4 ms·

So are you bullish or bearish on the prospect of LLMs replacing programmers? The industry is breathless with hype and media pieces on how traditional programmin

by ryandv 1y ago

So are you bullish or bearish on the prospect of LLMs replacing programmers? The industry is breathless with hype and media pieces on how traditional programming is obsolete; do your criticisms of my position also apply, were software engineering, rather than calculus, its subject?

inetknght 1y ago
> are you bullish or bearish on "bullish" and "bearish" are words that are useless in this context. Use them for stock markets where those words belong. > LLMs replacing programmers? The industry is breathless with hype and media pieces on how traditional programming is obsolete I believe LLMs fill a small (over-hyped) niche currently, but that niche will grow over time. I don't believe LLMs will, within our lifetimes, fill the specialized roles of high intelligence. It certainly will (and in many cases already does) fill the roles that don't require lots of thinking and intelligence. Could LLMs replace programmers? Definitely, right now. Could LLMs replace developers? Currently, it could replace some of the worst developers, and help but not replace some of the better developers. Could it replace engineers? Definitely not right now, and probably not for the next several decades. Is that a bad thing? To a lot of people, it is, especially in short term measured by job losses. But in the long run I believe LLMs are transformative in nature. Programming and development will be transformed to use LLMs as the tools that they are. Anyone who doesn't adapt to using an LLM will end up being a loser in their industry or be required to retrain to another role. That's generally good for the economy as a whole even if it's not great for the individual person. You could manually do a lot of things that a CNC mill can do. Or you can learn how to use a CNC mill. Or you could pay someone else to use their CNC mill. But a CNC mill is not a great thing to use for cutting raw lumber or assembling furniture. You "could" use a CNC mill for it, but it'd be extremely inefficient. You could use it to manufacture furniter pieces, but it's not going to assemble them. LLMs are kind've the same thing for computers. > do your criticisms of my position also apply, were software engineering, rather than calculus, its subject? I think yes. Software Engineering (as an Engineering discipline) is used where it matters (eg, regulated aviation, safety, etc). It's not used much (if at all) for basic websites or even most games. But when it is used for these things, it's a much better product. That will pretty much always be true even with LLMs guiding the developer (not engineer). For your example I replied to: I don't know much about nuclear reactors. But I know more than, say, 80% of the general population. I know enough to know that they can be dangerous but some designs are less-dangerous than others. Could I build a reactor myself? Probably, while the average person wouldn't even know where to start. I wouldn't do it without a lot more knowledge though because anything I come up with would be miles behind current state of the art in safety and procedure. I probably could identify nuclear fuel ores, but I have no idea how to refine it safely. I don't really care to learn that, it's not my passion. But I also know that refining it is dangerous if not done "properly" and so I wouldn't even want it as a hobby. Contrast that with robotics. Robotics can certainly be dangerous, but not nearly as dangerous as refining nuclear fuel. I know how robotics work, and how to work with them with some modicum of safety, and so I do that as a hobby. That doesn't mean I would encourage anyone to come along and just use an LLM to build robots because they'd be missing out on sooooooo many fundamentally important concepts and would have to rely on the LLM to tell them what's dangerous or what's not and how to implement safety features and procedures.
- ryandv 1y ago
  > What do you think is going to replace calculus? Until something better comes along, calculus is among the best introduction to higher theories of mathematics. > Programming and development will be transformed to use LLMs as the tools that they are. Anyone who doesn't adapt to using an LLM will end up being a loser in their industry or be required to retrain to another role [...] Mathematics will be transformed to use LLMs as the tools that they are. Any mathematician who doesn't adapt to using an LLM will end up being a loser in academia or be required to retrain to another role. The study of calculus will be replaced by tooling that carries out the mechanics of differentiation, integration, finding extrema, etc, making such education obsolete. > You could manually do a lot of things that a CNC mill can do. Or you can learn how to use a CNC mill. Or you could pay someone else to use their CNC mill. But a CNC mill is not a great thing to use for cutting raw lumber or assembling furniture. You "could" use a CNC mill for it, but it'd be extremely inefficient. You could use it to manufacture furniter pieces, but it's not going to assemble them. LLMs are kind've the same thing for computers. You could manually do a lot of things that a calculator, CAS, or LLM can do. Or you can learn how to use a CAS system or LLM, or pay someone to use their cloud service. But an LLM is not a great thing to use for deciding research programs or assembling a proof. You could use an LLM for it, but it'd be extremely inefficient. You could use it to formalize certain lemmata or minor results, but it's not going to compose them together into a larger theorem. > Could LLMs replace programmers? Definitely, right now. Could LLMs replace developers? Currently, it could replace some of the worst developers, and help but not replace some of the better developers. Could it replace engineers? Definitely not right now, and probably not for the next several decades. Could LLMs replace arithmeticians or calculators (in the original sense of the word, as a human occupation)? Definitely, right now. Could it replace mathematicians? Definitely not right now, and probably not for the next several decades. You distinguish between Software Engineers(tm) and lowly developers; I point out that you can draw the same distinction between mathematicians and lowly arithmeticians or "calculators" (in the original sense of the term as a human occupation), or between calculus and analysis. Leave the solution of this particular integral or the differentiation of this particular function, the mechanics of calculus, to the LLMs and the machines. Mathematicians need not concern themselves with such trivialities, and will prove results for all continuous functions in arbitrary metric spaces, or something similarly general and insightful.

by ryandv 1y ago

inetknght 1y ago

> are you bullish or bearish on "bullish" and "bearish" are words that are useless in this context. Use them for stock markets where those words belong. > LLMs replacing programmers? The industry is breathless with hype and media pieces on how traditional programming is obsolete I believe LLMs fill a small (over-hyped) niche currently, but that niche will grow over time. I don't believe LLMs will, within our lifetimes, fill the specialized roles of high intelligence. It certainly will (and in many cases already does) fill the roles that don't require lots of thinking and intelligence. Could LLMs replace programmers? Definitely, right now. Could LLMs replace developers? Currently, it could replace some of the worst developers, and help but not replace some of the better developers. Could it replace engineers? Definitely not right now, and probably not for the next several decades. Is that a bad thing? To a lot of people, it is, especially in short term measured by job losses. But in the long run I believe LLMs are transformative in nature. Programming and development will be transformed to use LLMs as the tools that they are. Anyone who doesn't adapt to using an LLM will end up being a loser in their industry or be required to retrain to another role. That's generally good for the economy as a whole even if it's not great for the individual person. You could manually do a lot of things that a CNC mill can do. Or you can learn how to use a CNC mill. Or you could pay someone else to use their CNC mill. But a CNC mill is not a great thing to use for cutting raw lumber or assembling furniture. You "could" use a CNC mill for it, but it'd be extremely inefficient. You could use it to manufacture furniter pieces, but it's not going to assemble them. LLMs are kind've the same thing for computers. > do your criticisms of my position also apply, were software engineering, rather than calculus, its subject? I think yes. Software Engineering (as an Engineering discipline) is used where it matters (eg, regulated aviation, safety, etc). It's not used much (if at all) for basic websites or even most games. But when it is used for these things, it's a much better product. That will pretty much always be true even with LLMs guiding the developer (not engineer). For your example I replied to: I don't know much about nuclear reactors. But I know more than, say, 80% of the general population. I know enough to know that they can be dangerous but some designs are less-dangerous than others. Could I build a reactor myself? Probably, while the average person wouldn't even know where to start. I wouldn't do it without a lot more knowledge though because anything I come up with would be miles behind current state of the art in safety and procedure. I probably could identify nuclear fuel ores, but I have no idea how to refine it safely. I don't really care to learn that, it's not my passion. But I also know that refining it is dangerous if not done "properly" and so I wouldn't even want it as a hobby. Contrast that with robotics. Robotics can certainly be dangerous, but not nearly as dangerous as refining nuclear fuel. I know how robotics work, and how to work with them with some modicum of safety, and so I do that as a hobby. That doesn't mean I would encourage anyone to come along and just use an LLM to build robots because they'd be missing out on sooooooo many fundamentally important concepts and would have to rely on the LLM to tell them what's dangerous or what's not and how to implement safety features and procedures.

ryandv 1y ago
> What do you think is going to replace calculus? Until something better comes along, calculus is among the best introduction to higher theories of mathematics. > Programming and development will be transformed to use LLMs as the tools that they are. Anyone who doesn't adapt to using an LLM will end up being a loser in their industry or be required to retrain to another role [...] Mathematics will be transformed to use LLMs as the tools that they are. Any mathematician who doesn't adapt to using an LLM will end up being a loser in academia or be required to retrain to another role. The study of calculus will be replaced by tooling that carries out the mechanics of differentiation, integration, finding extrema, etc, making such education obsolete. > You could manually do a lot of things that a CNC mill can do. Or you can learn how to use a CNC mill. Or you could pay someone else to use their CNC mill. But a CNC mill is not a great thing to use for cutting raw lumber or assembling furniture. You "could" use a CNC mill for it, but it'd be extremely inefficient. You could use it to manufacture furniter pieces, but it's not going to assemble them. LLMs are kind've the same thing for computers. You could manually do a lot of things that a calculator, CAS, or LLM can do. Or you can learn how to use a CAS system or LLM, or pay someone to use their cloud service. But an LLM is not a great thing to use for deciding research programs or assembling a proof. You could use an LLM for it, but it'd be extremely inefficient. You could use it to formalize certain lemmata or minor results, but it's not going to compose them together into a larger theorem. > Could LLMs replace programmers? Definitely, right now. Could LLMs replace developers? Currently, it could replace some of the worst developers, and help but not replace some of the better developers. Could it replace engineers? Definitely not right now, and probably not for the next several decades. Could LLMs replace arithmeticians or calculators (in the original sense of the word, as a human occupation)? Definitely, right now. Could it replace mathematicians? Definitely not right now, and probably not for the next several decades. You distinguish between Software Engineers(tm) and lowly developers; I point out that you can draw the same distinction between mathematicians and lowly arithmeticians or "calculators" (in the original sense of the term as a human occupation), or between calculus and analysis. Leave the solution of this particular integral or the differentiation of this particular function, the mechanics of calculus, to the LLMs and the machines. Mathematicians need not concern themselves with such trivialities, and will prove results for all continuous functions in arbitrary metric spaces, or something similarly general and insightful.