3 ms·

I get the sneaking suspicion that you don't actually do this shit. ChatGPT isn't an expert coder sitting by your side. It's not even remotely close. Go ahead an

I get the sneaking suspicion that you don't actually do this shit. ChatGPT isn't an expert coder sitting by your side. It's not even remotely close. Go ahead and get ChatGPT to walk you through how to implement even a non-novel CFD analysis that won't get you laughed out of the room. Get it to help you port an Ada Ravenscar runtime to an exotic (or even a not-so-exotic) processor. Try and have it generate non-trivial ladder logic programs for industrial controls. Try to get any help from it at all when doing microcontroller programming that isn't just "read the manual". My question for you is -- what exactly do you do? What programming could you possibly be doing that is so trivial that you actually believe that ChatGPT is capable of solving these things?

rvnx 2y ago
No need to get so upset ? It is like if I say that reading a book will help you learn faster about a domain than if you have to discover all by yourself. Yes self-learning is better, but it takes a much longer period of time, whereas if you have a tutor, then it saves lot of time (and companies don't often have such resources, which is where AI and books fill the gap). I strongly believe that LLMs are a serious helping tool for programming that helps programmers to onboard their project faster. Regarding more exotic techs, as a cousin of ChatGPT, Google Gemini used to be very very bad, but with Gemini 1.5-Pro you can feed it very long documents, and this is super helpful for specific implementation (e.g. the exotic processors), and it's, really, really not bad at programming, or at least pushing you in the right direction. Of course it's not autonomous (and whether it can be in the short-term on complex projects is unlikely), but it reduces the onboarding time, and this was the point raised in the conversation. A dev paired with a LLM is much much more productive. I suppose that you are concerned that it may push people to lose their jobs in the long-term. I am as well, but we still have some time ahead. I don't like this situation either, but I have to recognize that it is a very helpful co-programming tool.
- seabird 2y ago
  I'm not so much upset as I am flabbergasted. Reading the documentation is basically never the hard part. If it's where most of your work is going, you're not doing hard work. Gemini 1.5-Pro may be able to summarize documentation, but it's what isn't there that hurts you. It may make for a helpful reference, but that wasn't the initial claim. The claim was that "domain-specific knowledge can be solved with a $20 ChatGPT subscription", and being frank, that's just stupid. The difference between a smart person and a domain expert is orders of magnitude more than an LLM is able to paper over. It's a struggle to replace even the most trivial paper pushing with an LLM. I'm sure we'll find something one day, but it's going to be doing a hell of a lot more than an LLM.

rvnx 2y ago

No need to get so upset ? It is like if I say that reading a book will help you learn faster about a domain than if you have to discover all by yourself. Yes self-learning is better, but it takes a much longer period of time, whereas if you have a tutor, then it saves lot of time (and companies don't often have such resources, which is where AI and books fill the gap). I strongly believe that LLMs are a serious helping tool for programming that helps programmers to onboard their project faster. Regarding more exotic techs, as a cousin of ChatGPT, Google Gemini used to be very very bad, but with Gemini 1.5-Pro you can feed it very long documents, and this is super helpful for specific implementation (e.g. the exotic processors), and it's, really, really not bad at programming, or at least pushing you in the right direction. Of course it's not autonomous (and whether it can be in the short-term on complex projects is unlikely), but it reduces the onboarding time, and this was the point raised in the conversation. A dev paired with a LLM is much much more productive. I suppose that you are concerned that it may push people to lose their jobs in the long-term. I am as well, but we still have some time ahead. I don't like this situation either, but I have to recognize that it is a very helpful co-programming tool.

seabird 2y ago
I'm not so much upset as I am flabbergasted. Reading the documentation is basically never the hard part. If it's where most of your work is going, you're not doing hard work. Gemini 1.5-Pro may be able to summarize documentation, but it's what isn't there that hurts you. It may make for a helpful reference, but that wasn't the initial claim. The claim was that "domain-specific knowledge can be solved with a $20 ChatGPT subscription", and being frank, that's just stupid. The difference between a smart person and a domain expert is orders of magnitude more than an LLM is able to paper over. It's a struggle to replace even the most trivial paper pushing with an LLM. I'm sure we'll find something one day, but it's going to be doing a hell of a lot more than an LLM.