Skip to content
Tekoälli

Layer 2

Data

What the machine learns from, and why your own records are the asset.

On this page
  1. What happens in this layer
  2. A common misconception
  3. Key terms

A model learns from examples. Volume alone settles nothing: quality, coverage, accuracy, the right to use the material and whether it resembles the real situation all matter more than size.

What happens in this layer

  1. 01

    Data comes in many shapes

    Text, images and video, audio, source code, sensor readings, a company's transaction records, question-and-answer pairs written by people, and data the machine generated itself.

  2. 02

    The hard part is organising what you already have

    Records sit in different systems, the same word means different things in different places, permissions are unclear, and some of it is simply out of date.

  3. 03

    Sensitive material needs its boundaries first

    Not everything may be shown to everyone. Access rights have to be settled before the data is put within reach of a model.

  4. 04

    Your own data may be the whole advantage

    A company's own material is often worth more than the option of training a slightly larger model on everybody else's.

A common misconception

Key terms

Whole glossary →
Embedding
A numerical representation of content that makes it possible to search for similar things.
Vector database
A system for storing embeddings and searching them by similarity.
Synthetic data
Training material that was simulated or generated by AI.
Evaluation (eval)
A systematic test of the quality of a model or of a whole application.
Overfitting
The model memorises the examples instead of learning the thing.