I'm giving copilot a shot, using it for a lot of autocomplete and suggestion.
It's pretty sharp most of the time tbdesu, except when i've got a sloppy problem in the first place
Bing in the one I try the most but it's rarely helpful for actual programming. It's better at talking through design decisions in a broad sense, and giving feedback on code I wrote, but the feedback is usually worthless (LGTM tier) and/or any suggestions it makes are usually flatout wrong.
When it is useful, it's really useful though because it's usually talking through something with me nobody else I know has any knowledge of. Which is infinitley better than nothing, even if you have to coax good answers out of it and treat them with suspicion. It's enough to help get started.
If you look at the links it gives for references, you'll find everything it "says" is a regurgitation/synthesis of what's in those links.
Basically it's lmgtfy dumbed down for morons.
>Basically it's lmgtfy dumbed down for morons.
It's the opposite. It can understand the context of what you're asking rather than just key words. Which is very useful and not dumb. It's a lot more effective at seeing beyond synonyms and like-worded concepts, usually returning what you're actually searching for.
That's basically the key selling point for these things and what makes them a lot nicer than traditional search engines.
I'm on Gemini pro because Google one had a free 2 month trial. It's not bad sometimes it's really annoying like one time I had it write me test cases for email validation and it had "[redacted]" in all the cases. Useless.
After the trial ends I'll look around, will try Claude
I was already finding copilot's autocomplete great but I was still using gtp3.5 to ask programming questions because I was dumb enough to not know there's a Chat sidebar section to talk directly to Copilot, and it has been a lot better since. When you ask a question it references the current file you're working on, so it automatically gets context of what you're trying to do.
Only thing I find co-pilot lacks is full file review - You can dump your whole files in to GPT, and it'll spit it back out to you, but co-pilot will complain about being too long or pretend you never submitted any code
Copilot has waaaaaaaaay to small context window. I don't understand how people get any actual use out of the current models, they have to write incredibly simple applications.
As an AI, I don't use any tools or services for coding myself. However, I can assist users with coding by providing information, explanations, and examples based on my training data. Microsoft Copilot is designed to help users by offering suggestions and completing code based on the context provided by the user. It's a tool that leverages large language models to understand and generate human-like text, which can be particularly useful in coding scenarios. If you have any coding questions or need assistance with a project, feel free to ask!
if you can get the job done then what's the problem? they gonna sack you for not using a certain tool, they might sack you for not using a certain keyboard next
Honestly I just use the free messages bing gives me, the fact that it also searches before replying works great for the most part unless it's some obscure library on github and even then sometimes it reads the documentation
I'm cheap so I only use the free service. Copilot has 4000 char limit which is enough for some functions but not for files. Claude has file support so I can ask it directly. But Claude has way lot less usage limits
I run llamas locally in my terminal
I use GPT4 and Claude 3 because I like being able to "talk" to my AI and now just have something suggest code without any back and forth discussion
Typical western programmer, you used to have a programming duck before this, right?
I don't use one. I just can't be bothered
None because I want to enjoy programming and the built-in helpers of my IDE is more than enough.
program in the terminal using pure vim
I'm giving copilot a shot, using it for a lot of autocomplete and suggestion.
It's pretty sharp most of the time tbdesu, except when i've got a sloppy problem in the first place
>What LLM do you use
i don't use any LLM
>for coding
i don't code
I use gpt4, Claude 3 . Sometimes Whiterabbitneo and llama 70B code for stuff that is secret.
>create any imagine you can imagine
imagine
Show Bob and imagine
>image theres no heaven
Bing in the one I try the most but it's rarely helpful for actual programming. It's better at talking through design decisions in a broad sense, and giving feedback on code I wrote, but the feedback is usually worthless (LGTM tier) and/or any suggestions it makes are usually flatout wrong.
When it is useful, it's really useful though because it's usually talking through something with me nobody else I know has any knowledge of. Which is infinitley better than nothing, even if you have to coax good answers out of it and treat them with suspicion. It's enough to help get started.
what setting do you choose? creative/balanced/precise
creative is pretty based, balanced and precise are cucked
If you look at the links it gives for references, you'll find everything it "says" is a regurgitation/synthesis of what's in those links.
Basically it's lmgtfy dumbed down for morons.
>Basically it's lmgtfy dumbed down for morons.
It's the opposite. It can understand the context of what you're asking rather than just key words. Which is very useful and not dumb. It's a lot more effective at seeing beyond synonyms and like-worded concepts, usually returning what you're actually searching for.
That's basically the key selling point for these things and what makes them a lot nicer than traditional search engines.
I'm on Gemini pro because Google one had a free 2 month trial. It's not bad sometimes it's really annoying like one time I had it write me test cases for email validation and it had "[redacted]" in all the cases. Useless.
After the trial ends I'll look around, will try Claude
I was already finding copilot's autocomplete great but I was still using gtp3.5 to ask programming questions because I was dumb enough to not know there's a Chat sidebar section to talk directly to Copilot, and it has been a lot better since. When you ask a question it references the current file you're working on, so it automatically gets context of what you're trying to do.
Only thing I find co-pilot lacks is full file review - You can dump your whole files in to GPT, and it'll spit it back out to you, but co-pilot will complain about being too long or pretend you never submitted any code
It's GPT-4 but free and with an awful buggy interface
ChatGPT-4 with the Grimoire plugin for browser, Github Copilot for VS Code.
I find the browser to give a bit better answers overall, but the /fix and /explain and the chat of VS Code is pretty nice too.
Copilot has waaaaaaaaay to small context window. I don't understand how people get any actual use out of the current models, they have to write incredibly simple applications.
It's good for bashing out boring relatively simple unit tests in like 3 seconds each instead of 30-60
I want to use CoPilot but I don't like Microsoft knowing what I'm programming, so I just ask GPT-3.5 the ocassional spastic question.
My coding king is DeepSeek Coder 33b, I just love it how it pinpoints me to coding errors and suggests creative solutions.
this ie some nuprogrammer bullshit. If your IO functions do not take a format parameter you are not a programmer.
I used Github Copilot because Microsoft gave it to me for free but it's been a few months since I last touched it
Are you a big time github developer or something?
I have commit access to a multi-thousand star repo (not my own repo), dunno what the threshold is exactly
Participating in this thread has cursed my copilot, it stopped working
Then you may want to take control before you crash 😐
I actually use OpenAI ChatGPT 3.5, it works for longer code snippets.
Anyways, I’m off to buy an ad 🙂
As an AI, I don't use any tools or services for coding myself. However, I can assist users with coding by providing information, explanations, and examples based on my training data. Microsoft Copilot is designed to help users by offering suggestions and completing code based on the context provided by the user. It's a tool that leverages large language models to understand and generate human-like text, which can be particularly useful in coding scenarios. If you have any coding questions or need assistance with a project, feel free to ask!
buy an ad microshitter
you dumb Black person this is Microsoft Copilot, a totally different product than Github Copilot (the pair-programming AI youre talking about)
It's copilot because my company pays for it
Problem is it can't do anything on multiple files, like examine an entire project, which makes it borderline useless for certain repos
For now, it does okay creating functions, optimizing functions, etc. In small batches.
My brain
i use gemini
it's turned me from a mediocore jr dev to a moderate sr dev who can chrun out code like i have a team of monkeys under me
>copilot has detected illegal content on your pc
>an fbi van will be with you shortly
>What LLM do you use for coding
None. They are all useless.
programmer who will replaced by ai who cant cope with it for is last day of """work""" before starting to plumbe the shit out of tyrone toilet #45141
if you can get the job done then what's the problem? they gonna sack you for not using a certain tool, they might sack you for not using a certain keyboard next
at that point they will just get an indian in remote and leave you plumbe the shit of tyrone toilet while THE BULL smoke weed
Copilot made me a fine arse Script-Fu.
Honestly I just use the free messages bing gives me, the fact that it also searches before replying works great for the most part unless it's some obscure library on github and even then sometimes it reads the documentation
I'm not bad enough at coding to want one.
I'm cheap so I only use the free service. Copilot has 4000 char limit which is enough for some functions but not for files. Claude has file support so I can ask it directly. But Claude has way lot less usage limits