Hacker News Reader

Comment by markphip | Hacker News Reader

HN commenters cannot seem to decide if Copilot is not even worth $10 because what it can do is so basic and trivial or if it is somehow the "end of open source" because it can do so much.

The latter reaction seems hyperbolic. I have not used copilot yet so cannot comment on whether it is worth $10. It seems fairly useful but from what I have seen in demos it feels like fair use.

Open source code is open. It has likely been used to train or influence all sorts of static analysis, formatting and vulnerability tools. Are those also a problem?

Hopefully some court challenges will happen and we can get an answer as to the legality. Whenever things like this come up though, I wonder why people contribute to open source in the first place if they are going to be so bothered by their code being used by others. We all know this, or ought to, when we start contributing. There are companies that are going to make money from our contributions.

View on HN

My knee jerk reaction is outrage, but the more I think about it, the more I realise the only realistic audience for a tool like this is those who are learning or just not very competent yet. If I'm happy to waste my time answering people on Stack overflow, then why does it outrage me so much when a machine takes my knowledge and uses it to help beginners on my behalf? If co-pilot was a human that read the entirety of github and then helped people code for $10/month I wouldn't have any qualms, I'd probably commend them. The reality is copilot isn't capable of creating any real IP, it's just building off the very same foundations of open-source that I have also built my career off. I'm conflicted.

That said, I do believe it's outrageous there is no way to opt-out. I know that's very unpractical given the resources required to train the model, but presumably at some point it will be re-trained and there should be the ability to opt-out of future training cycles.

The longer the product exists, the less it will exist only for the entry level developer. It's naive to assume this is its final form.

I guess we should save our outrage ("the end of open source!") for some further developments in the space then?

Seems like we haven't even reached the slope yet, much less slipped on it.

i'm an outrage angel investor

I work on the ML field so my understanding of the product is not based on naivety. It's an NLP model so all it's doing is learning what "words" (or in this case syntax, function calls etc.) go together well. It has zero understanding of what each function call does and hence is incapable of doing anything beyond a glorified, contextually aware, copy & paste. It certainly wouldn't be able to create new IP without heavy influence from the developer. Writing the actual code is the least difficult part of software development to anyone above entry level.

these are great points. perhaps the question i should have asked was:

is this the death of open source licenses?

Other than the SCO Unix/Linux suit, I cannot recall OSS license discussions around a few lines of code. It usually seems to be about direct usage of OSS libraries or applications. That is why I cannot get too worked up about Copilot yet. Just practically speaking, it seems like the scope of what it would do is pretty minor.

So my answer to this question, would be No.

The problem I have with it isn't the few lines of code it generates, its the ingestion of the millions of repos from millions of devs, creating an AI model from that free code, and then charging for the product. I'm already a paying customer, I should get free access this thing that was built on free code and free labor that gives absolutely no attribution to the original authors.

my side concern (which may be overblown) is that this is the first step to completely replacing programmers and consolidating finances and power even further.

Edit:

additionally, I'm concerned about the "Death" of open source as in: if Microsoft can violate these licenses, what trust remains in the licenses? If people put restrictive licenses on their open source libraries, and now they know these can be violated, will they continue to support open source? And if so, what is the point of the license to begin with?

IANAL, but at this time, I do not see how any of the licenses have been violated. I guess we will learn more if someone steps up to challenge that. That seems to be the inflection point in your argument that makes it hard to discuss where you want to take this.

Going back to an earlier point, we do not know all of the analysis that has been done, whether academic or commercial, by examining open source code. Whether it is on issues of code quality or style, velocity, fix rates etc. Are all of those things also violating the license if that analysis leads to commercial success? What of GitHub itself and how it has improved or based features on what it learns from open source projects and how they use the service?

At this time, from what I have seen so far, I fail to see what is so different or special about Copilot. It did not illegally obtain the source code it used to train the model. Whether using the code to train the model is illegal itself, is another question but if it is then what other uses of the source code are illegal?

Reply on news.ycombinator.com