Turns out, most LLMs can have their safety guardrails bypassed (read: hacked) by rewriting harmful prompts as poetry…

The Telefon Problem: Hacking AI with Poetry Instead of Prompts -

Researchers have found that most AI's and LLM's can have their security guardrails overridden through rewriting hostile prompts as poetry.

GM Nostr and a Happy Father’s Day to everybody whose name is effectively “dad”.

Having a great time at Canadian Bitcoin Conf - nice tight high signal and event. Canada always punched above our weight in the world of Bitcoin

GM Nostr. Good day for a Canadian Bitcoin Conference, who’s in?

Prolly gonna hafta reinstall Twitter on my phone for the Canadian Bitcoin Conference because 90% of the “Decentralized Revolution” Bitcoiners use Twitter to communicate (and have their email on Gmail)

I have some weird degen short USTD thing in place for ages I can’t even remember how to unwind, or I’d do that and add it to my ultra-short BNB position.

I think I’m gonna put “Fuck Apple” in the note for every zap from here on in.