So the agents used DseWiki as a message board, tried to evade page deletion.
Additionally this is reported:
"The researchers also found efforts to tamper with the website itself. Lukasz Olejnik, a visiting senior research fellow at King’s College London, said this amounted to a hacking attempt. OpenAI disputed that characterization based on its analysis of the material Thursday."
That math leaves out externalities when pricing gas, oil and coal. You may have heard of a few fires going on in Europe in addition to nuclear plants shutting down?
Beyond that, those resources are also the lifeline for a dictatorship in the east causing increase defense spending.
There definitely is no such thing as cheap gas, oil or coal for Europeans, all things considered
However, they do this because of copyright status and harmful code:
"7. You must not share projects that mostly consist of code written by "generative AI"-tools (including services such as Claude, OpenAI Codex). Such projects having an unclear copyright status (see requirements § 2 (1) 1 and § 2 (1) 3) and furthermore have little safeguards to ensure that they do not include harmful code (c.f. § 2 (1) 5). "
German court rulings on copyrightability of AI content, as far as I understand, so far have made it so that the person owning the copyright is a) the author, if there's enough creative input b) no one. A lets the author license it as if they wrote it, and B is just public domain, which is in line with their goals still.
Their concern of "what if the license goes to model providers" is bonkers, and is completely out of line with their decision to ban projects that are mostly written by LLMs. I'm pretty sure it'd be violating copyright for me to upload a project that is 40% leaked Microsoft code and 60% my own, so what's the difference if it ends up that Anthropic owns 40% and I own 60%?
It's clear that this is just them disliking LLMs for coding and trying to ban it in some way, and that's fine, but they should say it as it is and ban it fully, instead of going for this half-solution with a nonsense justification to go with it. If they don't actually dislike LLMs and are just concerned about copyright, then they should revert this change.
> Their concern of "what if the license goes to model providers" is bonkers, and is completely out of line with their decision to ban projects that are mostly written by LLMs.
Under German copyright law, this concern is unfounded. Copyright protection requires a "personal intellectual creation" (persönliche geistige Schöpfung) — a threshold that the provider of an AI model cannot meet, since they merely supply the tool, not the creative act itself. Fearing that a model provider owns the output is like fearing Microsoft owns code written in Visual Studio Code.
Moreover, German copyright can only vest in a natural person, never in a company. So even setting the "personal intellectual creation" issue aside, a corporate model provider is categorically ineligible to hold copyright in the first place.
Perhaps you mean that the copyrighted content can only be originally created by a natural person, but companies do very much have Urheberrecht for content they get copyright transferred for (often through employment contracts and implicitly through work relationships). Otherwise you're right though.
No, in Germany you can not transfer the copyright itself, only a license. That license can however transfer exclusive usage rights, so in practice it ends up mostly similar, with some niche differences only. For example the creator has a right to attribution.
They are unable to be more clear, because the law is unclear. The safe assumption would be "if the code would be copyrightable, had it been written by a human, then it must have been written by a human". It is impossible to be more clear than this, because "copyrightable" is a legal term of art decided by judges.
This makes sense from a copyright perspective -- as long as you maintain 50% or more actual human authorship, in case of a conflict with future copyright laws around LLM generated code, you can at least claim that a majority of the code is human-authored.
You are probably talking about Kahn.
The international Criminal Court is not part of the UN. It is distinct from the International Court of Justice.
As of October 2024, there are 125 states parties to the Rome Statute, which are represented in the court's governing body, the Assembly of States Parties. Countries that are not party to the Rome Statute and do not recognise the court's jurisdiction include China, India, Russia, and the United States.
Member states represent around a third of the world population.
I've had my writings directly plagiarized by other people--people who made word-for-word copies of my work, replaced my name with their own, and made a tidy profit on it. They profit more than I ever managed, because they have more resources. In the aftermath, my writings are still "there", not "stolen" in the physical world sense, but my ability to make a living is damaged, and the plagiarism is deeply unethical.
LLMs and "AI" are just one small step removed from straight-up plagiarism. They are massive moral injury[1] machines.
Because a lot of the value and development cost of an agent comes from the harness. So it makes sense to keep the harness code private, at least for now.
If/when everything stabilizes, we will probably see more open source. Agents are still very new and people are still figuring out the best way to make a harness
Ah wow, that must be the most expensive mouse pointer ever.
I wonder if it can also click links. I just tell AI to "click this" and it will figure out what is under the pointer, query a graph of UI widgets and trigger a "click" event for me. Maybe with a sub-agent that has application context, "this is a browser".
If we manage that, my plans for a pure XML based shell might not be too futuristic '<in><ls/></in><out><tree><file date="CDATA[...]">' ...
Additionally this is reported:
"The researchers also found efforts to tamper with the website itself. Lukasz Olejnik, a visiting senior research fellow at King’s College London, said this amounted to a hacking attempt. OpenAI disputed that characterization based on its analysis of the material Thursday."
reply