Decoder. plain-English AI glossary

Multitask learning

● Core

Training one model on several tasks at once so they reinforce each other.

Think of it like

Like a decathlete whose sprint training also helps their long jump — skills share a common base and lift together.

Example

A single model is trained to translate, summarize, and answer questions, and gets better at each than if trained alone.

How it actually works

Multitask learning shares parameters across related tasks, letting the model learn common structure and regularize itself, often improving each task and data efficiency. The risk is negative transfer — conflicting tasks can drag each other down — and balancing their losses is fiddly. Modern instruction-tuned models are effectively massively multitask, trained on many task types together.

For product teams

One model that does many things can be cheaper and stronger than many narrow ones — if the tasks play nicely.

For engineers

Joint training on multiple tasks with shared parameters; shared structure helps, but negative transfer can hurt.

Related

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open DecoderAdd to Chrome