Got this thing which is puzzling me, and I feel like the Fediverse is just the place to find people who could take an educated guess :-)
The other day, I heard on the grapevine of someone saying that at their company, the LLM code is "better" and, specifically, "easier to read" than the code they were previously getting from humans.
"Easier to read" surprised me, because up till now my impression is/was that even if LLM-emanated stuff can be produced _faster_ than human-made code, it typically wouldn't be especially logical or elegant - if you even want to look under the bonnet at all.
I don't personally know the person who shared this opinion, so I'm not sure what's their vantage point or techieness level. But from the grapevine chain, my impression is they're at the least a reliable reporter of their own experience.
So now I am curious!
What would you guess this person was seeing, which gave them this impression?
E.g. are the LLMs actually not as bad on this particular front as I've got the impression of?
Has there been a recent breakthrough improvement in the readability of LLM code? (It's plausible to imagine that this company would have access to the latest "models" etc.)
Or... I don't have any particular reason to think that the human coders in the comparison would be unusually bad and thereby making the LLM look better, but does it sound plausible that perhaps there could be a tradition of half-hearted or absent _commenting_ at their place, so that a LLM could outdo it?
Or... what do people think?
#AskFedi #LLM #LLMCoding
