LLM

An LLM (Large Language Model) is the heart of ChatGPT, Claude, and Gemini. What it actually does is surprisingly simple — it's a program trained to read vast amounts of text from the internet and predict "what word comes next."

"Predicting the next word" might not sound impressive, but to do it really well, a model ends up having to absorb grammar, knowledge, and logic all at once. To correctly predict what comes after "The capital of South Korea is," it has to actually know the answer. So a machine that's good at predicting the next word turns out to also be good at translating, summarizing, and writing code.

When news mentions names like "GPT-5," "Claude Opus," or "Gemini," those are all product names for LLMs. Just as car companies release new models every year, AI companies keep releasing bigger, smarter LLMs — and those announcements are one of the most common topics in AI news.

"Anthropic unveiled a new LLM and claimed the top spot in performance" — this means a new "brain" has arrived. Often the chatbot service we use stays the same, but the brain inside it gets swapped out.

1. Ask a chatbot to "continue this sentence starting with 'The autumn sky...'" just once, then send the exact same question two more times in new conversations.
2. You'll get a different sentence each time — proof that the LLM isn't pulling out a fixed correct answer, but probabilistically picking a "plausible next word."
3. Now ask "What is the capital of South Korea?" three times. This time the answer is the same every time — the difference between a question where the probability strongly favors one answer and one where it doesn't is part of what gives an LLM its "personality."

The two terms get mixed together carelessly in articles, making them seem like the same thing. Model is the broad term for any AI program that has finished training, while large language model is the subset that handles text. A model that draws images or one that generates voices is still a model, but not a large language model.