1·
2 hours agoI only found it useful doing trivial chores such as converting between data structures, maybe create a test for a function, parsing and some regex. Anything deeper than that was full of errors or the it offered was suboptimal at best. It also fails a lot of times in fetching the relevant docs/sources for the discussion. I gave up trying after so many times it basically told me " go search for yourself"
I very much agree with your conclusions and general approach.
LLMs are great for certain tasks that are programming related and it does it very well. I, too, often find myself needing scripts that as long as they did what they were suppose to, I really didn’t care how.
Another thing I’ve noticed(which is probably related to amounts of training data) is that it can help better with simple Python tasks as opposed to how it handles simple rust tasks.
But you mentioned one of my main issues with. Ice been programming for 15 years or so, and still learning. All the available llms did crucial errors about fundamental tabd complex topics and got the answer so very wrong but also sounding very convincing. Couple it with lack of proper linking to the sources of the response, you might see why having it explain code might cause your learn wrongly. Although it is also possible to say this about randoms internet tutorials. I always try to remind myself that it’s a tool that produces output that always needs to be verified.