ChatGPT is hilariously incompetent… but on a serious note, I still firmly reject tools like copilot outside demos and the like because they drastically reduce code quality for short term acceleration. That’s a terrible trade-off in terms of cost.
As a software engineer, the number of people I encounter in a given week who either refuse to or are incapable of understanding that distinction baffles and concerns me.
That’s because it’s being advertised as a solution. That’s why you have people worried it’ll take their jobs when in reality it’ll let them do the job better.
The problem I have with it is that all the time it saves me I have to use on reading the code. I probably spend more time on that as once in a while the code it produces is broken in a subtle way.
I see some people swearing by it, which is the opposite of my experience. I suspect that if your coding was copying code from stack overflow then it indeed improved your experience as now this process is streamlined.
I don’t know if it does yet, but if ChatGPT starts providing source for every information, then it would make it much faster to find the relevant information and check their sources, rather than clicking websites one by one.
It used to have a problem with making a claims that were not relevant to or contradicted its own sources, but I don’t recall encountering that problem recently.
Biggest problem with it is that it lies with the exact same confidence it tells the truth.
Or, put another way, it’s confidently incorrect as often as it is confidently correct - and there’s no way to tell the difference unless you already know the answer.
it’s kinda hilarious to me because one of the FIRST things ai researchers did was get models to identify things and output answers together with the confidence of each potential ID, and now we’ve somehow regressed back from that point
i mean giving a confidence for recognizing a certain object in a picture is relatively straightforward.
But LLMs put together words by their likeliness of belonging together under your input (terribly oversimplified).the confidence behind that has no direct relation to how likely the statements made are true. I remember an example where someone made chatgpt say that 2+2 equals 5 because his wife said so. So chatgpt was confident that something is right when the wife says it, simply because it thinks these words to belong together.
My hairline has started receding very rapidly. There’s there’s these fine hairs all over my desk, and I see the photo I took when joining directly before turning on my camera every meeting.
Doesn’t sood good at all. I’m sorry to hear that, friend. I really hope there’s enough upsides there compared to working at a more mature company for you.
Sort of. Nobody’s cutting corners on aviation structural components, for example. We’ve been pretty good at maximizing general value output, and usually that means lower quality, but not always.
I’m going to say that’s the exception that proves the rule, assuming they were structural parts and not a minor controller chip for de-icing or something.
The company themself announced it without being prompted, and if whoever introduce these unapproved parts into a small number of engines is caught there’s going to be real hell to pay. The stuff that stops you from falling out of the sky is serious business, and is largely treated as such.
On the other hand, a software function that’s hacked together and inefficient will just fly below the radar, and most people will prefer two cheap outfits to one that’s actually well made for the same price, so quality goes right out the window.
It’s helped me a bit with resolving weird tomcat/Java issues when upgrading to RHEL8, though. It didn’t give me an answer, but it gave me ideas on where to look (in my case I didn’t realize fapolicyd replaced selinux)
That’s the point - you have the expertise to make proper sense of whatever it outputs. The people pushing for “AI” the most want to rely on it without any necessary expertise or just minimal efforts, like feeding it some of your financial reports and have generate a 5-year strategy only to fail miserably and have no one to blame this time (will still blame anyone else but themselves btw).
It’s not the most useless tool in the world by any means, but the mainstream talk is completely out of touch with reality on the matter, and so are mainstream actions (i.e. overrelying on it and putting way too much faith into it).
An unpopular opinion, I am sure, but if you’re a beginner with something - a new language, a new framework - and hate reading the docs, it’s a great way of just jumping into a new project. Like, I’ve been hacking away on a django web server for a personal project and it saved me a huge amount of time with understanding how apps are structured, how to interact with its settings, registering urls, creating views, the general development lifecycle of the project and the basic commands I need to do what I’m trying to do. God knows Google is a shitshow now and while Stackoverflow is fine and dandy (when it isn’t remarkably toxic and judgmental), the fact is that it cuts down on hours of fruitless research, assuming you’re not asking it to do anything genuinely novel or hyper-specific.
It helps a complete newbie like me get started and even learn while I do. Due to its restrictions and shortcoming, I’ve been having to learn how to structure and plan a project more carefully and thoughtfully, even creating design specs for programs and individual functions, all in order to provide useful prompts for ChatGPT to act on. I learn best by trial and error, with the ability to ask why things happened or are the way they are.
So, as a secondary teaching assistant, I think it’s very useful. But trying to use the API for ChatGPT 4 is…not worth it. I can easily blow through $20 in a few hours. So, I got a day and a half of use out of it before I gave up. :|
ChatGPT is hilariously incompetent… but on a serious note, I still firmly reject tools like copilot outside demos and the like because they drastically reduce code quality for short term acceleration. That’s a terrible trade-off in terms of cost.
I enjoy using copilot, but it is not made to think for you. It’s a better autocomplete, but don’t ever let it do more than a line at once.
Yup, AI is a tool, not a complete solution.
As a software engineer, the number of people I encounter in a given week who either refuse to or are incapable of understanding that distinction baffles and concerns me.
That’s because it’s being advertised as a solution. That’s why you have people worried it’ll take their jobs when in reality it’ll let them do the job better.
The problem I have with it is that all the time it saves me I have to use on reading the code. I probably spend more time on that as once in a while the code it produces is broken in a subtle way.
I see some people swearing by it, which is the opposite of my experience. I suspect that if your coding was copying code from stack overflow then it indeed improved your experience as now this process is streamlined.
deleted by creator
Same as ChatGPT is better web search.
id rather search the web than chatgpt because i fact check it anyway
I don’t know if it does yet, but if ChatGPT starts providing source for every information, then it would make it much faster to find the relevant information and check their sources, rather than clicking websites one by one.
Yep, ChatGPT4 allows optional calls to Bing now.
It used to have a problem with making a claims that were not relevant to or contradicted its own sources, but I don’t recall encountering that problem recently.
Biggest problem with it is that it lies with the exact same confidence it tells the truth. Or, put another way, it’s confidently incorrect as often as it is confidently correct - and there’s no way to tell the difference unless you already know the answer.
it’s kinda hilarious to me because one of the FIRST things ai researchers did was get models to identify things and output answers together with the confidence of each potential ID, and now we’ve somehow regressed back from that point
did we really regress back from that?
i mean giving a confidence for recognizing a certain object in a picture is relatively straightforward.
But LLMs put together words by their likeliness of belonging together under your input (terribly oversimplified).the confidence behind that has no direct relation to how likely the statements made are true. I remember an example where someone made chatgpt say that 2+2 equals 5 because his wife said so. So chatgpt was confident that something is right when the wife says it, simply because it thinks these words to belong together.
Oh boy do I have news for you, that’s basically the only thing middle managers care about, short tem acceleration
I’m still convinced that GitHub copilot is actively violating copyleft licenses. If not in word, then in the spirit.
Removed by mod
Western society is built on this principle
Tell me about it…
I left my more mature company for a startup.
I feel like Tyler Durden sometimes.
How you liking it? How many years have you aged in the months working at your startup?
My hairline has started receding very rapidly. There’s there’s these fine hairs all over my desk, and I see the photo I took when joining directly before turning on my camera every meeting.
Doesn’t sood good at all. I’m sorry to hear that, friend. I really hope there’s enough upsides there compared to working at a more mature company for you.
Sort of. Nobody’s cutting corners on aviation structural components, for example. We’ve been pretty good at maximizing general value output, and usually that means lower quality, but not always.
May want to show your roll a bit…
https://fortune.com/2023/10/03/delta-fourth-major-us-airline-fake-jet-aircraft-engine-parts-forged-airworthiness-documents-uk-company-aog/
I’m going to say that’s the exception that proves the rule, assuming they were structural parts and not a minor controller chip for de-icing or something.
The company themself announced it without being prompted, and if whoever introduce these unapproved parts into a small number of engines is caught there’s going to be real hell to pay. The stuff that stops you from falling out of the sky is serious business, and is largely treated as such.
On the other hand, a software function that’s hacked together and inefficient will just fly below the radar, and most people will prefer two cheap outfits to one that’s actually well made for the same price, so quality goes right out the window.
deleted by creator
It’s helped me a bit with resolving weird tomcat/Java issues when upgrading to RHEL8, though. It didn’t give me an answer, but it gave me ideas on where to look (in my case I didn’t realize fapolicyd replaced selinux)
That’s the point - you have the expertise to make proper sense of whatever it outputs. The people pushing for “AI” the most want to rely on it without any necessary expertise or just minimal efforts, like feeding it some of your financial reports and have generate a 5-year strategy only to fail miserably and have no one to blame this time (will still blame anyone else but themselves btw).
It’s not the most useless tool in the world by any means, but the mainstream talk is completely out of touch with reality on the matter, and so are mainstream actions (i.e. overrelying on it and putting way too much faith into it).
Yeah, it can speed up the process but you still have to know how to do it yourself when it inevitably messes something up.
An unpopular opinion, I am sure, but if you’re a beginner with something - a new language, a new framework - and hate reading the docs, it’s a great way of just jumping into a new project. Like, I’ve been hacking away on a django web server for a personal project and it saved me a huge amount of time with understanding how apps are structured, how to interact with its settings, registering urls, creating views, the general development lifecycle of the project and the basic commands I need to do what I’m trying to do. God knows Google is a shitshow now and while Stackoverflow is fine and dandy (when it isn’t remarkably toxic and judgmental), the fact is that it cuts down on hours of fruitless research, assuming you’re not asking it to do anything genuinely novel or hyper-specific.
It helps a complete newbie like me get started and even learn while I do. Due to its restrictions and shortcoming, I’ve been having to learn how to structure and plan a project more carefully and thoughtfully, even creating design specs for programs and individual functions, all in order to provide useful prompts for ChatGPT to act on. I learn best by trial and error, with the ability to ask why things happened or are the way they are.
So, as a secondary teaching assistant, I think it’s very useful. But trying to use the API for ChatGPT 4 is…not worth it. I can easily blow through $20 in a few hours. So, I got a day and a half of use out of it before I gave up. :|