4 ms·

> One thing sad to me now, with the focus on GPT or Diffusion prompts, is how much kids are no longer thinking about the actual systems or logical rules they're

by emrah 4y ago

> One thing sad to me now, with the focus on GPT or Diffusion prompts, is how much kids are no longer thinking about the actual systems or logical rules they're playing with, and just learning to trick something that's already inscrutable and smarter than them This is something I don't understand. Sure, not every kid is going to think about that but that's ok. We need some kids to think about it, probably a small subset.. Back when computers were new, not every kid tried to make games or understand how they worked either. Only a small subset and that's how it is..

q7xvh97o2pDhNrh 4y ago
The way I read OP's comment was more about quality rather than quantity. Put another way, I'd say that the right amount of people who should be learning that skill is probably zero. "Learning to write AI prompts" is a skill that only exists because AI hasn't evolved enough yet; in another 10-25 years, "learning to write AI prompts" will just be equivalent to "communicating clearly." And, yes, thinking and communicating clearly are valuable skills that everyone should learn, but that's just the point — people investing time in the "writing AI prompts" skill now aren't learning that. They're just learning to string random adjectives together, along with (at best) a slight bit of the iterative never-give-up spirit that comes with poking an inscrutable pile of matrices until it does what you want. [1] [1]: https://xkcd.com/1838/ https://xkcd.com/1838/
- jvalencia 4y ago
  I think the lament is that if people do not learn the logical side of things, they are unable to figure out how to reason about issues and errors. They are unable to easily go deep into topics like computer science.
  - serf 4y ago
    Do you think that the lament is wrong? I feel as if it stands to reason that someone without practice in common topics should do worse at a subject when encountered professionally or academically; the base mathematical principles behind programming like lambda calculus and the principles of recursion and boolean logic haven't changed, they've been masked. I feel like the use of many AI sub-systems really just trains someone in the use of that sub-system rather than any of the underlying principles. This feels obvious to me, am I missing something? Also : I don't think that specific 'AI sub-system' training is useless, that's the way our world is heading.. I just also happen to believe that it hides trees.
    - Espressosaurus 4y ago
      It feels like the app-ification programming. It's more accessible but leaves the users less able to deal with things it can't solve and completely incompetent to debug them. Hopefully I'm wrong. Also hopefully it doesn't put me out of a job in 5-10 years.
    - BirAdam 4y ago
      This is normal. The industry has many times more front-end web developers than full stack. It has many more full-stack developers than systems programmers. It has many more systems programmers than chip engineers. Each layer of abstraction makes things easier to learn (or at least more visible and therefore approachable) and therefore makes the number of people trained in any given discipline higher.
adamrezich 4y ago
but the same could be said for indie games developed with tools like Unity or Godot—people who make games with these tools don't really know 100% about what's going on with the actual underlying systems they're stitching together to make a game with, but they stitch things together and create a game nonetheless. on its face that's fine, but as time goes on, "making video games" has gone from being something that required intimate knowledge and understanding of the software and hardware at the foundation of the executable you're running, to domain-specific knowledge relative only to one specific tool/framework. if you've only ever made Unity games and now you want to work on a more complex simulation, you either find yourself increasingly having to fight "the Unity way of doing things", or take the plunge and unlearn your current way of thinking in domain-specific primitives such that you can learn to do things at a lower level.
- noduerme 4y ago
  I'd agree, with the caveat that I think it's okay to start at the top of the stack and work downwards, as long as the limitations of the tools keep you learning. I'm sure not every artist at Maxis could write Assembly code. There are more chances for visual artists now to step into the world of code than there were then. I started off as a graphic designer in an ad agency, learned the 1990s Dreamweaver way of writing web pages, and learned HTML/CSS and AS2/3 and JS and then PHP and Node and SQL as I hit various limitations with what I wanted to do. Now I write full stack business software. But I know people who went the other way, who learned the visual art and UI skills they wanted long after learning C++. So, unlearning your framework is a way of learning, too. The difference I see between interacting with a black-box AI versus any of those other things is that the AI isn't a constructor kit, it's a finished product that doesn't lead to furthering any skill set. It isn't like using Unity to build a city simulator; it's like playing the city simulator.
- lmm 4y ago
  Learning the minutiae of a particular piece of hardware isn't important. Learning to understand and debug a program's is very important. Unity or Godot still work on logic - if A causes B and C causes D but A and C doesn't cause B and D, there'll be a reason for that (even if that reason is just "a bug in the Unity implementation") and a programmer has a hope of learning what it is. Probabilistic tools don't have that, and I worry about the implications for learning. (For the same reason I've always been very skeptical about including tools like ElasticSearch or Splunk in a dataflow - even if they seem to work reliably, how can you be confident they will continue to do so in the future?)

by emrah 4y ago

q7xvh97o2pDhNrh 4y ago

The way I read OP's comment was more about quality rather than quantity. Put another way, I'd say that the right amount of people who should be learning that skill is probably zero. "Learning to write AI prompts" is a skill that only exists because AI hasn't evolved enough yet; in another 10-25 years, "learning to write AI prompts" will just be equivalent to "communicating clearly." And, yes, thinking and communicating clearly are valuable skills that everyone should learn, but that's just the point — people investing time in the "writing AI prompts" skill now aren't learning that. They're just learning to string random adjectives together, along with (at best) a slight bit of the iterative never-give-up spirit that comes with poking an inscrutable pile of matrices until it does what you want. [1] [1]: https://xkcd.com/1838/ https://xkcd.com/1838/

jvalencia 4y ago
I think the lament is that if people do not learn the logical side of things, they are unable to figure out how to reason about issues and errors. They are unable to easily go deep into topics like computer science.
- serf 4y ago
  Do you think that the lament is wrong? I feel as if it stands to reason that someone without practice in common topics should do worse at a subject when encountered professionally or academically; the base mathematical principles behind programming like lambda calculus and the principles of recursion and boolean logic haven't changed, they've been masked. I feel like the use of many AI sub-systems really just trains someone in the use of that sub-system rather than any of the underlying principles. This feels obvious to me, am I missing something? Also : I don't think that specific 'AI sub-system' training is useless, that's the way our world is heading.. I just also happen to believe that it hides trees.
  - Espressosaurus 4y ago
    It feels like the app-ification programming. It's more accessible but leaves the users less able to deal with things it can't solve and completely incompetent to debug them. Hopefully I'm wrong. Also hopefully it doesn't put me out of a job in 5-10 years.
  - BirAdam 4y ago
    This is normal. The industry has many times more front-end web developers than full stack. It has many more full-stack developers than systems programmers. It has many more systems programmers than chip engineers. Each layer of abstraction makes things easier to learn (or at least more visible and therefore approachable) and therefore makes the number of people trained in any given discipline higher.

adamrezich 4y ago

but the same could be said for indie games developed with tools like Unity or Godot—people who make games with these tools don't really know 100% about what's going on with the actual underlying systems they're stitching together to make a game with, but they stitch things together and create a game nonetheless. on its face that's fine, but as time goes on, "making video games" has gone from being something that required intimate knowledge and understanding of the software and hardware at the foundation of the executable you're running, to domain-specific knowledge relative only to one specific tool/framework. if you've only ever made Unity games and now you want to work on a more complex simulation, you either find yourself increasingly having to fight "the Unity way of doing things", or take the plunge and unlearn your current way of thinking in domain-specific primitives such that you can learn to do things at a lower level.

noduerme 4y ago
I'd agree, with the caveat that I think it's okay to start at the top of the stack and work downwards, as long as the limitations of the tools keep you learning. I'm sure not every artist at Maxis could write Assembly code. There are more chances for visual artists now to step into the world of code than there were then. I started off as a graphic designer in an ad agency, learned the 1990s Dreamweaver way of writing web pages, and learned HTML/CSS and AS2/3 and JS and then PHP and Node and SQL as I hit various limitations with what I wanted to do. Now I write full stack business software. But I know people who went the other way, who learned the visual art and UI skills they wanted long after learning C++. So, unlearning your framework is a way of learning, too. The difference I see between interacting with a black-box AI versus any of those other things is that the AI isn't a constructor kit, it's a finished product that doesn't lead to furthering any skill set. It isn't like using Unity to build a city simulator; it's like playing the city simulator.
lmm 4y ago
Learning the minutiae of a particular piece of hardware isn't important. Learning to understand and debug a program's is very important. Unity or Godot still work on logic - if A causes B and C causes D but A and C doesn't cause B and D, there'll be a reason for that (even if that reason is just "a bug in the Unity implementation") and a programmer has a hope of learning what it is. Probabilistic tools don't have that, and I worry about the implications for learning. (For the same reason I've always been very skeptical about including tools like ElasticSearch or Splunk in a dataflow - even if they seem to work reliably, how can you be confident they will continue to do so in the future?)