When I read stories about people 10x’ing their productivity with AI, I interpret them as having the same vibe as these “solvers”.
There are a lot of tasks that follow a pretty predictable but not automatable pattern that LLMs do a great job at, but that you couldn't leverage macros or other powerful editor tools to complete. Getting good at LLMs can greatly increase your productivity. But unless your job is just gluing well-defined interfaces together, spinning up demos, or writing documentation for existing code, you won't get 10x. 2x is very realistic for most programming jobs, though. Some tasks would potentially be below 1x, especially low level, novel, thorny logic. I'd never use it for developing new cryptographic primitives or a novel compression codec, for instance.
The phrasing you use is all marketing. When people share actual chat threads that they claim helped them, they always have the same issues I observe in my own use.
They won't do the thinking for you. You still have to know what you want out of it, you have to know your endpoint, and you have to be able to review and understand every line out of it, but they do save time. You might have tried to use them in situations where they aren't appropriate. They are text prediction programs, but they're extremely powerful text prediction programs. If you accept their unreliability and give them the right patterns and context to generate from, they can write a thousand lines of code in one go that look identical to what you would have written, and it takes 30 seconds to generate and a few minutes to review and test, compared to the half an hour at least that it would have taken otherwise. It can easily yield 10 times efficiency gain in places where it's really appropriate.
That said, it's not perfect by any means. To be really what I want, I need the same quality significantly faster, cheaper, and local. If my company wasn't paying for it, I certainly wouldn't pay to use this thing myself. This costs between $0.05 and $1 per request, which gets expensive fast, and the much cheaper models are less capable enough to save me so little time as to not be worth the bother really.
People have been sharing that, plus the problems they’ve solved or products they’ve made.
Though, most of these workflows are no longer really just “chats” at like chatgpt.com. A lot of that has changed.
i’m not saying you need to do it, or change anything, but… i’ve seen people do shit with a computer lately that 6 months ago would seem weird and alien.
Do you have video?
But those discussions don't crest the hill because the "verified" work consistently either fails to be addressing a challenge that the skeptics encounter or has qualities that the skeptics can't imagine themselves possessing.
The reality is that we all approach our work differently and all carry different standards in what we accept in the work we produce.
It's clear that generative AI genuinely helps some people improve their work or ease their burden of producing it, but it should also be clear at this point that the people it doesn't seem to help are not being failed for lack of having learned about its use cases. We're 2+ years in and are saturated with the hype and with such examples. It's just not as universally useful as some people (for whatever reason) want to believe.
It's fine if you've found a use-case that LLMs aren't so helpful, like nonfiction writing that you don't want to have to constantly verify for hallucinations. Obviously those exist.
But don't forget that they are calling into question whether other people really are super productive with LLMs, and that hints at a self-limiting belief.
It honestly sounds like a cult where I don’t know how to read the tea leaves and believe enough. I’ll pass happily and be so self limited :)
When AI is actually useful, there won’t be a debate.
AI even with flaws is good enough. I'm happy to pick up the profit on the difference in our beliefs and behaviors. It's easy to simply out compete.
Your original comment was suggesting that people who haven't found a use-case that works for them just haven't looked hard enough, despite how inescapable the topic and tools are.
Some hypothetical person that believes nobody might have a personally effective use case is a different one than what you seemed to be writing about, and completely unrelated to what I was writing about.
To be fair, the distinction gets muddy in both directions because some people invest their egos into/against this kind of thing and can't just accept differences as they are. But at that point, with those people, you're just stuck in the same class of endless flamewar debate as emacs vs vim or Xbox vs Playstation and there's really no point in giving time or attention to that.
I recently asked Claude to help deserialize a JSON Post request in an asp.net web API so that absent or null JSON properties get set to the coded default values. They gave me what was nearly the right answer (set NullValueHandling to Ignore) but led me down a merry garden path by saying I should also set DefaultValueHandling to Populate. Every time I said it wasn't working and gave the error message, it came up with additional code to put in, which never helped. Eventually I just started toying around with it myself and realised the solution, but it was maybe half a day down the drain.
Perhaps I could have taken a different approach from the start, but still, my impression is that the tools are not what they're made out to be.
If you got NullValueHandling, jumping to the library documentation would give you immediate accurate results.