213
AI Software Development – What Does The Data Say? (codemanship.wordpress.com)
submitted 2 weeks ago* (last edited 2 weeks ago) by to c/programming@programming.dev
67 comments fedilink hide all child comments
 

studies show a clear trend – output is up (more code, more commits, bigger diffs), but outcomes don’t reflect that trend. If anything, the average team is taking longer to ship worse software

you are viewing a single comment's thread
view the rest of the comments
[–] 1 point 1 week ago (1 child)

There’s a lot of levels at which I could reply to this comment, because it’s quite a philosophical topic, but let me clarify what I meant at the end of my last comment, before writing a wall of text here :)

As I said in my previous comment, faith is largely a means to make sense of our subjective experiences. We can refine our faith by talking about our experiences with other people.

So the most basic fallacy I was alluding to, from people who believe you can refine your faith by talking with an LLM is this: An LLM has no subjective experiences, because it doesn’t experience anything at all. This is one way how using the output from an LLM as a source of inspiration for your faith sets you up for psychosis. You start believing things that have no basis in reality or anyone’s genuine experiences. Rather, you may start believing in machine-generated approximations from which coherence has been lost.

But it seems your understanding was slightly different, in that people might develop their own sense of faith from experiences with an LLM itself rather than through the exchange of presumed experiences with the LLM. Do I understand correctly that you think some people attribute godlike qualities to the LLM, maybe through the presumption that it is an AGI? That would be a bit of a different discussion indeed :)

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 1 point 1 week ago (1 child)

    So the most basic fallacy I was alluding to, from people who believe you can refine your faith by talking with an LLM is this: An LLM has no subjective experiences, because it doesn’t experience anything at all. This is one way how using the output from an LLM as a source of inspiration for your faith sets you up for psychosis. You start believing things that have no basis in reality or anyone’s genuine experiences. Rather, you may start believing in machine-generated approximations from which coherence has been lost.

    Reasonable, though it could be argued they are full of subjective experiences, just all mixed together.

    But i expect it's probably more about the coherence of the experience that provides the meaning in this context.

    Coherence as in , one coherent explanation of a subjective experience by a person rather than a plausible sounding experience made up of many bits of experiences from a corpus of data.

    But it seems your understanding was slightly different, in that people might develop their own sense of faith from experiences with an LLM itself rather than through the exchange of presumed experiences with the LLM. Do I understand correctly that you think some people attribute godlike qualities to the LLM, maybe through the presumption that it is an AGI? That would be a bit of a different discussion indeed :)

    Not exactly, i think there are two different things here from my side.

    The first:

    I think people might attribute person like qualities to an LLM, the "ghost in the machine" as was mentioned earlier.

    I wasn't trying to imply someone who knows what an AGI is supposed to be would use that knowledge to assign godlike attributes to an LLM (a "god in the machine").

    Though , as a concept that would also be a valid example of "belief in the absence of proof".

    I don't expect most people to know the terminology or have even an idea of what's happening behind the scenes.

    The second:

    The part i'm struggling with is why "faith" in something like ability for meaning to come from LLM generated text (god-like or person-like) is any less acceptable than "faith" derived from perceived meaning in seemingly random (based on the knowledge of the time) events, like thunderstorms or plagues.

    They would both cover the idea of belief in something in the absence of proof, so how would someone decide one is less valid than the other.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 1 point 1 week ago

    Right, gotcha. So, a few things here:

    Reasonable, though it could be argued they are full of subjective experiences, just all mixed together.

    This argument wouldn’t really make sense to me, because an experience and the words describing said experience are clearly not the same. An LLM only possesses the later, but the actual experience is something conscious that an LLM does not have.

    Words are approximations, and often poor ones at that. For us humans, we can use them to describe experiences so that hopefully they bring understanding in the recipient, but whether the recipient truly understands and whether they can relate to the experience, also depends on their own experiences and their empathic abilities. None of that is at work inside an LLM.

    But indeed, thinking that an LLM does have such abilities would be an example of attributing person-like qualities to it.

    Now, if a person does develop a sense of faith from LLM output, is that more or less valid than them developing a sense of faith through a thunderstorm? That depends on the faith, I guess.

    If someone truly believes the LLM to be person-like (or godlike) that would be a nonsensical faith, as much as someone believing that firestorms are the wrath of god, for instance.

    If someone believes that an LLM-generated text provided them with a critical insight that turned their life around, and therefore LLMs are divinely inspired, it’s as harmless as someone believing that a thunderbolt that struck nearby gave them the necessary wakeup call that turned their life around, and therefore thunderstorms are guided by the divine.

    Of course in either case, if they start to revere LLMs/thunderstorms above other aspects of reality, it becomes problematic.

    Personally, as a more spiritually-inclined person, I think we should be able to say that both thunderstorms and LLMs are awe-inspiring, and I think both can, in theory, be a vessel for the divine. But ultimately, neither are divine in and of themselves.

    So isn’t there a difference between faith based on LLMs vs. faith based on firestorms? I think it really depends on the case, but there’s still an important distinction: A firestorm doesn’t pretend to be human, or doesn’t try to trick people into pretending it has humanlike qualities or experiences. So generally speaking, people tend not to develop weird faiths based on thunderstorms. But LLMs are, by design, imposters pretending to have humanlike qualities. This is why I think we tend to be dismissive of people believing in things told by LLMs, because it’s so easy to fall into the idea of ascribing real humanlike qualities to them and developing psychotic thoughts from there.

  • permalink
  • fedilink
  • source
  • parent