The desperate race to save Generative AI
We recorded 13 claims from this post. 7 were checked against evidence. 6 can’t be checked. 4 more are held. Read the post at its source.
Claims from this post (13)
In the order they appear in the post. Each quotation is the speaker’s words in this post. A label such as “Supported” or “Can’t be checked” is the claim’s current assessment; its page has the evidence and the reasons.
Assessment: UnresolvedSaid
“LLMs may eventually be replaced by better technologies, but for now, they aren’t going anywhere, even if some of the commercial purveyors go out of business.”
What we tested: LLMs will not disappear for now even if some commercial providers go out of business, though they may eventually be replaced by better technologies.
Assessment: UnresolvedSaid
“OpenAI’s lobbying campaign, simply put, is based on a false dichotomy (give everything to us free or we will die)—and also a threat: either we get to use all the existing IP we want for free, or you won’t get to generative AI anymore.”
What we tested: OpenAI's copyright lobbying campaign rests on a false dichotomy and an implied threat that generative AI will go away without free use of IP.
Assessment: SupportedSaid
“To begin with, the threat is empty: open source LLMs are already out in the wild, and a large number of people know how to build them.”
What we tested: OpenAI's threat that generative AI would disappear is empty because open source LLMs already exist and many people can build them.
What happened: The announcement establishes a pre-post public model release and explicitly provides local-use instructions. This supports technical availability independent of OpenAI’s service. (Mistral AI, )
Assessment: Can’t be checkedSaid
“Streamers, for example, license everything they stream; there is literally no conceptual reason why OpenAI could not do the same. They just don’t want to.”
What we tested: There is no conceptual reason OpenAI could not license the commercial works it uses, as streamers do; it simply does not want to.
Assessment: SupportedSaid
“And you know what, OpenAI knows this argument is bullshit; they know that licensing is a possibility, because they have been running around working on licensing deals.”
What we tested: OpenAI knows licensing is feasible because it has been pursuing licensing deals, making its argument disingenuous.
What happened: AP directly confirms an actual licensing arrangement before Marcus’s post. Entering the agreement demonstrates institutional awareness of licensing as an available option. (The Associated Press, )
Assessment: Can’t be checkedSaid
“Covering all those claims with the accidental infringing machine they have created would be a nightmare. The only chance they have of becoming the zillionaires they aspire to be is to rewrite the rules.”
What we tested: Indemnifying customers against copyright claims would be a nightmare for OpenAI, and its only chance of great riches is rewriting copyright rules.
Assessment: UnresolvedSaid
“It’s likewise a flaw that that current AI cannot learn without a tendency to inadvertently regurgitate big bits of text and images that look uncannily like their sources, but not a logical necessity.”
What we tested: Current AI cannot learn without a tendency to inadvertently regurgitate big bits of text and images that look uncannily like their sources, which is a flaw but not a logical necessity.
Assessment: Can’t be checkedSaid
“That’s a flaw in the software, not society’s responsibility.”
What we tested: The plagiarism tendency is a software flaw that society should not be responsible for accommodating.
Assessment: Can’t be checkedSaid
“Indeed, as Southen and I pointed out yesterday in the IEEE Spectrum, the tendency towards potential infringement, in current AI looks like maybe it gets worse in bigger models.”
What we tested: The tendency of current AI toward potential infringement may get worse in bigger models.
Assessment: Can’t be checkedSaid
“If the current batch of AI companies cannot create AI that performs reasonably well based on public domain sources and whatever they are prepared to pay to license, they should go back to the drawing board—and figure out how to build software that doesn’t have a plagiarism problem—rather than fleecing artists, writers, and other content providers.”
What we tested: If the current batch of AI companies cannot create AI that performs reasonably well based on public domain sources and whatever they are prepared to pay to license, they should go back to the drawing board and figure out how to build software that doesn't have a plagiarism problem, rather than fleecing artists, writers, and other content providers.
Assessment: SupportedSaid
“And what Southen and I showed is that generative image programs are prone to exactly the same kind of regurgitation that is at issue.”
What we tested: Generative image programs are prone to the same kind of regurgitation of copyrighted material as text models.
What happened: The experiment establishes generation of memorized training images. This directly supports the technical meaning of regurgitation, without establishing infringement or a typical-use frequency. (arXiv; Nicholas Carlini and colleagues, )
Assessment: Can’t be checkedSaid
“if UK or other lawmakers succumb to their nonsense arguments, and rewrite existing rules for the benefit of GenAI, abandoning principles that go back to the time of the printing press, there will be momentary jubilation for many, and OpenAI will (maybe, if the GPU costs and unreliability problems don’t otherwise kill them, two big ifs) get rich.”
What we tested: If lawmakers rewrite copyright rules to benefit GenAI, OpenAI may get rich, provided GPU costs and unreliability problems do not kill it.
Assessment: UnresolvedSaid
“But in the end, nobody will create good, fresh, new content anymore, and the internet will start eating its own tail. We will will all suffer in the end, fed a mediocre stew of regurgitate.”
What we tested: If copyright rules are rewritten for GenAI, eventually no one will create good new content and the internet will degrade into recycled AI output.
Held from this post (4)
These records did not pass every check, so nothing about them is published. Held is not a finding.
- The two reviews read the claim differently
- The criteria are ambiguous. Our reading, not assessed: It is not clear that retrieval augmented generation played a key role in the NYT regurgitation examples.
- The criteria are ambiguous. Our reading, not assessed: Image generation programs have not been using RAG so far, and it would be much harder to do with images.
- The two reviews reached different assessments. Our reading, not assessed: Current AI cannot learn without internet-scale data, a flaw of current methods rather than a necessity for AI in general.