AI glossary · Models
Model training
In one sentence
Model training is the process in which an AI model learns from a dataset: it's shown examples, its errors are measured and its parameters are adjusted, over and over, until its outputs are good.
Also known as: AI training · training a model · machine learning training · model fitting
What is model training?
A freshly created AI model knows nothing: its parameters hold random values and so do its answers. Training is what turns that pile of random numbers into something useful, by correcting it millions of times against examples.
Put it to work
AI agents that look after your customers
From €49/mo + VAT
Voice, WhatsApp and email with the phone line included: agent, numbers and CRM in one fee.
See plans and pricingIt’s the expensive part of AI. Training a large language model from scratch takes huge amounts of data, weeks or months of computing on thousands of specialised chips, and dedicated engineering teams. Only a handful of organisations do it; everyone else uses models that are already trained.
Related to this term
UTIO agents
- Available 24/7 in 26 languages
- Set it up yourself for €0: built from your website
- 14-day free trial, no card needed
From €49/mo + VAT
See plans and pricingHow it works
The basic loop runs again and again:
- Show an example (or a batch of examples) from the dataset.
- Compute the model’s answer using its current parameters.
- Measure the error with a function that compares that answer with the correct one.
- Adjust the parameters slightly in the direction that reduces the error. Neural networks do this with backpropagation and gradient descent.
Each full pass through the dataset is called an epoch. Throughout, accuracy is tracked on validation data to catch overfitting: a model that memorises the examples instead of learning to generalise.
Language models are usually trained in stages: pre-training on large volumes of text, instruction tuning on example conversations, and preference tuning, where people rate answers so the model learns which are more helpful and safer. Further adjustment on your own data is called fine-tuning.
Example
A small chain of opticians in Belfast wants to predict which customers are due an eye test each month so it can remind them. It trains a simple model on its visit history: age, date of the last prescription, lens type and whether the customer came back.
In the first passes the model gets a lot wrong. After several epochs the error drops and levels off. The team then tests it on customers from the last quarter, whom the model has never seen, to check it works on new cases and hasn’t just memorised the old ones.
What it means for your business
In all likelihood you’ll never train a model. For customer-facing work, the practical route is to use an already-trained model and give it your information with each request, which is what RAG does. Change a price, update the document, done: no retraining.
That’s how UTIO’s agents work: the WhatsApp agent and voice agent run on already-trained models and answer with what you give them. Inference, meaning day-to-day use of the model, is included in your plan, with no separate bill for tokens.
Frequently asked questions
What's the difference between training and inference?
Training is when the model learns: its parameters are adjusted using lots of examples. Inference is when the trained model does its job: it gets a question and answers without changing what it knows. Training happens rarely; inference happens on every request.
What is pre-training?
The first stage for language models: they learn to predict the next token across enormous amounts of text, which produces a general-purpose model. That model is then usually refined with example conversations and human ratings so it follows instructions and gives useful answers.
Does a chatbot learn from what I tell it?
Not during the conversation: its parameters don't change, it simply takes into account what's in its context. Whether a provider later uses your conversations to train new versions depends on its terms of use, which are worth checking.
Related terms
- Dataset
A dataset is an organised collection of examples, such as text, audio, images or records, used to train, fine-tune or evaluate an AI model. Its quality directly shapes what the model learns.
- Fine-tuning
Fine-tuning means further training an AI model that's already been pre-trained, using a small set of your own examples, so it specialises in a particular task, format or style without starting from scratch.
- Model parameters
A model's parameters are its internal numeric values, mainly the weights of a neural network, which are adjusted during training and store what it has learned. Their count is used as a rough measure of model size.
- Inference
Inference is when an already-trained AI model is put to use: it receives a new input, such as a question, an audio clip or an image, and computes a response from what it learned, without changing its parameters.
- Machine learning
Machine learning is the branch of AI in which a system learns to do a task from examples, rather than following hand-written rules, and gets better as it sees more data.
Updated on