Conversation

in the crypto grifting year of our lord 2022, this one is really up there in terms of audacity

750

github will probably get away with this — they have a lot of lawyers on retainer — but legal is not the same thing as ethical

918

if you only act ethically when it is convenient you are not in fact an ethical person

Quote Tweet

milo

@i2talics

· Jun 22

Replying to @ReinH

every rebuttal is just "yes but ... its useful for me. so i dont care"

727

reminder that 1. The MIT license, the most popular license on github, requires attribution 2. Copilot has been shown to reproduce substantial portions of its sources verbatim

152

1,170

what github / microsoft is counting on here is that open source developers do not have enough collective power to do anything to stop this, and sadly they are probably right

753

brb adding this clause to MIT and calling it the "fuck machine learning" or FML License

Quote Tweet

mffwww

@mffwww

· Jun 22

This is going to be settled in court eventually, and the sooner the better. Anyone should be able to apply a license that says “you may not use this to train any machine learning model”. twitter.com/ReinH/status/1…

706

danny "disco" mcClanahan

@hipsterelectron

Replying to

yeah the same way compiling someone else's code from source isn't a derivative work because i'm not including the actual source code in the output

danny "disco" mcClanahan

@hipsterelectron

(THIS IS SARCASM)

Show replies

Replying to

Welcome to the future, old man

Replying to

its okay to distribute a bunch of copyrighted works compressed together as long as the decompressor only accepts vague hints about which one you want to decompress and sometimes mixes them together a bit

127

forealiously

@forealiously

shelby reliability engineer

"eyyy dats a lovely code prompt youse got over dere, would be a shame if some proprietary code were to have a little 'accident' after it"

@shelbyspees

Replying to

honestly though? if I ever need to write a crud API I'm on that shit

Mat Savage (he/him/his)

@Mat__Savage

that’s my secret, all my APIs are crud

Replying to

It actually helps me with accessibility a lot, because I have a serious RSI and can’t type too much any more and it allows me to do “comment driven development”. You still have to know what you need and you have to be able to understand the code, (1/2)

BizTheDeveloper

@lizthedeveloper

Alex Bunardzic, white non-colonizer

But it does not allow non-devs to be devs, because you have to know what you want. I have actually tested this with some highly technical but non-coding grad students and they couldn’t really use it, even with a lot of guidance.

Show replies

Replying to

and

Interesting. So, any time I use someone's code & modify it to my needs, I am doing "code laundering"?

no?

Show replies

Replying to

Only for the time being, until its writing brand new code by itself.

This Tweet was deleted by the Tweet author. Learn more

Show replies

Replying to

I think it is doing far more than recomposing code. It is automating what you would have to do yourself otherwise: Study related code, identify useful patterns, apply to your context. That's more than derivative work. You can test that out by applying it to your self-made domain.

Šimon Tóth

@SimonToth83

If you study related code, identifying useful patterns and applying them to your context, you are violating copyright. Not even a hint of doubt.

Show replies

Replying to

The backlash against copilot is so weird to me. Filters are valuable. Understanding context is valuable. Filtering the incomprehensibly large corpus of open source "code other people wrote" to present an individual the code they need in their own specific context is valuable.

1. it's not built from *just* open source code. 2. Copyleft requires that derivative works are also open source. Copilot does not have such a requirement. Is it legal? Possibly. Does it bypass the intent of people who released their code under a copyleft license? Absolutely.

103

Show replies

Replying to

For better or for worse, it is very arguably a transformative work for copyright purposes — it’s basically just a high-dimensional index of words, which is a similar fact pattern to AG v. Google. Still a terrible look for Microsoft to not release the trained model, though.

it has been shown to reproduce large portions of its sources verbatim

114

Show replies

Replying to

It’s a very fair point to raise the question of what counts as a transformative work, and legally what can you sell and copyright works produced by machine learning. Humans take what they see and learn and produce their own unique works, adding value. Can machines do it too?

arborelia

@arborelia

my keyboard is heavily involved in producing both unique works and derivative ones, but, correctly, we do not assign it any of the credit

Replying to

I'm trying it from a junior perspective. While pairing with senior dev, Copilot makes magic. While I use it, same plugin propose dull words instead of code. My guess is that is much more trained on my work instead of other user's

Replying to

every rebuttal is just "yes but ... its useful for me. so i dont care"

101

Jarrich Van De Voorde

@jarrichvdv

That’s how everyone and everything works now. People are willingly parting with their privacy for the sake of convenience all the time for example

aws

@kcnaija

Automated

Replying to

CrashLoopBackOff Override

and

@socrabot_

What? I thought it was an intelligent machine that could figure out how to write code from knowing the syntax and semantics and best practices based AI rules such as code completion not from an existing code???

@freeone3000