Alignment is the work of making AI systems do what people actually intend, and behave in line with human values, rather than just what they were literally told.
In plain words
Large language models can produce answers that are untrue, toxic or just unhelpful, because predicting text was their goal, not doing what the user meant. Alignment covers the methods that steer them toward helpful, honest and safe behaviour, such as fine-tuning on human feedback, and the checks that test whether it worked. It matters more as systems become more capable.
A quick example
Imagine an early language model asked to explain something to a child: it might ramble or invent facts. After training with human feedback, the same kind of model follows the request more closely. That shift, toward what the user meant, is alignment in practice.
Where you'll meet it
- Post-training chat models with human feedback.
- Safety research on how to control more capable systems.
Related
Sources
- PaperTraining language models to follow instructions with human feedback, Ouyang et al. (arXiv) · read 29 Sept 2026
- PaperAI Alignment: A Comprehensive Survey, Ji et al. (arXiv) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.