On Thu, 16 Jul 2026 at 18:51, Wolfgang Jeltsch <[email protected]>
wrote:

>
> Well, if the LLM assists the “slow” programmers, who cannot deal with
> large contexts, doesn’t this mean that those programmers will care
> *less* about building abstractions, because the incentive to build them,
> which you describe, is largely gone?


In my experience LLMs are *not* good at building abstractions or even
reusing code, they can lead to code replication and code bloat.  And yes
the incentive structure to maintain discipline is not favorable. And yes it
will affect everyone, not just a particular category of programmers. They
may not feel the need to write better code because LLMs will take care of
understanding the code, they do not have to.  But this does not mean that
we must not use LLMs altogether. A responsible use can make a good
programmer very productive. You can take care of the big picture and
abstractions while LLMs can potentially help you fill in the local details
faster which you can review quickly. I think the entire discussion is
around the premise that most people will not be responsible, if that
problem arises and becomes a big burden then sure it warrants revisiting
the solution.


> Or do you mean to say that the LLM
> will not only maintain the context but also build the abstractions? If
> yes, is abstraction building really something that an LLM is useful for,
> and does it make sense for people who used to be strong in abstraction
> building to delegate this to a machine?


I was trying to say the opposite. You take care of the discipline and
abstractions, LLMs will take care of the context.

-harendra
_______________________________________________
ghc-devs mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to