday morning or Monday afternoon work for you?
Best,
...The model does not retrieve that exact email from memory. It generates a new response based on patterns it learned from many examples of professional emails, scheduling requests, apologies, and business communication.
LLM Capabilities
For many years, language models were mostly experienced as background technologies: autocomplete, search suggestions, translation, grammar correction, and text classification. They were useful, but they did not usually feel like flexible business collaborators. That changed quickly in the early 2020s. In 2022, ChatGPT made instruction-following LLMs visible to a broad public audience. In 2023, GPT-4 showed that these systems could perform more complex work across writing, coding, analysis, and professional-style tasks.
As models became larger, were trained on broader datasets, and were further adapted through instruction tuning and human feedback, they began to produce emergent capabilities: useful behaviors that were not explicitly programmed into the model or directly targeted by the training objective. As a result, modern LLMs are no longer just conversation partners. Some of their emergent capabilities are listed below.
| Emergent capability | Simple meaning | Business example |
|---|---|---|
| In-context learning | Learning the pattern from examples in the prompt. | Show three labeled customer comments; the model labels the fourth. |
| Instruction following | Carrying out a task described in natural language. | Rewrite a memo for a CFO audience. |
| Multi-step reasoning | Breaking a problem into smaller steps. | Compare pricing options using costs, demand, margin, and customer segments. |
| Code generation | Producing code from a description of the desired outcome. | Create a script that cleans a CSV and produces a summary table. |
| Tool use | Choosing when and how to use external tools. | Run Python to analyze a spreadsheet instead of guessing from the prompt. |
| Planning | Sequencing actions toward a goal. | Create a project plan for validating a new business idea. |
Exercise: Same message, different tones
Open an LLM such as ChatGPT, Claude, or Gemini. Ask it to write an email for the scenario below in three different tones.
You need to tell a teammate that their part of the project is late and ask when they can send it.
Try these three prompts:
Write this email in a friendly tone.
Write this email in a firm but professional tone.
Write this email in a concise executive tone.Compare the outputs. What changed across the three versions? Look for differences in word choice, sentence length, directness, politeness, and implied relationship with the recipient.