5 ms·

Chatgpt is usefull today. I had a sql query where each developer folded a new layer around it, without touching the center. So SELECT fieldnames FROM (SELEC

Chatgpt is usefull today. I had a sql query where each developer folded a new layer around it, without touching the center. So SELECT fieldnames FROM (SELECT otherfieldnames FROM ...) With splitting and merging the dataset and a few views, it went 7 layers deep. I asked chatgpt 4 to tell me which fields influenced which, and it peeled the onion with no problem. It even wrote a cleaned up query. It couldn't stop itself to give me a quick SQL refresher course, though. Even so, I am impressed and will experiment more with the cleanup of messy code. Other people are finding it usefull, too, even non IT people. I can't see this as doomed tech

CamperBob2 3y ago
It's incredible at programming, and not just at hacking SQL. It won't be long before we think of "Source code" as "Carefully-written English specifications that are compiled to an actual programming language by an LLM." We'll think of C++ the way we think of assembly now ("Climb the mountain and seek out the guru"), and we'll think of assembly the way we think of HDL now ("Climb the rest of the way up the mountain and spend the next decade or two training in the secret dojo.") Both high- and low-level programming languages will be specialties for a rarefied few.
- xk_id 3y ago
  You forgot to mention one single thing that your “predictions” rely on. One. Didn’t stop you from asserting them with 100% assurance tho.
  - CamperBob2 3y ago
    Personal experience writing nicely-functioning code in a language I barely know, with libraries I've never used before. Your mileage may vary, of course. Failing to take advantage of the best tools available isn't particularly adaptive, but you do you.
    - xk_id 3y ago
      This does not follow: > Personal experience writing nicely-functioning code in a language I barely know, with libraries I've never used before. > We'll think of C++ the way we think of assembly now. I encourage you to reevaluate your understanding of computer science.
      - CamperBob2 3y ago
        I encourage you to reevaluate your understanding of computer science. I have. I have concluded that anybody who spent the last several years studying the way things work now has wasted a lot of time (and likely a fair bit of money), unless their goal is to help develop the next generation of tools. Comp sci itself is relatively timeless, but the day-to-day practice of programming will be radically different in 5 years and unrecognizable in 10. And it's about freaking time.
        marak830 3y ago
        I'm 50/50 on the fence here. I agree with most of your assertions, but I think knowledge of the base level will always be useful. I do quite commonly use gpt4 to write my boiler plate, my knowledge helps me redirect it when it's going down the wrong path (or flat out hallucinating). Will it get better? Of course! But being able to understand enough to be able to modify it yourself won't disappear. Just become less useful. Apologies if I misunderstood your assertion, it's been a long week.
        CamperBob2 3y ago
        No, that's basically where I am as well -- GPT4 writes boilerplate and (most importantly) answers questions. It is not competitive with a CS education, nor is it trying to be. But it will eventually replace the people we currently think of as "programmers," which is what the vast majority of CS students end up doing for a living. Unless the IP courts, a nuclear attack, or some neo-Luddite terrorists stop it. :-P GPT4 is certainly error-prone, but rapidly growing less so. Over the past couple of years, it has gone from a rubber-duck debugging partner to a capable (if over-imaginative) junior developer. What I'm starting to see now is that it is moving from "junior developer" to genuine guruhood. Notice how it hallucinates exactly nothing in the session at https://chat.openai.com/share/345d0bca-7fc3-4b91-a1f1-95b1e8e22c01 https://chat.openai.com/share/345d0bca-7fc3-4b91-a1f1-95b1e8... , even when I was being a dumbass and arguing with it. It no longer falls over itself apologizing for its 'error', but stands its ground when it is correct. The GP's comments happened to come at the same time I was engaging GPT4 in that session. I couldn't help but think that this is exactly what it was like to argue with an OG Luddite, or with someone who insisted that horses were never going to give way to automobiles or buses or trains or bikes or anything else.
xk_id 3y ago
Completely irrelevant if it will hallucinate the next time you use it. A broken clock is useful twice a day too, but it’s also worthless.
- CamperBob2 3y ago
  You need to spend some time actually using a current-generation GPT. You have absolutely no idea what you're talking about.
- hyperman1 3y ago
  I think this is too harsh. It is technology with good uses but also limitations, just like all other technology. I'd say use it, but don't trust it fully. Hallucination seems a consequence of it being incapable to say: I don't know. Messy code is a problem with good enough answers to make it less of a risk to go completely of the rails. Also, weirdly, you can ask it to look critically at its own answer, and give edge cases. But at the end of the day, I am responsible for the SQL, not ChatGPT. I have to look over the response, test it,... A non-trivial amount of time. ChatGPT is a great tool but not magic. See also the now infamous case of the lawyer with the hallucinated precedents. It is a good idea for a lawyer to let ChatGPT do some research. But what went wrong was the validation: Someone should have looked up the cited case law, and noticed they were hallucinations. This is misuse of a good but not perfect tool.
- MattPalmer1086 3y ago
  This only makes sense if you are thinking of these tools as replacements for people. If you see them as a force multiplier, something that makes many tasks much more efficient, then, it is not irrelevant. My own experience in using them like the author to get fully working code with libraries I was not familiar with demonstrates this to my satisfaction. It didn't get everything right on the first try, but I managed to try out 4 different libraries, with gpt rewriting the code for each to get more or less the same results in less than a day. Would have taken me at least 2 or 3 days without it.
  - xk_id 3y ago
    > If you see them as a force multiplier Gambling on a random generator because you don’t want to spend time improving your skills is not a “force multiplier”.

CamperBob2 3y ago

It's incredible at programming, and not just at hacking SQL. It won't be long before we think of "Source code" as "Carefully-written English specifications that are compiled to an actual programming language by an LLM." We'll think of C++ the way we think of assembly now ("Climb the mountain and seek out the guru"), and we'll think of assembly the way we think of HDL now ("Climb the rest of the way up the mountain and spend the next decade or two training in the secret dojo.") Both high- and low-level programming languages will be specialties for a rarefied few.

xk_id 3y ago
You forgot to mention one single thing that your “predictions” rely on. One. Didn’t stop you from asserting them with 100% assurance tho.
- CamperBob2 3y ago
  Personal experience writing nicely-functioning code in a language I barely know, with libraries I've never used before. Your mileage may vary, of course. Failing to take advantage of the best tools available isn't particularly adaptive, but you do you.
  - xk_id 3y ago
    This does not follow: > Personal experience writing nicely-functioning code in a language I barely know, with libraries I've never used before. > We'll think of C++ the way we think of assembly now. I encourage you to reevaluate your understanding of computer science.
    - CamperBob2 3y ago
      I encourage you to reevaluate your understanding of computer science. I have. I have concluded that anybody who spent the last several years studying the way things work now has wasted a lot of time (and likely a fair bit of money), unless their goal is to help develop the next generation of tools. Comp sci itself is relatively timeless, but the day-to-day practice of programming will be radically different in 5 years and unrecognizable in 10. And it's about freaking time.
      - marak830 3y ago
        I'm 50/50 on the fence here. I agree with most of your assertions, but I think knowledge of the base level will always be useful. I do quite commonly use gpt4 to write my boiler plate, my knowledge helps me redirect it when it's going down the wrong path (or flat out hallucinating). Will it get better? Of course! But being able to understand enough to be able to modify it yourself won't disappear. Just become less useful. Apologies if I misunderstood your assertion, it's been a long week.
        CamperBob2 3y ago
        No, that's basically where I am as well -- GPT4 writes boilerplate and (most importantly) answers questions. It is not competitive with a CS education, nor is it trying to be. But it will eventually replace the people we currently think of as "programmers," which is what the vast majority of CS students end up doing for a living. Unless the IP courts, a nuclear attack, or some neo-Luddite terrorists stop it. :-P GPT4 is certainly error-prone, but rapidly growing less so. Over the past couple of years, it has gone from a rubber-duck debugging partner to a capable (if over-imaginative) junior developer. What I'm starting to see now is that it is moving from "junior developer" to genuine guruhood. Notice how it hallucinates exactly nothing in the session at https://chat.openai.com/share/345d0bca-7fc3-4b91-a1f1-95b1e8e22c01 https://chat.openai.com/share/345d0bca-7fc3-4b91-a1f1-95b1e8... , even when I was being a dumbass and arguing with it. It no longer falls over itself apologizing for its 'error', but stands its ground when it is correct. The GP's comments happened to come at the same time I was engaging GPT4 in that session. I couldn't help but think that this is exactly what it was like to argue with an OG Luddite, or with someone who insisted that horses were never going to give way to automobiles or buses or trains or bikes or anything else.

xk_id 3y ago

Completely irrelevant if it will hallucinate the next time you use it. A broken clock is useful twice a day too, but it’s also worthless.

CamperBob2 3y ago
You need to spend some time actually using a current-generation GPT. You have absolutely no idea what you're talking about.
hyperman1 3y ago
I think this is too harsh. It is technology with good uses but also limitations, just like all other technology. I'd say use it, but don't trust it fully. Hallucination seems a consequence of it being incapable to say: I don't know. Messy code is a problem with good enough answers to make it less of a risk to go completely of the rails. Also, weirdly, you can ask it to look critically at its own answer, and give edge cases. But at the end of the day, I am responsible for the SQL, not ChatGPT. I have to look over the response, test it,... A non-trivial amount of time. ChatGPT is a great tool but not magic. See also the now infamous case of the lawyer with the hallucinated precedents. It is a good idea for a lawyer to let ChatGPT do some research. But what went wrong was the validation: Someone should have looked up the cited case law, and noticed they were hallucinations. This is misuse of a good but not perfect tool.
MattPalmer1086 3y ago
This only makes sense if you are thinking of these tools as replacements for people. If you see them as a force multiplier, something that makes many tasks much more efficient, then, it is not irrelevant. My own experience in using them like the author to get fully working code with libraries I was not familiar with demonstrates this to my satisfaction. It didn't get everything right on the first try, but I managed to try out 4 different libraries, with gpt rewriting the code for each to get more or less the same results in less than a day. Would have taken me at least 2 or 3 days without it.
- xk_id 3y ago
  > If you see them as a force multiplier Gambling on a random generator because you don’t want to spend time improving your skills is not a “force multiplier”.