My work procured me a copilot license and it really is disappointing. It’s wrong or overly vague most of the time, and just generally doesn’t answer questions in a useful way. Copilot in the IDE is marginally more useful but it rarely gives you an efficient answer and obviously it’s never anything novel, it’s just copy and pasting from stackoverflow answers (or questions even.)
I have copilot at work and honestly it’s not worth the price.
The number of times it has created grossly inaccurate meeting notes or summary items basically means I can’t trust it to be shared with someone who wasn’t there, so it’s mostly just there as a roll the dice memory jogger for participants.
The components embedded in office apps like PowerPoint are absolutely useless, and that’s where I really wanted it to help.
I’ve found it to have extremely limited value, but not zero. It’s been useful as a shortcut for things I can already do myself. For instance, I can easily get syntax for a param block, or build a window form. Could I do it myself? Absolutely, and pretty easily. And I can recognize when it’s right vs wrong. But it’s marginally faster to have copilot do it instead of digging up the documentation.
It’s more like a party trick than a trillion dollar revolution. The $20/month for a full time dev is probably around the break even point for the labor savings. It’s not going to save THAT much time.
sigmaklimgrindset@sopuli.xyz
on 13 Feb 16:00
collapse
You’re not IN blahaj sonny 😎
homesweethomeMrL@lemmy.world
on 13 Feb 14:47
nextcollapse
It says in multiple places, “do not believe the answers you get from this thing. Seriously, it’s more like a magic 8 ball than anything useful. You get what you get. Don’t come bitching to us.”
_haha_oh_wow_@sh.itjust.works
on 13 Feb 16:08
nextcollapse
I’ve used co-pilot too: Not impressed. AI is largely unreliable garbage, at least at the moment.
CosmicTurtle0@lemmy.dbzer0.com
on 13 Feb 17:01
nextcollapse
Early co-pilot (before chatgpt) was pretty decent. You typed in a comment what you wanted to do and you could cycle through code examples easily. While it wasn’t 100% accurate, it was close enough to get you what you needed and took care of the annoying work.
As soon as ChatGPT got integrated, shit went downhill fast and I uninstalled it. It was a night and day difference.
You’re thinking github copilot while the article is about Microsoft copilot, which is a completely different product. Of course, confusion is completely understandable and their naming is almost as bad as their operating system.
Copilot is one of two LLMs I’ve briefly tried. It was noticeably better than Gemini was at the time, but still seemed entirely pointless. Nothing it (or Gemini) offered to do were things I wanted help with. I enjoy research and writing, so why would I outsource those things and burn down an acre of rainforest in the process?
_haha_oh_wow_@sh.itjust.works
on 13 Feb 18:25
collapse
Gemini is fucking abysmal, Google should be embarrassed.
I can think of a bunch of other things that Google should be embarrassed about, but Gemini is uniquely humiliating because of how proud they are of it and how hard they are pushing it.
I’d be more forgiving, but what MS Copilot sucks at the worst is working with Office files:
PowerPoint: completely useless
Excel: useful for some tasks if I reformat all my sheets into tables, so mostly useless
Word: can’t even answer why the formatting of the file from my coworker is a mess (which is every coworker), let alone try to fix it. But it will gladly take a succinct piece of information and make it into an insufferable 50 pages of fluff. It summarizes documents well and helps with reviews.
OneNote: creating content it’s ok, but useless at retrieving info.
Outlook: It tries to help, mostly confirms that my language is may be perceived as cold and offensive. Which is good, because that was my intention.
It is surprisingly helpful with PDFs and extracting data and cleaning up formatting.
Unsurprising, it does great with CSV and other open data formats. Maybe there is a lesson here?
snooting@sub.wetshaving.social
on 13 Feb 18:58
collapse
I think the lesson here is that proprietary binary formats are way more difficult for a LLM to parse.
If only we could use plaintext for everything…
I was wrong, see comment from @NocturnalEngineer@lemmy.world
NocturnalEngineer@lemmy.world
on 13 Feb 19:57
collapse
They’ve been using OOXML for their file formats since 2007.
snooting@sub.wetshaving.social
on 13 Feb 20:19
collapse
Oh, it looks like you’re right! TIL
werefreeatlast@lemmy.world
on 13 Feb 20:43
collapse
I got a better one…put a rat in a can, place two papers with two different answers… As a question and let the rat go. Each paper is coated in delicious cheese.
Just figure out if you trust a rat…do you go for the answer an idiot picked? Or do you go for the answer that was less tasty?
HeyThisIsntTheYMCA@lemmy.world
on 14 Feb 10:15
collapse
Often the trick behind a lot of these ridiculous methods is they make you hope for a specific outcome, thus exposing a previously hidden desire
threaded - newest
Colour me surprised
I had low expectations for AI, and it is even less useful than I expected.
My work procured me a copilot license and it really is disappointing. It’s wrong or overly vague most of the time, and just generally doesn’t answer questions in a useful way. Copilot in the IDE is marginally more useful but it rarely gives you an efficient answer and obviously it’s never anything novel, it’s just copy and pasting from stackoverflow answers (or questions even.)
I miss functioning web searches.
AI is filled with as much, if not more, SEO designed slop as web searches do now but it is presented as if the user clicked “I feel lucky”.
I have copilot at work and honestly it’s not worth the price.
The number of times it has created grossly inaccurate meeting notes or summary items basically means I can’t trust it to be shared with someone who wasn’t there, so it’s mostly just there as a roll the dice memory jogger for participants.
The components embedded in office apps like PowerPoint are absolutely useless, and that’s where I really wanted it to help.
I’ve found it to have extremely limited value, but not zero. It’s been useful as a shortcut for things I can already do myself. For instance, I can easily get syntax for a param block, or build a window form. Could I do it myself? Absolutely, and pretty easily. And I can recognize when it’s right vs wrong. But it’s marginally faster to have copilot do it instead of digging up the documentation.
It’s more like a party trick than a trillion dollar revolution. The $20/month for a full time dev is probably around the break even point for the labor savings. It’s not going to save THAT much time.
Agreed, it’s not completely devoid of value, but it’s definitely not saving me any super meaningful time.
There’s some labor efficiency but I would be really surprised if it even got to an hour a week in productivity gains.
I’ve heard that copilot is specifically useless. Mostly gets in the way.
No fucking way.
I like Copilot. [Blajah does not support downvotes.]
blajah
It seems that some people don’t get the joke. At least, I assume it’s a joke.
Yes, thank you hahaha. What’s the downvote count up to? I can’t see them due to my instance.
Here’s the universal link:
lemmy.ca/post/39050251
2 upvotes (including mine), 8 downvotes.
You’re not IN blahaj sonny 😎
It says in multiple places, “do not believe the answers you get from this thing. Seriously, it’s more like a magic 8 ball than anything useful. You get what you get. Don’t come bitching to us.”
I’ve used co-pilot too: Not impressed. AI is largely unreliable garbage, at least at the moment.
Early co-pilot (before chatgpt) was pretty decent. You typed in a comment what you wanted to do and you could cycle through code examples easily. While it wasn’t 100% accurate, it was close enough to get you what you needed and took care of the annoying work.
As soon as ChatGPT got integrated, shit went downhill fast and I uninstalled it. It was a night and day difference.
You’re thinking github copilot while the article is about Microsoft copilot, which is a completely different product. Of course, confusion is completely understandable and their naming is almost as bad as their operating system.
Wait those are distinct?! Holy shit. And since Microsoft owns GitHub, that’s even less clear. Unbelievable.
Copilot is one of two LLMs I’ve briefly tried. It was noticeably better than Gemini was at the time, but still seemed entirely pointless. Nothing it (or Gemini) offered to do were things I wanted help with. I enjoy research and writing, so why would I outsource those things and burn down an acre of rainforest in the process?
Gemini is fucking abysmal, Google should be embarrassed.
I can think of a bunch of other things that Google should be embarrassed about, but Gemini is uniquely humiliating because of how proud they are of it and how hard they are pushing it.
I’d be more forgiving, but what MS Copilot sucks at the worst is working with Office files: PowerPoint: completely useless
Excel: useful for some tasks if I reformat all my sheets into tables, so mostly useless
Word: can’t even answer why the formatting of the file from my coworker is a mess (which is every coworker), let alone try to fix it. But it will gladly take a succinct piece of information and make it into an insufferable 50 pages of fluff. It summarizes documents well and helps with reviews.
OneNote: creating content it’s ok, but useless at retrieving info.
Outlook: It tries to help, mostly confirms that my language is may be perceived as cold and offensive. Which is good, because that was my intention.
It is surprisingly helpful with PDFs and extracting data and cleaning up formatting.
Unsurprising, it does great with CSV and other open data formats. Maybe there is a lesson here?
I think the lesson here is that proprietary binary formats are way more difficult for a LLM to parse.If only we could use plaintext for everything…I was wrong, see comment from @NocturnalEngineer@lemmy.world
They’ve been using OOXML for their file formats since 2007.
Oh, it looks like you’re right! TIL
I got a better one…put a rat in a can, place two papers with two different answers… As a question and let the rat go. Each paper is coated in delicious cheese.
Just figure out if you trust a rat…do you go for the answer an idiot picked? Or do you go for the answer that was less tasty?
Often the trick behind a lot of these ridiculous methods is they make you hope for a specific outcome, thus exposing a previously hidden desire