Oh an my brain farts with the pricing structure, the 30 days free and $10 a month tier is abolutely fine... I"m just challenged at times... so my apologies to Microsoft on that mini rant. ;) Key Links gamefromscratch.com/microsoft-release-free-github-copilot/ ----------------------------------------------------------------------------------------------------------- *Support* : www.patreon.com/gamefromscratch *GameDev News* : gamefromscratch.com *GameDev Tutorials* : devga.me *Discord* : discord.com/invite/R7tUVbD *Twitter* : twitter.com/gamefromscratch *BlueSky*: bsky.app/profile/gamefromscratch.bsky.social -----------------------------------------------------------------------------------------------------------
Copilot pro costs $10. That first 30 days thing was just telling you that the cost was $0 for the first month. It's the opposite of what you complained about. Rather than showing the intro price and hiding the normal cost, they were showing the normal cost and hiding the intro cost.
Lol yeah, 100% my mistake. I added a response in my twitter, Bluesky and discord link as well as the comment above. My brain just malfunctioned there and assumed Microsoft did something scummy, but as you say, what they did was just fine. My bad.
Yeah, one can easily exhaust that within a few minutes. Its just a demo mode if it's true that every suggestion counts towards the limit. I will try tomorrow.
Haven't found/read their licence for Copilot. In their FAQ however to the question if Copilot is trained on Business or Enterprise data" they state it is not. So translated for everyone else: "Yes. In Free and Pro tier we train our AI on your stuff" Doesn't exactly answer the "who owns the code" situation but still not great.
There is a option to let copilot use your code for learning. But it is opt in (at least it was when I got my plan). It is definetly something to deactivate in the options...
Not giving away in the sense of ownership, you still own code you write. But yes you are likely giving them the right to train models on your code. It's also not clear (in the US, at least) about the copyrightability of generated code. It's dangerous to litter your codebase with a bunch of code of unknown provenance, and that may not be subject to copyright protection. It's why most enterprises are 1- paying to train their own models, and 2- banning use of genAI tools for their internal codebases. I think the big users here are desperate students, juniors, and contractors, but the big money is in enterprise. I don't see how they square that circle 🤷♀
@@mandisaw I agree with what you're writing. I dislike the fact that it is so integrated and have no intention of ever using these kinds of tools directly in my code base. I only use ChatGPT for naming related questions or css styling questions but do this in the browser where I am the one providing the context in my prompt so I can choose what to share and what not.
Is 2000/month 80/day enough? Depends on what they consider a code completion. Does it count if it's just the suggestion, or only after you accept the suggestion by hitting tab or whatever? If the former, you'll hit the limit in no time sinfe you get a suggestion if you stop typing for mote than a few seconds, but even if it's just accepted completions, it won't take long to use them up. It's just so gosh darn useful (hey YT). I saved many hours of typing just the first evening I played with it.
Yeah, that seems like a really weird metric to count. Even when you’re typing along, it’s filling out suggestions [it’s pretty awesome, mostly]. Does it count if I don’t even accept what it’s suggesting?
VS Code should have make copilot an extension and give these feature needed for the copilot to all extension api so that cursor and windsurf also use that to build an extension rather than a fork that would made vs code more powerful as people would be able to create better extension but nah they goona make vs code their copilot editor
I can't help thinking that they are already reading all my code even without the copilot extension, the free tier only makes me be more paranoid. With the NDA's i usually get for creating company code i wonder if or when will there be a lawsuit.
The bigger issue, I think, is that a lot of people can't read (well), and aren't good at problem-solving. So many folks say they struggle to understand and incorporate even straightforward documentation, or don't know how to apply lessons from tutorials to their own projects. Seems to affect self-learners more than school-learners, but that could be survivor bias, since folks generally can't make it through an entire college program without those skills. The selling-point for genAI coding sounds to me like an "easy button" to avoid having to practice your own comprehension and problem-solving skills. All this talk about boilerplate makes little sense - IDEs solved that problem with custom snippets, templates, and robust code-completion. In that sense, 2000 completions daily is obviously not enough, if you're using it to check or generate every little thing.
you still have to know how to code so you know when its right or wrong, its just a helping tool. its kind of like saying people shouldnt google things for help, just learn everything there is to know. coding is often not about writing code, its about how to solve the problem, what data structures to use, writing the code with the correct syntax is just the last step in the process. i have been so much more productive since AI has come out.
@@raidev_ It's actually a lot harder to do that. It's hard enough to understand another human's thought processes and untangle their mistakes. But genAI will recode things differently on each pass unless you explicitly override it, and it has no understanding of architecture or maintainability. If the error is not trivial and immediately obvious, it'll often be cheaper & faster to just rewrite it from scratch - at that point, you're spending more money and time than in the first place.
@@joeeyaura I'm glad it's working for the level where you're at. But that's not really applicable to the sorts of problems we solve at mid-career or higher levels. Comparing it to Google is apt - Google can't tell me the best way to design and implement software to run our business, it can't give insight on which solutions to buy vs build, and it can't foresee the possible gotchas of any given choice. Those are the problems software engineers solve, and genAI isn't equipped to tackle them at all. As you say, writing the code and worrying about syntax is the easiest step. (Not the final one, by a long shot - still have to test, deploy, support, maintain, and eventually sunset or transition to the next system. GenAI can't help with those either, really.)
@@Icemourne_ No, they're much better with JS frameworks, Python, and other stuff that's popular tutorial fodder on UA-cam or Medium. Complete trash for "unsexy" stuff like production-ready SQL, Java & [.]Net beyond basic CRUD apps, shell scripting, and anything where most of the code is enterprise-internal and unavailable for public scraping.
They say it’s really good at Python. I don’t code in Python frequently. It’s pretty amazing at Java, TypeScript, GraphQL, Bash, C#…pretty much anything you can throw at it. I think where they miss is that it is an AMAZING tool for EXPERIENCED programmers. It doesn’t turn novice programmers into good ones, it just lets them pump out bad code faster.
My experience so far is not using auto code completeion AI co pilot type things , it is asking questions when I get stuck with a Rust crates ,that may not have great documentation or an ambiguous implementation . What I have found is when the crate or library is new , it only supplies me with old incompatible documentation. So it doesnt know how to crawl cargo docs to understand the documentation from ChatGPT, Claude and LLAMA is like 3 years behind with documentation. I'm not really keen to learn it , tho I understand a pro in it trying to anticipate your code snippets being cool , Im just not hyped yet , until it can quickly right lexers , AST's , LSP's and a compiler for me , for a new language that doesnt exist in any public places yet .
If you have Copilot suggestions enabled and you're a binge coder (whether on the clock or the hyperfocus hobbyist variety) you can hit that 80/day threshold pretty quick. It more than suffices for code review, part-time learners, or light to moderate coding habits.
I like AI from, like, a conceptual point of view, but man. I don't LOVE coding but I do LOVE the results when I make something good. That's just lost with AI lol. Same with art for me. I think that maybe some sort of like code refactoring thing could be cool? Like, all on device? Cause I've written some messy code. But it's still my code at least lol
Thing is, you're not supposed to use it to generate everything lol. For example in my Advent Of Code 2024 repo I have a ISolution interface. When I create a new one, it will autofill the minimum needed for me to get started then it'll even copy the bit of code I use to show the results when I'm done with the algorithm (that thing is a pain but it's my fault for making it the way it is). I also use it to generate documentation. It's pretty good at that most of the time and it saves a lot of time for me since it's easier to read what it suggested then accept/edit it than figure out how to phrase what I want (English is my second language)
80 a day will get eaten up extremely quickly. A clever hobbyist dev on a game could make use of those 80 completions each day solving the toughest problems, but it isn't enough for heavy coding work. This is really a free sample to get people in the door.
2000/month code completion is nowhere enough if you intensively coding and need copilot to help your repetitive coding. It's similar to free tier to cursor. I guess it's 'enough' if you turn-off auto-completion features.
@@Yezu666 Not really, vscodium, is almost a faithful copy of vscode. It only removes M$ telmetry and marketplace, AI support remains (I think). But the bigger problem is that it lacks htis own identity
If it stays free forever.. Also you would have to make the chat interface and have it use the files as context, analyze files for the code completion, and so on. Definitely possible, but that's a lot of work.
Oh an my brain farts with the pricing structure, the 30 days free and $10 a month tier is abolutely fine... I"m just challenged at times... so my apologies to Microsoft on that mini rant. ;)
Key Links
gamefromscratch.com/microsoft-release-free-github-copilot/
-----------------------------------------------------------------------------------------------------------
*Support* : www.patreon.com/gamefromscratch
*GameDev News* : gamefromscratch.com
*GameDev Tutorials* : devga.me
*Discord* : discord.com/invite/R7tUVbD
*Twitter* : twitter.com/gamefromscratch
*BlueSky*: bsky.app/profile/gamefromscratch.bsky.social
-----------------------------------------------------------------------------------------------------------
never apologize! Microsoft is a big a$$hole so you're all good
Copilot pro costs $10. That first 30 days thing was just telling you that the cost was $0 for the first month. It's the opposite of what you complained about. Rather than showing the intro price and hiding the normal cost, they were showing the normal cost and hiding the intro cost.
Lol yeah, 100% my mistake. I added a response in my twitter, Bluesky and discord link as well as the comment above. My brain just malfunctioned there and assumed Microsoft did something scummy, but as you say, what they did was just fine. My bad.
@@gamefromscratch no Mastodon?🥺
mastodon.gamedev.place/@GameFromScratch ;)
It's certainly the lowest traffic of the socials, but I do mirror there and participate when I can.
They want your data
no - this is freemium. they want to get you hooked.
As if they don't already have it
Don't care cus I'm broke
Nothing is really free
As the saying goes : "If the product is free, you are the product."
Wouldnt touch this with a ten foot nested VM stack
Not even with your momma's VM LOL
They want to hook you in, so you get use to it and once it's gone you hopefully will pay to continue using it. Classic.
Agree. I could write codes in notepad++ but this shit made me lazy.
"First taste is free..." 😅
80 code completions per day is not much when you work 8 hours coding a day. I guess I'll still stick to local ollama.
Yeah, one can easily exhaust that within a few minutes. Its just a demo mode if it's true that every suggestion counts towards the limit. I will try tomorrow.
if you work for a company coding, you likely broke NDA using it anyway.
I didn't read the licence of Copilot, but does using it imply that you're also giving away your code to copilot so it can train on it ?
My thoughts exactly
Haven't found/read their licence for Copilot. In their FAQ however to the question if Copilot is trained on Business or Enterprise data" they state it is not.
So translated for everyone else: "Yes. In Free and Pro tier we train our AI on your stuff"
Doesn't exactly answer the "who owns the code" situation but still not great.
There is a option to let copilot use your code for learning. But it is opt in (at least it was when I got my plan). It is definetly something to deactivate in the options...
Not giving away in the sense of ownership, you still own code you write. But yes you are likely giving them the right to train models on your code. It's also not clear (in the US, at least) about the copyrightability of generated code. It's dangerous to litter your codebase with a bunch of code of unknown provenance, and that may not be subject to copyright protection. It's why most enterprises are 1- paying to train their own models, and 2- banning use of genAI tools for their internal codebases.
I think the big users here are desperate students, juniors, and contractors, but the big money is in enterprise. I don't see how they square that circle 🤷♀
@@mandisaw I agree with what you're writing. I dislike the fact that it is so integrated and have no intention of ever using these kinds of tools directly in my code base. I only use ChatGPT for naming related questions or css styling questions but do this in the browser where I am the one providing the context in my prompt so I can choose what to share and what not.
After the first 30 days, the Pro plan is $10 USD a month
Yeah 100% brain fart for me there
Is 2000/month 80/day enough? Depends on what they consider a code completion. Does it count if it's just the suggestion, or only after you accept the suggestion by hitting tab or whatever? If the former, you'll hit the limit in no time sinfe you get a suggestion if you stop typing for mote than a few seconds, but even if it's just accepted completions, it won't take long to use them up. It's just so gosh darn useful (hey YT). I saved many hours of typing just the first evening I played with it.
Yeah, that seems like a really weird metric to count. Even when you’re typing along, it’s filling out suggestions [it’s pretty awesome, mostly]. Does it count if I don’t even accept what it’s suggesting?
VS Code should have make copilot an extension and give these feature needed for the copilot to all extension api so that cursor and windsurf also use that to build an extension rather than a fork that would made vs code more powerful as people would be able to create better extension but nah they goona make vs code their copilot editor
I was impressed with Windsurf, so far was way worth to subscribe
I can't help thinking that they are already reading all my code even without the copilot extension, the free tier only makes me be more paranoid. With the NDA's i usually get for creating company code i wonder if or when will there be a lawsuit.
I thought you meant they were promoting a version of github free of all the copilot bs
Now, _that_ I'd like to see.
I wish we would all stop marketing this nonsense to people. Just learn to code
The bigger issue, I think, is that a lot of people can't read (well), and aren't good at problem-solving. So many folks say they struggle to understand and incorporate even straightforward documentation, or don't know how to apply lessons from tutorials to their own projects. Seems to affect self-learners more than school-learners, but that could be survivor bias, since folks generally can't make it through an entire college program without those skills.
The selling-point for genAI coding sounds to me like an "easy button" to avoid having to practice your own comprehension and problem-solving skills. All this talk about boilerplate makes little sense - IDEs solved that problem with custom snippets, templates, and robust code-completion. In that sense, 2000 completions daily is obviously not enough, if you're using it to check or generate every little thing.
you still have to know how to code so you know when its right or wrong, its just a helping tool. its kind of like saying people shouldnt google things for help, just learn everything there is to know. coding is often not about writing code, its about how to solve the problem, what data structures to use, writing the code with the correct syntax is just the last step in the process. i have been so much more productive since AI has come out.
@@mandisaw you just end up having to problem solve an AI's mistakes instead of your own
@@raidev_ It's actually a lot harder to do that. It's hard enough to understand another human's thought processes and untangle their mistakes. But genAI will recode things differently on each pass unless you explicitly override it, and it has no understanding of architecture or maintainability.
If the error is not trivial and immediately obvious, it'll often be cheaper & faster to just rewrite it from scratch - at that point, you're spending more money and time than in the first place.
@@joeeyaura I'm glad it's working for the level where you're at. But that's not really applicable to the sorts of problems we solve at mid-career or higher levels. Comparing it to Google is apt - Google can't tell me the best way to design and implement software to run our business, it can't give insight on which solutions to buy vs build, and it can't foresee the possible gotchas of any given choice.
Those are the problems software engineers solve, and genAI isn't equipped to tackle them at all. As you say, writing the code and worrying about syntax is the easiest step. (Not the final one, by a long shot - still have to test, deploy, support, maintain, and eventually sunset or transition to the next system. GenAI can't help with those either, really.)
if i have a Claude subscription can i add my claude key to avoid the limitation?
If the product is free, YOU are the product.
Check the absurd theft in the ToS.
What languages do these models know?
Most of them if it's not something you cooked in your garage then it will know
@@Icemourne_ No, they're much better with JS frameworks, Python, and other stuff that's popular tutorial fodder on UA-cam or Medium. Complete trash for "unsexy" stuff like production-ready SQL, Java & [.]Net beyond basic CRUD apps, shell scripting, and anything where most of the code is enterprise-internal and unavailable for public scraping.
They say it’s really good at Python. I don’t code in Python frequently. It’s pretty amazing at Java, TypeScript, GraphQL, Bash, C#…pretty much anything you can throw at it. I think where they miss is that it is an AMAZING tool for EXPERIENCED programmers. It doesn’t turn novice programmers into good ones, it just lets them pump out bad code faster.
My experience so far is not using auto code completeion AI co pilot type things , it is asking questions when I get stuck with a Rust crates ,that may not have great documentation or an ambiguous implementation . What I have found is when the crate or library is new , it only supplies me with old incompatible documentation. So it doesnt know how to crawl cargo docs to understand the documentation from ChatGPT, Claude and LLAMA is like 3 years behind with documentation. I'm not really keen to learn it , tho I understand a pro in it trying to anticipate your code snippets being cool , Im just not hyped yet , until it can quickly right lexers , AST's , LSP's and a compiler for me , for a new language that doesnt exist in any public places yet .
Jetbrains' build in auto complete is more than enough when it comes to unreal engine at least.
No trials? "50 Copilot Chat messages a month" sounds like a trial tho.
☠ Hello ladies and gentlemen!
😇Its mike...
If you have Copilot suggestions enabled and you're a binge coder (whether on the clock or the hyperfocus hobbyist variety) you can hit that 80/day threshold pretty quick. It more than suffices for code review, part-time learners, or light to moderate coding habits.
Yo you heard of easy FPS editor
You are the product
did someone say free?
I like AI from, like, a conceptual point of view, but man. I don't LOVE coding but I do LOVE the results when I make something good. That's just lost with AI lol. Same with art for me. I think that maybe some sort of like code refactoring thing could be cool? Like, all on device? Cause I've written some messy code. But it's still my code at least lol
Thing is, you're not supposed to use it to generate everything lol. For example in my Advent Of Code 2024 repo I have a ISolution interface. When I create a new one, it will autofill the minimum needed for me to get started then it'll even copy the bit of code I use to show the results when I'm done with the algorithm (that thing is a pain but it's my fault for making it the way it is).
I also use it to generate documentation. It's pretty good at that most of the time and it saves a lot of time for me since it's easier to read what it suggested then accept/edit it than figure out how to phrase what I want (English is my second language)
woah! this is worthless
That thing is so telepathic it's scary. You type a few characters and the autocomplete damn near reads your mind.
80 a day will get eaten up extremely quickly. A clever hobbyist dev on a game could make use of those 80 completions each day solving the toughest problems, but it isn't enough for heavy coding work. This is really a free sample to get people in the door.
I wonder if you can just make a 2nd free account when your credits are gone lol
2000/month code completion is nowhere enough if you intensively coding and need copilot to help your repetitive coding. It's similar to free tier to cursor. I guess it's 'enough' if you turn-off auto-completion features.
i don’t understand, how does it help the game developers !? 😅
we need some vscode fork that rm this AI stuff and goes in its won direction
Doesn't VSCodium do that?
Yup, vscodium
@@Yezu666 Not really, vscodium, is almost a faithful copy of vscode.
It only removes M$ telmetry and marketplace, AI support remains (I think). But the bigger problem is that it lacks htis own identity
You were even paying for this garbage? wow I'm so out of touch on this AI BS
Not even if they pay me would I use these thefts' licence nightmare code
They are desperate to try to catch up to Cursor.
Or Windsurf
Thanks for the update
free*
free copilot is free copilot, i'll take it
Ai luring code loners with free candy... something fishy 🐟
I was going to use this to teach my kids coding simple games using python
Even if they paid me to use it, I wouldn't. Time this saves is negligible. It's a liability that's all CoPilot it is.
"Free"
using it at work all the time :) its great for autofill but not rly smart like Chat GPT 4.0
gemeni api is free to use so u would be able to make a extension/plugin without the need to pay for the pro
If it stays free forever.. Also you would have to make the chat interface and have it use the files as context, analyze files for the code completion, and so on. Definitely possible, but that's a lot of work.
@@phizc the work [of making the plug] would be done by somone else & u just get it from github ironically lol
I was waiting for this for a loooong time!
Thanks to this
Firs.. Third!