Generative AI
Generative AI is AI that produces new content, text, images, audio or code, rather than only analysing or labelling existing input. In translation, the generative model is what writes the target text. The same generative ability that makes it useful is what lets it produce fluent output that is not grounded in the source.
How it works
A generative model learns the patterns of its training data, then samples from those patterns to create output that fits a prompt or an input. For translation, the input is the source text and the output is the target text.
Because it generates rather than retrieves, it is not bound to say only true things. Keeping generative output faithful, to a source, to facts, to terminology, is the job of the surrounding process, not the model alone.
How SourceTarget uses it
Where a generative model is the best drafting engine for a job, SourceTarget uses it inside the same human-in-the-loop workflow as any other machine output. Confidence and quality scoring, and human review, are what keep generative fluency from turning into published error.