I had a similar take after my first experience using AI to help me code. I put it aside as a curiosity. But when I went back recently, it's not that it's perfect, but the improvement in that time was massive. Does that mean it will continue to improve at that pace? Not necessarily, but we haven't seen the end state yet, so anything we say is just a judgment on what we have at the moment.
But do you use it now to help you code and if yes, how? The negative effects of relying to heavily on AI while coding are greatly discussed, hence I am wondering what a „good“ use case would be.
> The negative effects of relying to heavily on AI while coding are greatly discussed, hence I am wondering what a „good“ use case would be.
Really depends on your perspective. For some executives, a "good" use case may be the equivalent of burning goodwill to generate cash: push devs to use AI extensively on everything, realize a short term productivity bump while their skills atrophy (by haven't atrophied yet), then let the next guy deal with the problem of devs that have fully realized the "negative effects of relying to heavily on AI."
That’s a pretty dark perspective but it would imply that those executives are some kind of evil geniuses that grasp the extent of this situation. I personally try to count this kind of behavior on the statistics one of the ignobels present: 80% of asked uni professors felt they’re above the average (iq wise).
I haven't used it directly on anything except little test projects. But my general view is that it's like being an editor as opposed to a writer. I have to have mastered the craft of writing to edit someone else's copy.
I couldn’t agree more, thanks for answering! Anecdotally I’ve witnessed people using and talking big about ML/ LLM‘s while being in shock when learning about the fact that there are fundamentally basic statistical concepts behind those.
Not OP, but I specifically like to use AI to explain obtuse sections of code that would take me longer periods of time to understand by reading.
If I have a bug reported and I’m not sure where it is, pasting the bug report into an LLM and asking it to find the bug has yielded some mixed results but ultimately saved me time.
Interestingly enough, I also was wondering if I could improve my efficiency by condensing written text. The idea would be to remove the usual padding or „slop“ you have within most of the modern web environment.
Wouldn’t you loose a bit of that brain power if you stop to make those connections yourself while trying to understand those code sections?
