Comment by pnt12

> Basically AI algorithms learn patterns, not abstraction

I've been playing around with chat GPT and trying to find useful cases.

I asked it to come up with python assignments, such that I could train for a certification, and I provided the syllabus, which includes the datetime and calendar module. Among the many syggestions, one stood out to me: creating a CLI which accepts a date and uses those modules to calculate the following full moon and print it.

Sounded cool, but there's a catch: those modules don't have any information about moon cycles!

I guess it matched some pattern regarding dates and moon cycles to produce this bad answer. It's a creative suggestion, but quite unfeasible!

We cannot use these models as experts who provided answers, but instead as creative, sometimes nonsense assistants, who can give their input but we can clearly filter the good ideas from the nonsense.