Fuzzy and forgiving

LLMs make the interaction with software fuzzy and forgiving.

Computer programs are (were?) rigid and precise. You put the right input, you get some output. And by right, I mean what the author of the software considered to be correct.

Now, put an LLM in front of your software. It can interview you (the user) to gather the inputs, clarify parameters and suggest examples, then morph what you say into the format the software requires and, eventually, run the program and present the outputs.

Of course, helpful error messages and hints produced by the software are still good to have, just that you probably won't read them yourself anymore.

The question is: how to ensure the correctenss of both data streams? What goes in and what is presented as output. A plausible but wrong output is worse than no output at all.

And it remains to be seen if this will be the default mode of working in the next couple of years.