What's anthropic going to do, sue them? Insist in court that LLM recreating copyrighted code is a violation of copyright???
Ɲєιƚ Ƃιɍƌ 凤
jonny (good kind)_I_VERIFIED_THIS_IS_NOT_CODE_OR_FILEPATHS and the docstring explains this is "to confirm you've verified the message contains no sensitive data." Like the LLM resorts to naming its variables with prompt text to remind it to not leak data while writing its code, which, of course, it ignores and prints the error directly.
So the reason that Claude code is capable of outputting valid json is because if the prompt text suggests it should be JSON then it enters a special loop in the main query engine that just validates it against JSON schema (it looks like the schema just validates that something in fact and object and its keys are strings) and then feeds the data with the error message back into itself until it is valid JSON or a retry limit is reached.
This code is so eye wateringly spaghetti so I am still trying to see if this is true, but this seems to be how it not only returns json to the user, but how it handles all LLM-to-JSON, including internal output from its tools. There appears to be an unconditional hook where if the JSON output tool is present in the session config at all, then all tool calls must be followed by the "force into JSON" loop.
If that's true, that's just mind blowingly expensive
edit: please note that unless I say otherwise all evaluations here are just from my skimming through the code on my phone and have not been validated in any way that should cause you to be upset with me for impugning the good name of anthropic
edit2: this is both much worse and not as bad as i thought on first read - https://neuromatch.social/@jonny/116326861737478342
Attached: 3 images OK i can't focus on work and keep looking at this repo. So after every "subagent" runs, claude code creates *another* "agent" to check on whether the first "agent" did the thing it was supposed to. I don't know about you but i smell a bit of a problem, if you can't trust whether one "agent" with a very big fancy model did something, how in the fuck are you supposed to trust another "agent" running on the smallest crappiest model? That's not the funny part, that's obvious and fundamental to the entire show here. HOWEVER RECALL [the above JSON Schema Verification thing](https://neuromatch.social/@jonny/116325123136895805) that is unconditionally added onto the end of every round of LLM calls. the mechanism for adding that hook is... JUST FUCKING ASKING THE MODEL TO CALL THAT TOOL. second pic is registering a hook s.t. "after some stop state happens, if there isn't a message indicating that we have successfully called the JSON validation thing, prompt the model saying "you must call the json validation thing" this shit sucks so bad they can't even ***CALL THEIR OWN CODE FROM INSIDE THEIR OWN CODE.*** Look at the comment on pic 3 - "e.g. agent finished without calling structured output tool" - that's common enough that they have a whole goddamn error category for it, and the way it's handled is by just pretending the job was cancelled and nothing happened.
USER_TYPE=ant that seems to be read directly off the node env var accessor. No idea how much of that would be serverside verified but boy is that sloppy. They are often labeled in comments as "anthropic only" or "internal only," so the intention to gate from external users is clear lol
@whitequark @jonny Apparently some have had DMCA takedowns filed against them, so here are a couple links still working as of this writing:
@whitequark @jonny Additionally:
Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflo...
@martenson
@IvanDSM
Sorry I removed the link to that repo because i thought it was just the unpacked source, but it turns out they're trying to convert attention to the repo into their own product.
Here's another blogpost, there are a million, I don't claim this one is particularly good but at least it seems to come attached to the actual source
https://kuber.studio/blog/AI/Claude-Code's-Entire-Source-Code-Got-Leaked-via-a-Sourcemap-in-npm,-Let's-Talk-About-it
@martenson @IvanDSM @jonny Okay, but what repo? We're operating off a Fedi trademark vaguepost.
Edit: found an article with links: https://dev.to/gabrielanhaia/claude-codes-entire-source-code-was-just-leaked-via-npm-source-maps-heres-whats-inside-cjo
@traecer @jonny @srvanderplas I think this is unrelated to the question anyways. If AI-generated code is considered a derivative work of some GPLed code, then GPL does apply to it. No need to think about linking code or dependencies.
And as you say, courts seem to generally consider AI-generated code as public domain, so I would guess GPL is pretty much unenforcable in this context.
I am not a lawyer either though xD
@ell1e @traecer @jonny @srvanderplas The `isEven` example is really funny. xD
Yeah, I mean, if the AI-generated code is a blatant copy of some code in the training data, I don't think there will be much of a doubt that it's a copyright violation.
But when the generated code starts diverging from the source, I think it's a more legally gray area. I would also consider it a copyright violation, but it's not me who needs to be convinced about this. It's judges. And I haven't seen them agree yet.
@LaquinArt @traecer @jonny @srvanderplas The person in the video is a lawyer, just to let you know. Also there's this: https://www.twobirds.com/en/insights/2025/landmark-ruling-of-the-munich-regional-court-(gema-v-openai)-on-copyright-and-ai-training
My main intention was to give you resources that may inform you about how settled (or not settled) what you previously said really is. Not that I know though, since I'm not a lawyer. This isn't legal advice.
But there are plenty of sources saying that LLMs directly copying seems to be a regular event, not a rarity: https://dl.acm.org/doi/10.1145/3543507.3583199
@LaquinArt @jonny @srvanderplas
"I should be able to distribute with any license I want (even a proprietary one)"
Nope, that's the viral nature of the GPL. If you link to GPLed code and intend to distribute your new code, it MUST be GPL as well. That way helps to ensure the FSF's idea of "software freedom" aka "copyleft". This is why the Linux kernel license has an explicit exception to GPLv2 to ensure Linux syscalls can be made by user space code without distributing the user space code under the GPL. (See: https://www.kernel.org/doc/html/latest/process/license-rules.html) It's also one the reasons so many open/free source projects use dual licenses like Perl or Firefox, and more permissive licenses like the Apache 2.0 and MIT licenses are so popular.
What you have described is the Lesser GPL (LGPL) license, and yes, code under the LGPL does NOT require your code to have any particular license, provided you distribute any changes to the original (assuming you intend to distribute the original code at all).
@traecer @jonny @srvanderplas Did you read the syscall exception yourself?
‘This exception is used together with one of the above SPDX-Licenses to mark user space API (uapi) header files so they can be included into non GPL compliant user space application code.’
The exception allows you to *include GPL code*, which I also said triggers GPL.
The mere referencing does not trigger GPL so long as you don't include a GPL work. Otherwise, reimplementing APIs would be illegal. And we know it's not.
@traecer @jonny @srvanderplas Oh, look! The following paragraph makes this even more clear:
‘NOTE! This copyright does *not* cover user programs that use kernel services by normal system calls - this is merely considered normal use of the kernel, and does *not* fall under the heading of "derived work".’
@traecer @jonny @srvanderplas Upon further investigation, I see that the FSF considers that a work dynamically linking a GPLed work is covered by GPL. From what I know, this has never been proven in court, and I don't think it would hold up.
Sure, once you run the program, there exists a combined work that should be subject to GPL. But this combined work is generated by the user and never distributed, so GPL is never triggered.
@ell1e @traecer Strictly speaking, it’s not talking about that: “Fair use” is not a legal concept within that court’s jurisdiction.
EU law allows to ignore any and all copyright for “data mining”, and OpenAI tried to argue that since their business is data mining, they never have to care about copyright to begin with. This particular ruling simply says that if your product reproduces lyrics of an entire song, that isn’t just data mining, it is in fact copying.
So what the court says is: Under current EU law, you’re allowed to copy as much data as you want for *training* your LLM, but that doesn’t mean you’re also allowed to actually provide LLM as a service to the public. (IANAL)
Note that this particular ruling is not legal precedent and it’s already being appealed.
Ethically? Absolutely 100%
Legally? Well, you see, the tech CEOs are very good friends with all three branches of the US government, so not in the USA or Israel anyway.
Andrew
beemoh
✧✦Catherine✦✧
Jamie Gaskins
Toast, Anonymous Fedi Fungus
The Orange Theme
nash
Brodeuse LucileDT
Dan Sugalski
Preston Maness ☭
[HANDMAIDEN] xan
Lars Marowsky-Brée 😷
José Albornoz
wohali
some kind of orange shape
elle
IvanDSM
martenson
bluestarultor
Susan Vanderplas
traecer
Laquin
ell1e coding things
Arne 🌻🌈
Cassandrich
eestileib (she/hers)
mirabilos