The three words, broken down
LLM stands for large language model. Each word is actually doing work, so it helps to take them one at a time.
A language model is software built to work with human language, specifically to predict which words are likely to come next in a piece of text. That idea existed long before ChatGPT.
Large is the part that changed everything. It means the model was trained on a huge amount of text, far more than earlier versions, and built with far more internal capacity to learn from all of it. Scale turned out to matter more than anyone expected.
Why size made such a big difference
Older language models could do simple things: finish a sentence, suggest the next word as you typed. They could not hold a real conversation, follow multi-step instructions, or explain a complicated topic clearly. They simply had not seen enough to learn those skills.
Once models were trained on vastly more text with vastly more computing power behind them, something changed. Abilities nobody explicitly trained for, holding a coherent conversation, reasoning through a problem step by step, writing in a specific style, started showing up on their own. Researchers refer to this as an emergent ability: a skill that appears once a model gets large enough, without anyone building it in directly.
The engine, not the car
Here is a distinction worth having clear in your head. ChatGPT is not itself an LLM. ChatGPT is a product, a chat interface with memory, settings, and other features built around an LLM that does the actual reading and writing underneath.
Think of the LLM as the engine and the chat app as the car built around it. The same basic engine can power different cars with different features. This is why you will hear both the model name and the product name used somewhat interchangeably, and why that can get confusing if nobody explains the difference.
What LLMs are actually good and not good at
Knowing the term is one thing. Knowing what it implies about the tool in front of you is more useful.
- → Strong at language tasks: writing, summarizing, explaining, translating, and following instructions given in plain English.
- → Weak at knowing what is actually true. An LLM predicts likely words, not verified facts, which is why it can state something false with total confidence.
- → Limited to what it learned during training, plus whatever you give it directly in a conversation. It does not automatically know about things that happened after its training ended, unless it is connected to a live search feature.