Generative AI
Generative AI produces content, such as text, images, audio or code, from learnt patterns and the instructions it receives
What it means, how it is used and where it comes up, with its related concepts.
What it is and how it works
It can support creative and operational tasks, but its outputs need evaluating according to their use. Generating a convincing answer does not prove that it is true, original or appropriate. Implementation should define what is automated, what is reviewed and how errors are detected.
Also known as
GenAI
Topics
Frequently asked questions about Generative AI
How do you choose a first task to apply generative AI to?
Choose a well-defined task whose output can be evaluated and corrected before it is used. The relevant errors should be known, and there should be examples that allow the output to be compared with current work. A frequent task is not automatically a good fit: if checking the answer is harder than producing it, the benefit may be limited. The pilot should account for the review effort, not just the generation effort.
What should a person review in AI-generated content?
They should check the aspects that could cause an error in that particular use: facts, references, omissions, suitability for the audience and consistency with the brief. Correcting the style is not enough. When the content cites a source, it has to be opened to verify that it supports the claim. Review needs explicit criteria and enough time; assigning it without those conditions does not guarantee a reliable result.
Do you need to re-evaluate an AI solution if the model or its instructions change?
Yes, when the change could alter behavior that matters for the task. It is worth keeping a test set with typical and difficult cases to compare results across versions. You should also record what has changed and which issues remain open. An improvement in one capability does not prove an improvement in all of them; the decision to use the new version should be based on the intended context.