6 ms·

I've seen it already. A small business owner (one man show) friend of mine with zero developer experience was able to solve his problem (very custom business sp

I've seen it already. A small business owner (one man show) friend of mine with zero developer experience was able to solve his problem (very custom business specific data -> calendar management) in a rough way using ChatGPT. But it got past about 300 lines long and really started to get bad. He'd put dozens of hours of time on his weekend to getting it to where it was by using ChatGPT over and over, but eventually it stopped being able to make the highly specific changes he needed as he used it more and more. He came to me for some help and I was able to work through it a bit as a friend, but the code quality was bad and I had to say "to really do this, I'd need to consult, and there's probably a better person to hire for that." He's muddling along but is looking for low cost devs to contract with on it now that he's getting value out of it though. And I suspect that kind of story will continue quite a bit as the tech matures.

imiric 2y ago
> And I suspect that kind of story will continue quite a bit as the tech matures. Don't you think that this tech can only get better? And that there will come a time in the very near future when the programming capabilities of AI improve substantially over what they are now? After all, AI writing 300 line programs was unheard of a mere 2 years ago. This is what I think GP is ignoring. Spreadsheets couldn't to do every task an accountant can do, so they augmented their capabilities. Compilers don't have the capability to write code from scratch, and Python doesn't write itself either. But AI will continually improve, and spread to more areas that software engineers were trained on. At first this will seem empowering, as they will aid us in writing small chunks of code, or code that can be easily generated like tests, which they already do. Then this will expand to writing even more code, improving their accuracy, debugging, refactoring, reasoning, and in general, being a better programming assistant for business owners like your friend than any human would. The concerning thing is that this isn't happening on timescales of decades, but years and months. Unlike GP, I don't think software engineers will exist as they do today in a decade or two. Everyone will either need to be a machine learning engineer and directly work with training and tweaking the work of AI, and then, once AI can improve itself, it will become self-sufficient, and humans will only program as a hobby. Humans will likely be forbidden from writing mission critical software in health, government and transport industries. Hardware engineers might be safe for a while after that, but not for long either.
- srcreigh 2y ago
  AI can’t write its own prompts. 10k people using the same prompt who actually need 5000 different things. No improvements to AI will let it read vague speakers’ minds. No improvement to AI will let it get answers it needs if people don’t know how to answer the necessary questions. Information has to come from somewhere to differentiate 1 prompt into 5000 different responses. If it’s not coming from the people using the AI, where else can it possibly come from? If people using the tool don’t know how to be specific enough to get what they want, the tool won’t replace people. s/the tool/spreadsheets s/the tool/databases s/the tool/React s/the tool/low code s/the tool/LLMs
  - imiric 2y ago
    > AI can’t write its own prompts. What makes you say that? One model can write the prompts of another, and we have seen approaches combining multiple models, and models that can evaluate the result of a prompt and retry with a different one. > No improvements to AI will let it read vague speakers’ minds. No improvement to AI will let it get answers it needs if people don’t know how to answer the necessary questions. No, but it can certainly produce output until the human decides it's acceptable. Humans don't need to give precise guidance, or answer technical questions. They just need to judge the output. I do agree that humans currently still need to be in the loop as a primary data source, and validators of the output. But there's no theoretical reason AI, or a combination of AIs, couldn't do this in the future. Especially once we move from text as the primary I/O mechanism.
  - heavyset_go 2y ago
    I agree with your point, just want to point out that models have been trained on AI generated prompts as synthetic data.
- layer8 2y ago
  > But AI will continually improve There is a bit of a fallacy in here. We don’t know how far it will improve, and in what ways. Progress isn’t continuous and linear, it comes more in sudden jumps and phases, and often plateaus for quite a while.
homarp 2y ago
> as the tech matures. Then it means you can use the matured tech and build in one day a superb service. And improve it the next day.
rurp 2y ago
This makes sense to me. When updating anything beyond a small project, keeping things reliable and mantainable for the future is 10x more important than just solving the immediate bug or feature. Short term hacks add up and become overwhelming at some point, even if each individual change seems manageable at the time. I used to work as a solo contractor on small/early projects. The most common job opportunity I encountered was someone who had hired the cheapest offshore devs they could find, seen good early progress with demos and POCs, but over time things kept slowing down and eventually went off the rails. The codebases were invariably a mess of hacks and spaghetti code. I think the best historical comp to LLMs is offshore outsourcing, except without the side effect of lifting millions of people out of poverty in the third world.
faeriechangling 2y ago
A large amount of the problems in the world don't require computer programs over a few hundred lines long to solve, so LLMs will still see use by DIY types. People may underestimate how difficult it is for an LLM to write a long or complex computer program though. It makes sense LLMs do very well at pumping out boilerplate and leetcode answers or trivial programs, but it doesn't nessecarily track that it would be they would that good at writing complex sophisticated and unique custom software. It may in fact be much further away from doing that than a lot of people anticipate, in a self-driving is just around the corner kind of way.

imiric 2y ago

> And I suspect that kind of story will continue quite a bit as the tech matures. Don't you think that this tech can only get better? And that there will come a time in the very near future when the programming capabilities of AI improve substantially over what they are now? After all, AI writing 300 line programs was unheard of a mere 2 years ago. This is what I think GP is ignoring. Spreadsheets couldn't to do every task an accountant can do, so they augmented their capabilities. Compilers don't have the capability to write code from scratch, and Python doesn't write itself either. But AI will continually improve, and spread to more areas that software engineers were trained on. At first this will seem empowering, as they will aid us in writing small chunks of code, or code that can be easily generated like tests, which they already do. Then this will expand to writing even more code, improving their accuracy, debugging, refactoring, reasoning, and in general, being a better programming assistant for business owners like your friend than any human would. The concerning thing is that this isn't happening on timescales of decades, but years and months. Unlike GP, I don't think software engineers will exist as they do today in a decade or two. Everyone will either need to be a machine learning engineer and directly work with training and tweaking the work of AI, and then, once AI can improve itself, it will become self-sufficient, and humans will only program as a hobby. Humans will likely be forbidden from writing mission critical software in health, government and transport industries. Hardware engineers might be safe for a while after that, but not for long either.

srcreigh 2y ago
AI can’t write its own prompts. 10k people using the same prompt who actually need 5000 different things. No improvements to AI will let it read vague speakers’ minds. No improvement to AI will let it get answers it needs if people don’t know how to answer the necessary questions. Information has to come from somewhere to differentiate 1 prompt into 5000 different responses. If it’s not coming from the people using the AI, where else can it possibly come from? If people using the tool don’t know how to be specific enough to get what they want, the tool won’t replace people. s/the tool/spreadsheets s/the tool/databases s/the tool/React s/the tool/low code s/the tool/LLMs
- imiric 2y ago
  > AI can’t write its own prompts. What makes you say that? One model can write the prompts of another, and we have seen approaches combining multiple models, and models that can evaluate the result of a prompt and retry with a different one. > No improvements to AI will let it read vague speakers’ minds. No improvement to AI will let it get answers it needs if people don’t know how to answer the necessary questions. No, but it can certainly produce output until the human decides it's acceptable. Humans don't need to give precise guidance, or answer technical questions. They just need to judge the output. I do agree that humans currently still need to be in the loop as a primary data source, and validators of the output. But there's no theoretical reason AI, or a combination of AIs, couldn't do this in the future. Especially once we move from text as the primary I/O mechanism.
- heavyset_go 2y ago
  I agree with your point, just want to point out that models have been trained on AI generated prompts as synthetic data.
layer8 2y ago
> But AI will continually improve There is a bit of a fallacy in here. We don’t know how far it will improve, and in what ways. Progress isn’t continuous and linear, it comes more in sudden jumps and phases, and often plateaus for quite a while.

homarp 2y ago

> as the tech matures. Then it means you can use the matured tech and build in one day a superb service. And improve it the next day.

rurp 2y ago

This makes sense to me. When updating anything beyond a small project, keeping things reliable and mantainable for the future is 10x more important than just solving the immediate bug or feature. Short term hacks add up and become overwhelming at some point, even if each individual change seems manageable at the time. I used to work as a solo contractor on small/early projects. The most common job opportunity I encountered was someone who had hired the cheapest offshore devs they could find, seen good early progress with demos and POCs, but over time things kept slowing down and eventually went off the rails. The codebases were invariably a mess of hacks and spaghetti code. I think the best historical comp to LLMs is offshore outsourcing, except without the side effect of lifting millions of people out of poverty in the third world.

faeriechangling 2y ago

A large amount of the problems in the world don't require computer programs over a few hundred lines long to solve, so LLMs will still see use by DIY types. People may underestimate how difficult it is for an LLM to write a long or complex computer program though. It makes sense LLMs do very well at pumping out boilerplate and leetcode answers or trivial programs, but it doesn't nessecarily track that it would be they would that good at writing complex sophisticated and unique custom software. It may in fact be much further away from doing that than a lot of people anticipate, in a self-driving is just around the corner kind of way.