Bet it was scraped from a labs confidential upcoming, unpublished research.

It’s a nice trick, but then what, when researchers start using fully private selfhosted models or quit LLMs all together and the limitless data exfiltration ends?

Fantastic.

3 years late, but I guess it wouldn’t be Qualcomm without a helping of disappointment.

I mean, hallucinations are possible, especially if you have long-running agents. However, I also suspect direct instruction, simply as the alternative would take too long, be excessively expensive and the “newsworthy” result subject to chance.

Can’t build a PR strategy on that.

So, this is going to continue to escalate, until whoever gave the text generator the software tools needed to do this and gave it the access without any kind of firewall, gets charged with the appropriate felonies and is put in jail.

To be clear, this is illegal activity, and the LLM can only do this if a human provides the means. That human == criminal scum.

So, like, if i do this, my IP is logged, cross referenced by my ISP, then the police come knocking at my door, as I have committed a felony (and probably a number of other violations along the way).

Are these laws now invalid or what?

Nice setup! Might check out some services I didn’t know about yet.

But I might add one thing, because I am/was in the same boat: Friends Don’t Let Friends Use Ollama