Slop is Plagiarism
2 days ago: "LLMs Are Not Much More Than Plagiarism Engines"
Some months ago Andy explained what Anthropic was doing with copyrights (popularising the excuses by manipulating the language we use to communicate the issues and laws) and how truly evil this company is. If you did not read Andy's analysis, you should. Anthropic isn't "woke", it's one of the most evil so-called 'hey hi' companies out there. It's very, very evil. It also lies a lot, as Akira has noted (false claims, marketing based on lies).
"Anthropic's products are not secret," an associate has said about the post quoting Akira, "they cost too much to operate, especially since the company *loses* a lot of money on *each* query. Paying customers are not only few and far between, they also can't fill the gap."
The paid-for press (paid to proliferate hype about slop) tells us Anthropic is "worth" almost 1,000 billion dollars, but the company isn't even profitable. It's just losing a ton of money; companies start banning or restricting Anthropic "services" (disservices) as costs go nuts, they spin out of control. That's aside from legal liability issues.
On the Web, we've noticed somewhat of a decline in slopfarms since last year. In Google News we saw this thing yesterday:
It is slop.
But we saw no other example, hence we shelved "Slopwatch". People are getting fed up. Metrics and analytics show decline in use of LLMs. That's despite some conceited bosses attempting to impose LLMs on staff (in the face of resistance and pushback).
About the above legal liability issues, consider this new message (not so new, but the moderation in that mailing list is lagging, slow):
it concludes with the observation that LLM output, and AI output in general, is ineligible for copyrights unless that output is modified by a human for the purposes of creative expression:
In Part II of the report on Copyright and Artificial Intelligence,
https://www.copyright.gov/ai/
For further clarification, there are written guidelines on the scope of copyright in the context of slop which has been handled meaningfully by humans:
"As described above, in many circumstances these outputs will be copyrightable in whole or in part—where AI is used as a tool, and where a human has been able to determine the expressive elements they contain. Prompts alone, however, at this stage are unlikely to satisfy those requirements."
https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-2-Copyrightability-Report.pdf
However, slop which has not been handled meaningfully by humans remains ineligible for copyright:
"Consistent with the Office’s policies described above, applicants have a duty to disclose the inclusion of AI-generated content in a work submitted for registration and to provide a brief explanation of the human author’s contributions to the work"
Thus the problem raised in Andy's original message, that LLMs can be and are being misused to strip licensing (and attribution) from projects' code bases. The code goes into the large language model with a license and a copyright holder. Calling that infringement a training set doesn't change its nature. Then the code comes back out of the model as slop, and while slop in general is ineligible for copyright status the original code remains strongly protected by copyright, whether plagiarized by an LLM or not.
"... a work “autonomously created by artificial intelligence without any creative contribution from a human actor” was “ineligible for registration”.
https://www.copyright.gov/ai/ai_policy_guidance.pdf
And, again, as mentioned in an earlier reply, there is the further problem of the LLMs separating users from projects and vice versa. That is a whole other discussion about duplicated effort, missing communications, and general lost opportunities.
/Lars
PS. COPE (Committee on Publication Ethics) writes the following:
"COPE joins organisations, such as WAME and the JAMA Network among others, to state that AI tools cannot be listed as an author of a paper."
https://publicationethics.org/guidance/cope-position/authorship-and-ai-tools
If those companies take copyrighted work and then turn it into something that lacks the notices and cannot be copyrighted, what does that make those companies?
Go back to reading Andy's article. After all those fake (false) valuations and paid-for hype we'll have a belated reality check or "correction" as Wall Street calls it; when they say the capital is "misallocated" (citing Microsoft) they mean lost. Trillions of dollars down the drain, invested in a dud (wasted on a lie). █