Comment by qsera
7 hours ago
>just as top notch quality as they produced before, just that now they are much faster.
No, they are not. There write programs like they write prose. There is 100% adherence to grammar and spelling when they write prose, but yet what they write is unintelligible and needlessly verbose.
When they write computer programs there is 100% adherence to good practices, but eventually the program becomes so unintelligible and verbose that only an LLM can make any sense of it going forward. But even then, changes result in more things breaking than they fix. But the number of bugs slashed goes through the roof. Management is happy!
>Of course you can misuse it.
I think the proper way to use it is to use it as a better search tool. But I don't think AI marketing and valuation is going to be sated by such a use case.
> > just as top notch quality as they produced before, just that now they are much faster.
> No, they are not.
It fascinates me with what conviction you are writing this. You don't know any of the people involved nor how they work or the code they produce. Yet you are certain that what I wrote is false.
I'm speculating now, but you are likely imagining that the LLM-produced code is used as-is, producing a growing slop pile. The cases I'm describing are not like that. Those folks know exactly what they want, how the architecture and the code should look like, and they are getting the LLM to produce it like that. There are quite a few iterations and manual polishing steps too. In the end, the code looks like as if they had written it themselves, just that the whole process was 5x as fast because there is so much quite standard stuff automatically added which would just have been slower by hand.
If you have not experienced this yourself then I understand your attitude, but rest assured, this is possible and I'd caution you to dismiss this.