LLM (Large Language Model)

Autocomplete, but enormous

LLM stands for Large Language Model. At its heart, it does one deceptively simple thing. Given some text, it predicts the next word. If you give it "The cat sat on the", it probably continues with "mat".

Your phone's keyboard does this too. The difference is scale. An LLM has read a huge slice of the internet and has billions of dials tuned for the job. Predicting the next word this well forces the model to understand grammar, facts, logic, even code.

One word at a time

When ChatGPT or Claude "writes" you an answer, it's predicting one word, adding it to the page, then predicting the next, thousands of times, fast. There's no separate "essay plan" hiding inside. The plan emerges from doing next-word prediction with superhuman skill.

This also explains the famous weakness. An LLM says whatever is most likely. Nothing checks whether it is true. When likely and true disagree, you get confident nonsense. People call this a hallucination.

The models on this site

GPT, Claude, Gemini, Grok, Llama, and DeepSeek are all LLMs. They share the same core idea, and they differ in their training data, sizes, and tuning. That is why they have different personalities and different scores on the comparison table.