logoalt Hacker News

prometheus1992 • today at 7:44 PM • 12 replies • view on HN

Is it possible that brian conrad somehow uploaded the proof to anthropic to cross check something and then the anthropic guys got the proof?

- it doesn't have to be the full proof - even if portions were uploaded, that should also be enough or at least calls for a thorough check on ants side. the human proof has been in the works for 12 years, who knows when something was uploaded somewhere or some info just stayed on a google drive which was later sold.


Replies

Ungh • today at 7:51 PM

While it's unlikely here, this is one reason I tend to keep confidential information entirely offline in notebooks these days. Some fucker is going to copy and paste or index anything I communicate with them to an LLM. They are doing that now with this paper.

nostromo • today at 8:01 PM

Quick public service announcement: both Anthropic and OpenAI allow you to turn off allowing them to train using your data.

➕ show 12 replies
benmusch • today at 8:29 PM

The proofs are entirely different approaches

xp84 • today at 8:10 PM

Well, this guy says his thing is 100% human, "No AI." So in the even that happened, he doesn't have the best ground to stand on to cry foul, since he would be admitting to academic fraud. Using AI to help "check" something = not "100% human."

➕ show 1 reply
billnad • today at 8:27 PM

Considering Anthropic has stolen the worlds information I have to think that this is exactly the case

➕ show 1 reply
Aurornis • today at 8:24 PM

Anthropic’s proof is reportedly 250 times longer by page count. It’s going to take time to determine if they even go in the same path, or if they hold up at all.

JadeNB • today at 8:34 PM

Brian is an incredibly smart guy, and knows that Anthropic is in this area. I don't think he would go uploading somebody else's work without thinking through the implications carefully.

ls612 • today at 7:48 PM

My understanding is that the proofs are completely different, so this is not likely.

catigula • today at 8:18 PM

Yes.

afavour • today at 8:05 PM

One half of me says that’s a crazy conspiracy theory. The other says… yeah, that’s completely plausible.

In general it’s kind of sad to think how often this is going to occur in the future. The provenance of the data sets AIs are trained on is so vague that we’ll never know exactly what came from where, and which human ultimately thought it up.

agnosticmantis • today at 8:16 PM

Regardless of what happened here, this is one reason we need more open models (smaller, bigger, better) publically available despite Anthropic's constant fear mongering and regulatory capture efforts.

menaerus • today at 7:51 PM

Yes of course, as well as it was the case with other several hundreds of proofs also. OpenAI btw