Comment Reasonably useful for some menial tasks (Score 1) 248
Scanning somewhat predictable documents to pick out specific data and then using traditional text matching tools to verify if the information.
From my experience there is about 5% error rate due to the randomness of MML engines, but building a simple loop that sends the request back through the engine with an adapted prompt specifying to ignore a specific text, usually gets the correct result. Stopping and erring out after 3 loops seems to bring the error rate down to less than 1% - So, what is this good for? Well, it can save lots of time in text input, where a document needs to be read and inputted into a form and saved to a database for search and statistics. - Cutting hours of the menial "read document and type into form in web app" tasks away.
So far my experience with code generation is laughable at best. And text generation is equally bad.