4 ms·

You learn a lot from ChatGPT though. If you want to know the ordinary answer to some essay question, cgpt will give you that. If you leave it at that and hand i

You learn a lot from ChatGPT though. If you want to know the ordinary answer to some essay question, cgpt will give you that. If you leave it at that and hand it in, maybe Chomsky is right. But it's also a essay writing productivity tool. You sketch out what you want mentioned and cgpt will fill in the gaps, like those apprentices that helped the renaissance masters. I keep coming back to the phrase "small intentions" when I discuss cgpt: both in essays and code, it is already quite good at filling in details. With less time spent on details, students can spend more time on the architecture. Ultimately, we'll have more people able to practice higher level creativity. This is already a thing that you do in math class; after your primary school years, you get a calculator to work out the arithmetic.

smt88 4y ago
The problem with using ChatGPT for "filling in details" is how often it is confidently, utterly wrong. My friend tried to use it to write a legal letter and it completely invented a state law and cited it very specifically. For him, it was annoying and amusing. For other people, mistakes like that could be very serious.
mrfox321 4y ago
Is the high level structure of writing any more important than the low level structure? Maybe it will tend towards everyone sounding the same. Which then gets amplified as LLMs train on writing that is essentially mutations of itself. Me no like.
- lordnacho 4y ago
  > Is the high level structure of writing any more important than the low level structure? Well you can make a point about the fall of Rome with badly cited sources or spelling mistakes. But if your essay says Rome fell due to conquest by the Han dynasty it can't be saved by any number of correctly spelled, beautifully cited details. ChatGPT will let you write an essay making any point while filling in little things.
thundergolfer 4y ago
It sounds like you're advocating a system of education, thinking, and work that is exemplified by 'Software Architecture Astronauts', people who have not recently or ever contended with the details, the nuts and bolts, of actual software development and maintenance. To be honest, the idea that you can produce valuable intellectual work by outsourcing the details is absurd. This would have seen me absolutely fail my challenging philosophy and engineering coursework, and end up a fundamentally incompetent dilettante in work and life.
- sitkack 4y ago
  We can do both, ChatGPT can code. So the Astronauts should be able to produce working systems that pass functional tests. The majority of the detractors of these models don't see the forest for the trees. It only gives new capabilities, they remove nothing.
  - thundergolfer 4y ago
    > It only gives new capabilities, they remove nothing. This is not how technology works in the world. See the arguments of for example McLuhan. Technologies are appendages for the human body or intellect, and at the same time they numb the part of that they amplify. Also the point of the Architecture Astronauts idea is that they don't produce working systems. It's a specific pejorative, it doesn't refer generally to people that work at a high-level without harming the systems they engage with.
    - sitkack 4y ago
      You clearly have a bone to pick with software architects, if anything it should reduce the amount of Space Travel for said architects. McLuhan, a media analysis astronaut. Average is the message. > Technologies are appendages for the human body or intellect We previously had fleets of people calculating tables of numbers for hours per day. Simple calculators removed that need and increased the speed of results by orders of magnitude. You seem to be arguing for handicap system, an intellectual conservatism. Is that correct? What does ChatGPT remove from ones ability to write low level code or for a software architect to both architect a system AND produce a working model?
meroes 4y ago
The model for mathematics is something like a “good book” of once-entered, unchanging posable questions and theorems. More may be added, but once added there is a sense of permanent definitiveness and relevance. For your example and beyond-mathematics calculator analogy, maybe the dates of historical events are equally definitive, but I don’t see much beyond that being so. Science goes through transformations. Either the questions remain the same (what is there, what exists), but the answers to them change as science develops. Or, the questions change, and the answers don’t (what did plogiston, aether, or general relativity predict). Your expanded “calculator” paradigm seems unfounded. It’s not analogous to math.
- sitkack 4y ago
  > Your expanded “calculator” paradigm seems unfounded. Please explain. GPT makes a utility out something that previously was not. The GP didn't say it was exact.
  - meroes 4y ago
    They didn't say exact. Exact isn't the concept I'm concerned about. There is something unique about mathematics in that its answers never change. And that static structure can be tapped into with calculators. A "calculator" outside this domain reeks of potential danger to me. Asking a question to some other field whose answers are open to revision doesn't give one transparent access to timeless knowledge. There is always going to be a caveat that the answers given by GPT are clouded by time and circumstance. You can't just copy the answers GPT gives you and call that knowledge, like calculator results do allow. I would hope GPT knowledge comes with the understanding that GPT v.XX, exposed to Y circumstances, produced Z results. We don't do that for calculators.
wpietri 4y ago
> after your primary school years, you get a calculator to work out the arithmetic Do we? I do a lot of mental arithmetic as a software developer. When I was an exchange student my host-dad was an engineer and he was incredibly good at it. As he was thinking through a problem he'd do all sorts of ballpark math to get a feel for things. Eventually, yes, the numbers would all get crunched properly. But even then he'd also be doing the math mentally to make sure that something hadn't gotten screwed up. A person who uses a calculator for everything starting from a young age just doesn't develop that skill. And I think there will be a similar problem with ChatGPT. Can an excellent writer get a performance boost starting with ChatGPT? The jury's out, but it's not impossible. But can one become an excellent writer by tarting up LLM-generated prose? I doubt it. Learning to write is a process of sweating the details. Especially so when the output is an essay. Real higher-level creativity requires foundational work. I think at best you'll get the sort of content that comes when an uninspiring politician hires a mediocre ghost writer. That said, a lot of writing is bullshit. And I mean that both in the Frankfurt sense ("speech intended to persuade without regard for truth") [1] and in the Graeber sense ("completely pointless, unnecessary, or pernicious") [2]. And that's where I think ChatGPT will shine. Marketing copy to fill in the "lorem ispum" space on the product page? ChatGPT to the rescue. Long sections of TPS reports that nobody will read? ChatGPT again. And once they expand ChatGPT to being able to make PowerPoint decks, a zillion ineffective middle managers will be able to return their focus to Candy Crush, harassing the interns, and creating pointless meetings. I think that's also bad, of course. Lowering the cost of bullshit production looks like a net societal negative to me. But it would be successful in the traditional tech industry metrics of "gains users" and "produces revenue". [1] https://en.wikipedia.org/wiki/On_Bullshit https://en.wikipedia.org/wiki/On_Bullshit [2] https://en.wikipedia.org/wiki/Bullshit_Jobs https://en.wikipedia.org/wiki/Bullshit_Jobs
- lordnacho 4y ago
  > A person who uses a calculator for everything starting from a young age just doesn't develop that skill. But is that a reasonable fear? It's actually quite hard to use a tool without understanding it, so you won't get a lot of people reliant on calculators who don't understand calculation. I've certainly never met someone for whom it was a problem. About using tools you don't understand, this is easy to try. Get some film editing software or 3D rendering software (assuming you don't know this area, like me) and see where you get to. I've never been able to use a tool without understanding at least something about what it does.

smt88 4y ago

The problem with using ChatGPT for "filling in details" is how often it is confidently, utterly wrong. My friend tried to use it to write a legal letter and it completely invented a state law and cited it very specifically. For him, it was annoying and amusing. For other people, mistakes like that could be very serious.

mrfox321 4y ago

Is the high level structure of writing any more important than the low level structure? Maybe it will tend towards everyone sounding the same. Which then gets amplified as LLMs train on writing that is essentially mutations of itself. Me no like.

lordnacho 4y ago
> Is the high level structure of writing any more important than the low level structure? Well you can make a point about the fall of Rome with badly cited sources or spelling mistakes. But if your essay says Rome fell due to conquest by the Han dynasty it can't be saved by any number of correctly spelled, beautifully cited details. ChatGPT will let you write an essay making any point while filling in little things.

thundergolfer 4y ago

It sounds like you're advocating a system of education, thinking, and work that is exemplified by 'Software Architecture Astronauts', people who have not recently or ever contended with the details, the nuts and bolts, of actual software development and maintenance. To be honest, the idea that you can produce valuable intellectual work by outsourcing the details is absurd. This would have seen me absolutely fail my challenging philosophy and engineering coursework, and end up a fundamentally incompetent dilettante in work and life.

sitkack 4y ago
We can do both, ChatGPT can code. So the Astronauts should be able to produce working systems that pass functional tests. The majority of the detractors of these models don't see the forest for the trees. It only gives new capabilities, they remove nothing.
- thundergolfer 4y ago
  > It only gives new capabilities, they remove nothing. This is not how technology works in the world. See the arguments of for example McLuhan. Technologies are appendages for the human body or intellect, and at the same time they numb the part of that they amplify. Also the point of the Architecture Astronauts idea is that they don't produce working systems. It's a specific pejorative, it doesn't refer generally to people that work at a high-level without harming the systems they engage with.
  - sitkack 4y ago
    You clearly have a bone to pick with software architects, if anything it should reduce the amount of Space Travel for said architects. McLuhan, a media analysis astronaut. Average is the message. > Technologies are appendages for the human body or intellect We previously had fleets of people calculating tables of numbers for hours per day. Simple calculators removed that need and increased the speed of results by orders of magnitude. You seem to be arguing for handicap system, an intellectual conservatism. Is that correct? What does ChatGPT remove from ones ability to write low level code or for a software architect to both architect a system AND produce a working model?

meroes 4y ago

The model for mathematics is something like a “good book” of once-entered, unchanging posable questions and theorems. More may be added, but once added there is a sense of permanent definitiveness and relevance. For your example and beyond-mathematics calculator analogy, maybe the dates of historical events are equally definitive, but I don’t see much beyond that being so. Science goes through transformations. Either the questions remain the same (what is there, what exists), but the answers to them change as science develops. Or, the questions change, and the answers don’t (what did plogiston, aether, or general relativity predict). Your expanded “calculator” paradigm seems unfounded. It’s not analogous to math.

sitkack 4y ago
> Your expanded “calculator” paradigm seems unfounded. Please explain. GPT makes a utility out something that previously was not. The GP didn't say it was exact.
- meroes 4y ago
  They didn't say exact. Exact isn't the concept I'm concerned about. There is something unique about mathematics in that its answers never change. And that static structure can be tapped into with calculators. A "calculator" outside this domain reeks of potential danger to me. Asking a question to some other field whose answers are open to revision doesn't give one transparent access to timeless knowledge. There is always going to be a caveat that the answers given by GPT are clouded by time and circumstance. You can't just copy the answers GPT gives you and call that knowledge, like calculator results do allow. I would hope GPT knowledge comes with the understanding that GPT v.XX, exposed to Y circumstances, produced Z results. We don't do that for calculators.

wpietri 4y ago

> after your primary school years, you get a calculator to work out the arithmetic Do we? I do a lot of mental arithmetic as a software developer. When I was an exchange student my host-dad was an engineer and he was incredibly good at it. As he was thinking through a problem he'd do all sorts of ballpark math to get a feel for things. Eventually, yes, the numbers would all get crunched properly. But even then he'd also be doing the math mentally to make sure that something hadn't gotten screwed up. A person who uses a calculator for everything starting from a young age just doesn't develop that skill. And I think there will be a similar problem with ChatGPT. Can an excellent writer get a performance boost starting with ChatGPT? The jury's out, but it's not impossible. But can one become an excellent writer by tarting up LLM-generated prose? I doubt it. Learning to write is a process of sweating the details. Especially so when the output is an essay. Real higher-level creativity requires foundational work. I think at best you'll get the sort of content that comes when an uninspiring politician hires a mediocre ghost writer. That said, a lot of writing is bullshit. And I mean that both in the Frankfurt sense ("speech intended to persuade without regard for truth") [1] and in the Graeber sense ("completely pointless, unnecessary, or pernicious") [2]. And that's where I think ChatGPT will shine. Marketing copy to fill in the "lorem ispum" space on the product page? ChatGPT to the rescue. Long sections of TPS reports that nobody will read? ChatGPT again. And once they expand ChatGPT to being able to make PowerPoint decks, a zillion ineffective middle managers will be able to return their focus to Candy Crush, harassing the interns, and creating pointless meetings. I think that's also bad, of course. Lowering the cost of bullshit production looks like a net societal negative to me. But it would be successful in the traditional tech industry metrics of "gains users" and "produces revenue". [1] https://en.wikipedia.org/wiki/On_Bullshit https://en.wikipedia.org/wiki/On_Bullshit [2] https://en.wikipedia.org/wiki/Bullshit_Jobs https://en.wikipedia.org/wiki/Bullshit_Jobs

lordnacho 4y ago
> A person who uses a calculator for everything starting from a young age just doesn't develop that skill. But is that a reasonable fear? It's actually quite hard to use a tool without understanding it, so you won't get a lot of people reliant on calculators who don't understand calculation. I've certainly never met someone for whom it was a problem. About using tools you don't understand, this is easy to try. Get some film editing software or 3D rendering software (assuming you don't know this area, like me) and see where you get to. I've never been able to use a tool without understanding at least something about what it does.