The way the law is being enforced now, this should be an entirely legitimate argument. A snowball's chance in hell though that it holds up without a legal team like OpenAI has.

We know that training LLMs on LLM-generated text leads to an absolute collapse in quality.

This is often repeated, and true. But needs to be qualified.

Modern LLMs use tons and tons of “augmented” data, which is code for LLM generated or massaged data. Some is even generated during training, and judged; papers on that are what made Deepseek famous.

Training on LLM trash will, of course, yield greater trash, and obviously good text has to come from something real. But that’s because slop is slop. And there are issues with “deep frying” LLMs, yes, but simply training on LLM on LLM output does not necessarily reduce quality. It often helps, significantly.

And we also know that AI has been showing up in papers so if they haven’t, then this will be quite unreliable.

Now this is a problem.

TBH LLMs would be pretty good at flagging papers for humans to check, similar to what Wikipedia is already doing. But yeah, if you just feed a prompt bad papers, LLMs just assume the context is true, generally, and that’s a tremendous problem.

I would be surprised if it was something that they trained themselves, and not an off the shelf model hooked up to a search.

That's all it should do. We're nowhere near an AI that could be an arbiter of truth. Hell, most AI couldn't even be trusted to parrot the literature accurately.

I would find this extremely useful as a tool to help me find sources that I then review myself - similar to how I use Wikipedia. But the danger is in people trying to use it for more.

Of course not.

Yeah they added “Don’t hallucinate” to the prompt.

And without hallucinations ???

Likely not

Have they solved the huge unsolved problem no one else has solved

yeah, no.

It probably uses Retrieval Augmented Generation, which can still hallucinate, but usually does a better job for niche questions and it can even provide a source sometimes depending on how you set it up

Obviously not, because that's not possible.

What fun would that be?

I'll keep the hallucinations for myself, tyvm.

Per sci-hub.ru this has been available since March 6th.

"Hear the good news: recent advances in artificial intelligence enabled Sci-Hub to launch a robot that gives scientifically-grounded responses to questions. The robot starts with searching for relevant literature in Sci-Hub database, then turns to selecting and reading most recent studies, and composes the answer based on this information. The answer includes all the references, and each referenced article can be read on Sci-Hub with one click.

Unlike question-answering robots that were based upon the early generation of neural networks, Sci-Hub bot does not hallucinate and is not making up scientific facts and does not cite sources that do not exist. To support its statements, Sci-Bot uses articles from Sci-Hub database. Questions can be asked in any language, and answers can be saved on server and shared.

The alpha version only supports answerig one question, and a more advanced variation that supports conversation mode is coming soon. Right column displays example questions that has been answered by robot - push the question to see the generated answer."

You have to go into each article and check the key points, trust me.

It is a god-tier liar.

It's where a lot of the pirate sites have found refuge from the Western copyright cartels. It's not necessarily a government-affiliated site just because it's got an .ru domain.

.ru is not government domain, it's just the normal russian domain... Literally every country except for america uses the country code top level domain for everyday use.

Because Alexandra Elbakyan lives in Russia. One of the official sci hub homes is .ru also