I forget a lot of things, simple dumb stuff like type conversions or specific keywords spelling. Copilot takes care of 99% of that so I can focus on my higher level spec.
If anything sometimes it's too agressive. I start typing a word and it's already building up the rest of the application in a different direction...
Also, that little "building confidence in yourself" rider that you added suggests that you think the OP doesn't have confidence in themselves. Careful about those assumptions; in this case it comes across as a little patronizing.
They certainly do have confidence, but it doesn't hurt building upon it. I don't think confidence is a discrete 0 or 1 variable. It's a continuous variable, from 0 to infinity.
By the way, I made a question, not a statement. I welcome the argument on whether it's worth the time to memorize the stuff Autopilot can autocomplete. That's something to measure and debate.
> Careful about those assumptions; in this case it comes across as a little patronizing.
Thanks for the heads up. This is important. But I didn't make any assumption. It seems maybe you made an assumption about me making an assumption?
If that's the case, don't worry as I did not feel patronized.
It's not a silver bullet / holy grail / MacGuffin / etc.
There's a community of people obsessed with spaced repetition. None of them seem to have accomplished spectacular feats of learning. There's a good reason for that.
(The flip side, however, is that many people who have accomplished spectacular feats of learning DO often use spaced repetition, but among a broader repertoire).
> None of them seem to have accomplished spectacular feats of learning
> many people who have accomplished spectacular feats of learning DO often use spaced repetition
---
Second note:
> Spaced repetition is a good tool for learning vocabulary.
This might be a misconception? The first ever published paper about spaced repetition was about vocabulary.
But research has shown its merits are valid for many other types of knowledge.
By the way, memorizing coding vocabulary is arguably similar to human language vocabulary.
It's not contradictory.
I'll give an example: Landmines are a good tool for slowing an enemy army. However, if your military consists of *just* landmines, it won't be very effective. That doesn't make landmines a bad tool. Indeed, even a super-weapon, like the first jet fighter, won't win a war if it's your *only* tool.
Learning -- even a language -- is a complex process, and you need many tools. Spaced repetition is awesome for factoids. If you want to learn a language, you need to memorize vocabulary. SR is great for that. If you add 5-15 minutes of spaced repetition to a good language program, it will help a lot. If that's where you spend a majority of your time, you'll learn very little. However, SR won't help you practice a broad range of skills around listening, speaking, understanding communication styles, or quite a few other things.
Ditto with physics and math. If you know equations, it accelerates everything else. However, the bulk of the knowledge isn't factual or procedural; it's conceptual. Simply memorizing formulas won't help you. On the other hand, in most cases, once you learn conceptual knowledge in physics, you never forget it.
"Coding vocabulary" isn't in my top-50 problem with junior developers. Naming variables, algorithms, systems design, etc. are. Most of those don't align to SR. I'd take a programmer who spends 8 hours coding over one who spends 8 hours memorizing library calls in an SR system.
Note 2:
The spacing effect is /somewhat/ broadly applicable (but far from universal), but spaced repetition specifically is only helpful for factoid-style knowledge. You can look over the different classifications of knowledge, skills, and abilities (factual, conceptual, procedural, declarative, etc.).
It’s not always right but it’s right enough it’s a big timesaver.
That assumes I once had the knowledge of what is "correct".
I just don't quite remember it right now. Then I rely on Autopilot to complete it.
Sometimes I may not feel sure enough to judge whether Copilot is right. I'll need to dig the documentation anyways.
Other times I'll feel sure. But in how many of those I'll be wrong? And what will be the consequences?
> what will be the consequences?
for me, usually a failed build or unit test :P low stakes stuff
I've had this experience too. Usually it's meh, but at one point it wrote an ENTIRE function by itself and it was correct. IT WAS CORRECT! And it wasn't some dumb boilerplate initialization either, it was actual logic with some loops. The context awareness with it is off the charts sometimes.
Regardless I find that while it's good for the generic python stuff I do in my free time, for the stuff I'm actually paid for? Basically useless since it's too niche. So not exactly worth the investment for me.
I find I spend my time reviewing Copilot suggestions (which are mostly wrong) rather than thinking about code and actually doing the work.
When you disable auto complete, it can still be fired via a keyboard shortcut, which is how I use it.
https://github.com/github/copilot-docs/blob/main/docs/jetbra...
"editor.inlineSuggest.enabled": false
There's a ton of developer effort that went into Copilot and those devs should be paid fairly. But the majority of what fuels Copilot is the millions of lines of open source code submitted every day.
I think I'd feel a lot better about it if they committed a good chunk of that money back into the same open source communities they depend on. Otherwise its a parasitic (or at least not fully consensual) relationship
GitHub hosting open source communities is a way of giving back.
Hosting code "for free" is part of its business model. It's not a way of "giving back"
Github Advanced Security costs me $200+/month, but if I make my repository public, I get all of that for free.
Certainly feels like a way of giving back.
Of course it benefits them too, but it doesn’t have to be purely altruistic to be a net positive.
They could even change github so it benefits their code crawling.
No. They're not. They're advertising that they are.
They are providing it to a very small set of high-profile OSS maintainers some opaque algorithm picked out. Having high-profile adopters is just good business.
One fundamental aspect of being open source is not limiting the purposes of use. If we now say that "code-generation AI training" is not allowed without prior approval (in addition to the license itself), then it's not open source anymore...
If I write AGPL code, and co-pilot scans it and makes a very similar program to it for a FAANG, who then proceeds to compete with my open-source tool by using the creative ideas generated there-in, but with a proprietary tool, that's not very fair. That's why I chose the license I did.
FAANG is more than welcome (indeed, encouraged) to use my code for any purpose permitted under the license. That includes everything except making it proprietary.
I've tried running copilot with the starting lines of my code. It generated code with identical creative ideas. It was the equivalent of taking Star Trek, and generating a new movie with the same plot line, but with names changed. That's not legal.
My code was specific enough that this wasn't just chance or other similar code. I work in a pretty narrow domain.
I did use copilot for coding myself, and a lot of what it generated was unique. But it is also a good paraphrasing tool. Running a movie script backwards and forwards through Google Translate to get different phrasing, and then swapping out new names, does not a new movie make. Ditto here.
I'm not defending Microsoft's market tactics, for obvious reasons, but we do have to consider that anyone can publish whatever insignificant code as OSS and become an "OSS maintainer" out of nothing.
They have to draw the line somewhere. Nowhere they draw will make everyone happy.
But! I think the savings are even bigger because there is no context switch. If I have my browser open I might find myself going to hackernews, checking my email, looking at my stackoverflow notifications, browsing twitter - whatever. Copilot is not only faster, it keeps my focus on the code without giving me a chance to get distracted. In some sense, for this example, Copilot saved me 5-10 seconds by not needing to Google something. In another sense, it might have saved me an hour because I didn't decide to just check something on twitter while I had my browser open.
I find that if my brain is willing to be distracted, it’s made some sort of calculation saying that the cost of being distracted isn’t going to have a significant bearing on deliverables…and I’m pretty sure I’m no better or worse at estimations than anyone else.
Like, writing a really boring unit test might only take 60 seconds, but if Copilot can do it for me (even if I have to quickly scan it for correctness) that saves me… well something other than 60 seconds. It sure feels like a big deal.
Boring and repetitive but of infinite value if you can detect early that your deployment broke something.
If a boring/boilerplate unit test like this can deliver value, other b/b unit tests are probably going to similar impacts and hence, saying they all “decrease” value is reductionist.
Yeah, and it should come almost automatically from the tooling. If you are writting this, you have a tooling problem.
Goal of library/language/framework designers: limit boilerplate and unnecessary code
Goal of tool/IDE designers: make it easy to not spend time on boilerplate and unnecessary code.
Both sides have built in limitations that will keep them from completely solving the problem. Terse, highly DRY code tends to also be highly abstracted and hard to read, with a lot of implicit behavior. On the other side, large amounts of generated code lead to tool lockin and cruft accumulation.
If co-pilot’s autogenerated test cases can help prevent this head smacking, it will have proved that basic/boilerplate code was valuable.
The site checker included in the tooling is just a more mature version of the boilerplate unit test co-pilot gives us.
Btw, I have no skin in the game -never used co-pilot…just surprised that HN commenters can be so dismissive of wanting to get the basics right - like having some test coverage.
It wasn't much help in designing/implementing classes in Java or .NET, but when it came to implementing unit tests it practically wrote everything after i named the class and designated it as a test. It was able to extract all the different methods from the classes being implemented, and create appropriate unit tests based on that.
Now, it was school homework, so not representative of a complex business application, but if i can just handle the basics/boilerplate, it would be worth it.
Assuming a (European) work week of 37.5 hours per week, $10 comes down to $0.06 per working hour, and if it can just save me 5 minutes of work every day it will be worth it.
For example, I just had to convert some OCaml code to Rust. I wrote the first few conversions, and then I would just paste the OCaml code in a comment, and it would auto-complete the equivalent code in Rust. I just had to check that it was correct, which it was most of the time, rinse and repeat and wow. One would have to be blind to not find copilot impressive, really, it's the future.
Yeah, it makes mistakes, sometimes it shows you i.e. the most common way to do something, even if that way has a bug in it.
Yes, sometimes it writes a complete blunder.
And yes again, sometimes there are very subtle logical mistakes in the code it proposes.
But overall? It's been *great*! Definitely worth the 10 bucks a month (especially with a developer salary). :insert shut up and take my money gif:
It's excellent for quickly writing slightly repetitive test cases; it's great as an autocomplete on steroids that completes entire lines + fills in all arguments, instead of just a single identifier; it's great for quickly writing nice contextual error messages (especially useful for Go developers and the constant errors.Wrap, Copilot is really good at writing meaningful error messages there); and it's also great for technical documentation, as it's able to autocomplete markdown (and it does it surprisingly well).
Overall, I definitely wouldn't want to go back to writing code without it. It just takes care of most of the mundane and obvious code for you, so you can take care of the interesting bits. It's like having the stereotypical "intern" as an associate built-in to your editor.
And sometimes, fairly rarely, but it happens, it's just surprising how good of a suggestion it can make.
It's also ridiculously flexible. When I start writing graphs in ASCII (cause I'm just quickly writing something down in a scratch file) it'll actually understand what I'm doing and start autocompleting textual nodes in that ASCII graph.