AI glossary Models & Architecture
What is training in AI and machine learning?
← All glossary termsThe process of teaching a model from data by adjusting its parameters, from general pre-training to task-specific fine-tuning.
Explained
Training is the process by which a machine learning model learns from data. The model makes predictions, an error measure compares them with the expected result, and an optimisation algorithm adjusts its parameters to reduce that error, over many passes through the data. For large language models, training happens in stages: pre-training on vast amounts of general text, then fine-tuning on specific tasks or domains, and often alignment with human feedback. Training needs data, computing power and expertise; inference is what happens afterwards, when the trained model is used.
The stages of training a language model
- Pre-training: the model learns general language and knowledge by predicting the next token across a very large corpus. This is the most expensive stage and is done by model providers.
- Fine-tuning: the pre-trained model is trained further on a smaller, specific data set to adapt it to a task, a domain or a format.
- Alignment: techniques such as reinforcement learning from human feedback teach the model to follow instructions and behave as intended.
Training vs inference
Training changes the model: it adjusts the parameters using data. Inference uses the model as it is to answer a request.
Training happens occasionally and in bulk; inference happens every time someone uses the system. For most companies the running cost of AI is inference, while training, if done at all, is usually limited to fine-tuning an existing model.
When a company should train a model
Most enterprise uses do not need training. Prompting a good model and grounding it in company data with retrieval covers many cases.
Fine-tuning makes sense when the model must follow a specific format or style consistently, master a narrow domain vocabulary, or when a smaller tuned model can replace a larger and more expensive one. Training from scratch is rare outside companies that build models. Whatever the choice, a separate evaluation set is needed to prove the trained model is better.
Frequently asked questions
What does training an AI model mean?
It means showing the model many examples and adjusting its internal parameters so that its outputs get closer to the expected ones. After training, the model can apply what it learned to new inputs.
What is the difference between pre-training and fine-tuning?
Pre-training teaches a model general knowledge from a very large, broad data set. Fine-tuning takes that pre-trained model and trains it further on a smaller, targeted data set so that it performs better on a specific task or domain.
What is the difference between training and inference?
Training is when the model learns and its parameters change. Inference is when the finished model is used to produce answers or predictions, without changing it.
Do companies need to train their own AI models?
Usually not. Many uses work with an existing model, good prompts and the company's data provided at query time. Training or fine-tuning is worth it when a task needs consistent behaviour, specialised knowledge or lower running costs that an existing model cannot deliver.
Decision guides that use this term
Where it shows up at Thinkia
Further reading
Related terms
Sources
- Brown et al. · Language Models are Few-Shot Learners (arXiv)
- Ouyang et al. · Training language models to follow instructions with human feedback (arXiv)
- Goodfellow, Bengio and Courville · Deep Learning (MIT Press)
Updated: 2026-10-21