LLM.

An LLM, short for Large Language Model, is the technology behind AI assistants such as ChatGPT, Claude, and Gemini. According to Anthropic, LLMs are language models with a very large number of parameters that have been trained on massive amounts of text and can write texts, answer questions, and summarize content. ChatGPT is powered by OpenAI’s GPT models, Claude by Anthropic’s models, and Gemini by Google’s models. For your business, what matters most is what such a model does well and where it reliably falls short.

How an LLM Works

An LLM predicts what comes next, one step at a time. According to Anthropic, during the initial training—known as pretraining—the model learns to predict the next word based on the text that has come before. It is only through further training, including human feedback, that it becomes an assistant capable of following instructions.

The model does not process whole words, but rather tokens—that is, parts of words. OpenAI states as a rule of thumb that one token in English corresponds to about four characters or three-quarters of a word. OpenAI and Anthropic bill for their APIs based on tokens, and the context window—that is, how much text a model can process at once—is also measured in tokens. According to Anthropic, this working memory is distinct from the training data.

What Your Business Should Keep in Mind

Expect to see fabricated information. Anthropic notes that even advanced language models can generate text that is factually incorrect or doesn’t fit the given context, a phenomenon they call “hallucination.” That’s why a human always verifies numbers, prices, deadlines, and legal information before they appear on your website or in an offer.

Provide the model with the facts instead of relying on its knowledge. The knowledge gained from training does not include your prices or processes and is limited to a specific point in time. Among other things, Anthropic recommends instructing the model to work only with the documents provided and to support its statements with quotes from those documents.

Evaluate models based on your task, not on version numbers. New models are released frequently. In projects, we’ve found that a clearly worded prompt with good examples often makes more of a difference than switching to the latest model. So test them with a handful of real-life scenarios from your daily life.

Frequently Asked Questions About LLMs

Does an LLM learn from my inputs? That depends on the provider and the plan. Whether inputs are used for training is governed by the provider's privacy policy and terms of service, which you should review before using the service.

Is an LLM the same thing as generative AI? An LLM is one part of it. Generative AI also includes models that generate images, audio, or videos.

Can an LLM act on its own? On its own, it only generates text. Only when it gains access to tools and plans its own actions do we refer to it as agentic AI.

If you want to know which language model is right for which task in your business, take a look at how we approach AI for businesses.