The point of posting a link is to get someone to read it, no? Posting a URL that isn't even a hyperlink is peak inefficiency because nobody is gonna read it. Wasted effort.
At least we can click HH's shitty MSN links and can sort of make out what it's about in the URL. Your garbage X links provide an awful UX for your customers. There's nothing.
The results have been mixed for me. Yes for your need I think it's fine.
But when just asking for some basic methodology for bulk synchronizing data for example, both llm's have given me completely incorrect solutions that assumed that a utility function in postgres worked the exact same in redshift.
For algorithms for some basic workflow, both will give decent starting points but still several things to rip out and do properly.
Biggest win has been implementing an agent to replace our old documentation stuff, although if your data is bad because your documentation team was bad...how useful is something like that.
Yeah I’ve been really slow on usage just because it is kinda frustrating. But I’d say lately I’m getting some value out of it as I use for specific well defined things that are small enough in scope. Like if the alternative was looking up some not so good docs or something and trial and erroring it.
And since I don’t really do like marketing or creative stuff it’s not as good.
I was thinking I’d try having some docs generated next for some public stuff I’m building. We shall see.
An AI plugin into your IDE isn't a bad idea especially if you have developers that struggle or have a lot of shit kicked back to them. It's just been kind of the same as everything else for me: the good engineers are even better now with it, and the bad engineers still just find a way to be bad, but it's not a situation where just an AI can completely replace them in a private, proprietary code environment where you lose domain knowledge.
Related but different experience from a much less technical side.
It's been the same experience for us at an implementation level for OPM. Good PM's benefit from having one more tool or "assistant". For the rest It's mostly worthless. It can increase productivity but not replace it.
The tell on it's limitations was when lots of major companies pulled back on capital investment for "digital transformation".
We’re playing around with some stuff for our sales team crm shit but only because all the premium features got rolled into the normal package.
I mean it’s definitely somewhat useful in a lot of different applications but now that the whole overblown hype train is ending I think people’s expectations have been lowered and they are willing to get some value out of it in more specific ways.
Comments
Expending multiple tldr replies to explain how you don’t have time to copy paste.
The point of posting a link is to get someone to read it, no? Posting a URL that isn't even a hyperlink is peak inefficiency because nobody is gonna read it. Wasted effort.
At least we can click HH's shitty MSN links and can sort of make out what it's about in the URL. Your garbage X links provide an awful UX for your customers. There's nothing.
lol you didn't even understand what I said about AI because you've never used it for software engineering in practice.
imagine thinking people in tech are efficient
I have been playing around with my boy Claude.
Helped me write the auth for a client library to work with an api that uses fucking oauth 1.0 lol
I feel like it mostly got it right and it’s a pretty good use case (something that’s pretty well defined but super tedious to write)
Maybe we can ask it why buck is such a fag?
So it's pretty much confirmed he doesn't work in tech, right?
The results have been mixed for me. Yes for your need I think it's fine.
But when just asking for some basic methodology for bulk synchronizing data for example, both llm's have given me completely incorrect solutions that assumed that a utility function in postgres worked the exact same in redshift.
For algorithms for some basic workflow, both will give decent starting points but still several things to rip out and do properly.
Biggest win has been implementing an agent to replace our old documentation stuff, although if your data is bad because your documentation team was bad...how useful is something like that.
Yeah I’ve been really slow on usage just because it is kinda frustrating. But I’d say lately I’m getting some value out of it as I use for specific well defined things that are small enough in scope. Like if the alternative was looking up some not so good docs or something and trial and erroring it.
And since I don’t really do like marketing or creative stuff it’s not as good.
I was thinking I’d try having some docs generated next for some public stuff I’m building. We shall see.
An AI plugin into your IDE isn't a bad idea especially if you have developers that struggle or have a lot of shit kicked back to them. It's just been kind of the same as everything else for me: the good engineers are even better now with it, and the bad engineers still just find a way to be bad, but it's not a situation where just an AI can completely replace them in a private, proprietary code environment where you lose domain knowledge.
Related but different experience from a much less technical side.
It's been the same experience for us at an implementation level for OPM. Good PM's benefit from having one more tool or "assistant". For the rest It's mostly worthless. It can increase productivity but not replace it.
The tell on it's limitations was when lots of major companies pulled back on capital investment for "digital transformation".
We’re playing around with some stuff for our sales team crm shit but only because all the premium features got rolled into the normal package.
I mean it’s definitely somewhat useful in a lot of different applications but now that the whole overblown hype train is ending I think people’s expectations have been lowered and they are willing to get some value out of it in more specific ways.