5 Million People Tried Microsoft's AI Coding Tool 'GitHub Copilot' in the Last 3 Months (techcrunch.com) 41
Microsoft's AI coding assistant "GitHub Copilot" has now had 20 million "all-time users," a GitHub spokesperson told TechCrunch.
That means 5 million people have tried out GitHub Copilot for the first time in the last three months — the company reported in April the tool had reached 15 million users.
Microsoft and GitHub don't report how many of these 20 million people have continued to use the AI coding tool on a monthly or daily basis — though those metrics are likely far lower.
Microsoft also reported that GitHub Copilot, which is among the most popular AI coding tools offered today, is used by 90% of the Fortune 100. The product's growth among enterprise customers has also grown about 75% compared to last quarter, according to the company... In 2024, Nadella said GitHub Copilot was a larger business than all of GitHub was when Microsoft acquired it in 2018. In the year since, it seems GitHub Copilot's growth rate has continued in a positive direction.
Microsoft and GitHub don't report how many of these 20 million people have continued to use the AI coding tool on a monthly or daily basis — though those metrics are likely far lower.
Microsoft also reported that GitHub Copilot, which is among the most popular AI coding tools offered today, is used by 90% of the Fortune 100. The product's growth among enterprise customers has also grown about 75% compared to last quarter, according to the company... In 2024, Nadella said GitHub Copilot was a larger business than all of GitHub was when Microsoft acquired it in 2018. In the year since, it seems GitHub Copilot's growth rate has continued in a positive direction.
I am one of those 5 million ... (Score:4, Interesting)
... accidentally. I was trying to set it up to use a local model, but thanks to various dark patterns and opaque language, it ended up uploading code to Gitlab. There was probably minimal harm because it's an open source project anyway, but I was frosted anyway.
Re: (Score:1)
Re: (Score:2)
That's odd. I thought the idea of the new fangled AI thing was that its users did not need any intelligence of their own. Otherwise what's the point. Clearly AI does not work, by your own logic.
Re: And most (Score:1)
So what happens if.. (Score:2)
Has anyone done a deep dive on the EULA yet?
I assume that anyone who even thinks of using this tool has to allow MS to own all copyright, MS can copy and use anything the user creates in the tool in perpetuity, and the author gives up all rights and has to agree to an absolute gag order and mandatory arbitration by a Fox "news" anchor in case of any disagreement.
Microsoft considered harmful...
Re: (Score:2)
Here's a link to the EULA: https://github.com/customer-te... [github.com]
About code ownership:
GitHub does not own Suggestions. You retain ownership of Your Code
Re: (Score:2)
Has anyone done a deep dive on the EULA yet?
My company's legal department did a deep dive into GitHub Copilot's EULA and approved it for development use. At the same time they also said we can't post code into ChatGPT.
What did it do? (Score:3)
I think it would be more interesting to look at what Copilot did for those 20M users.
Was it mostly boiler plate? How much of the generate code was correct? How much of the Copilot code makes into production?
Re:What did it do? (Score:4, Informative)
From my own experience:
> Was it mostly boiler plate?
A good bit of it was, but speeding that up without having to dig through a bunch of templates saves a ton of time.
> How much of the generate code was correct?
A lot of it, but I generally only use it to complete a few lines of code at once, and do the high level thinking on my own. If you use AI that way, it's a great productivity tool.
> How much of the Copilot code makes into production?
A lot of it.
Re: (Score:1)
"A good bit of it was, but speeding that up without having to dig through a bunch of templates saves a ton of time."
That may be true, but it is a far cry from what is claimed of AI coding tools.
"If you use AI that way, it's a great productivity tool."
That may be true, but it is a far cry from what is claimed of AI coding tools.
"A lot of it."
How do you know?
Re: (Score:2)
My experience mirrors yours.
You definitely have to know what you're doing. On one hand, its suggestions are often very helpful. On the other, it does a lot of dumb stuff. For example, ask it to refactor a function, and it will oblige, but sometimes it inserts the new version inside the old one, instead of replacing it. So I have to clean up after it, but it still saved me time.
Re: (Score:2)
You definitely have to know what you're doing.
Hence the rub: if I already know what I'm doing, I don't really need it, even if it saves a bit of time. On the other hand I tried several times to use it for things I knew nothing about (a new communication protocol, a very obscure bash add-on, a complex set of mixed-language libraries never meant to work together, etc...) and I got absolutely nowhere: all the code generated didn't work, wasn't even relevant and a big waste of time overall. I couldn't even coax it to give me even starting points.
Re: (Score:3)
Productivity tools are most effective in the hands of an experienced user.
For example, I like to do my own home renovation projects, and I've done many of them over a period of decades. I love using power tools, not because I "need" them, but because they save me time and make me more productive.
AI is like that for coding. I don't "need" them, but it saves me time and makes me more productive.
For me, even if I use it in an area where I'm not very experienced, like that new API, I do find it extremely helpfu
Re: (Score:1)
Same. I'm a Gen-X C++ developer (and have a 4-digit slashdot ID) and I have it integrated into both VS Code AND neovim.
I don't use prompts to write huge swaths of code. 99.99% of the time, I just let it help me auto-complete a few lines of code at a time. It can keep my mind moving along at a higher pace than normal.
Re: (Score:1)
I was working on a twenty-year-old Java project which had very few unit tests last year. In a couple of days with Copilot I added tests for a couple of hundred classes; for the simple ones it generated the entire test and I just needed to run it, for more complex classes it needed more help since it didn't understand things like not trying to mock final classes. The tests also found a handful of bugs that had been in the code for years so I fixed them.
Last week I had to write a test for old C++ code which m
Re: (Score:2)
I was working on a twenty-year-old Java project which had very few unit tests last year. In a couple of days with Copilot I added tests for a couple of hundred classes; for the simple ones it generated the entire test and I just needed to run it, for more complex classes it needed more help since it didn't understand things like not trying to mock final classes. The tests also found a handful of bugs that had been in the code for years so I fixed them.
What kind of code coverage did you get from your tests?
Re: (Score:2)
Cannot be good or they would have given us that. Instead they push a meaningless "has looked at it" number.
Re: (Score:2)
I think it would be more interesting to look at what Copilot did for those 20M users.
While true, Microsoft has a long tradition of only showing the numbers that make themselves look good..
As an example, when Windows Phone 7 was released, they cited several millions of downloads of their IDE, instead of declaring the total number of phones sold. That is why we are up to WindowsPhone 17 today.
How do you train the AI (Score:1)
I've tried a few of these and I can tell you that the code they generate comes almost exclusively from stack overflow posts. I know because the code usually doesn't work unless it's algorithm code like you're asking it to generate a binary tree implementation or something.
So I end up looking for working code and that gets me to stack overflow every time. I know that the generic documentation for the apis does not work for current llms. They can't really make sense of ref
Re: (Score:2)
Re: (Score:2)
When Stack Overflow goes down?
You do not. But it is all short-term hype and get-rich-quick now. These assholes do not even start to think about strategy or how things ae supposed to work tomorrow.
Re: (Score:2)
How do you train the AI...when Stack Overflow goes down?
They've been trying to hire software engineers to write code as training data for them. I don't know how successful it has been, but I admit I would write code for them if they paid me enough.
100 million people tried influenza (Score:1)
About just the same level of intentionality. Microsoft just forces thist stuff on you. For the o365 copilot they will even charge you for it despite not wanting that shoveled garbage
What They Refuse To Tell You (Score:3)
They refuse to tell you how many of those people continue to use Copilot.
Do 10 million a month continue to use it? That would be impressive.
Did 20 million look at it, say 'meh', and never return? That would not be terribly impressive. Although it would be very very telling.
Re: (Score:2)
Yeah. I mean a few weeks ago on github, I accidentally clicked the preconfigured "generate me a Pong game" prompt and played around with it for all of ten minutes.
So I mean, technically I've tried Copilot . . .
And it was alright, pretty cool even, but come on. It's not as if I was flocking to it, they put it in front of everyone on Github, a thing they bought.
Re: What They Refuse To Tell You (Score:2)
The number might also be somewhat inflated by people who have a "work" account and a "personal" account, or an "employer" account and a "contract client" account, or who have switched jobs within the past year.
Re: (Score:2)
Used for like 5 months but now turned off...
Too often the code generate is not correct - actually analyzing the code and checking it is more effort for me than just writing it myself...
Available but so far not used (Score:2)
So far I've not used it but one day I'm going to check if it can list my appointments that come in by E-mail for the next day, including changing the time zones.
No way (Score:2)
Thing that sales and marking have hyped up and pushed harder than anything else since Windows 95 gets some clicks. People around the world are surprised that the clicks are low.
More news at 11
Fancy Autocomplete (Score:4, Interesting)
These "AI" tools are simply the next level of "autocomplete"
We started with basics like tab completion of symbols
Eventually moving to full code analysis to figure out what symbols could be tab completed
Then we moved on to function/method parameter completion
From there we got code sniplets that an IDE could inject into our code
And now we have those sniplets being dynamically modified based on parameters that we're giving it
Its never been complete, its never been perfect, but if we look at "AI" in the way I just described and use it accordingly, it most certainly does speed up development.
I recently used ChatGPT to write some code for an RP2040 to create an emulated network card for the Apple IIe. It knew nothing of the actual protocol or bus, but I was able to feed it enough information on the spec to get about 80-90% there. From that, I just filled in the rest of the code to ensure it worked correctly.
For a notice, AI most certainly can hamper the troubleshooting and learning aspects of programming. For a seasoned professional, it definitely can speed up the process of producing specific types of library code.
It really isn't at the point where it is making full applications, not even close. But for algos/libraries it is doing a pretty damn good job right now.
Re: (Score:2)
Re: Fancy Autocomplete (Score:2)
Re: (Score:2)
Looks like a lie by misdirection (Score:2)
Apparently, the numbers on actual use are pretty bad, so they now go for "Looked at it!", which is worse than meaningless.
Tried and gave up... (Score:2)
Too often the answer was long and incorrect and I had to retype it while the copilot kept generating other long incorrect answers...