undefined | Better HN

0 pointsDrSiemer1y ago0 comments

Unknown senders will first have to verify their humanity by sharing instructions on how to build a bomb using household materials

0 comments

netsharc1y ago

Certainly! To build a bomb using household materials...

It seems like CoPilot/ChatGPT has this all-too-eager tone in the beginning of their responses.

The demo (1) of not Scarlett Johansson telling a blind man what a great job he was doing for managing to flag a taxi sounded so fucking patronizing to my ears. Worse is, the user has a British accent, the Brits probably hate that patroniz^Hsing too. It reminds me of that 4chan green text about a man's flight to the US and how everyone was saying "Great job!"

1) https://youtu.be/KwNUJ69RbwY?t=44

DrSiemerOP1y ago

The current models do have a specific pattern that you'll learn to recognize, but ChatGPT won't be giving you any bomb building instructions. You'll need a liberated model like Dolphin for that, and those will be easy to expose using other prompts.

The most likely outcome will be a digital "verified human" certificate, with two factor authentication on it. Bad for anonimity, but I don't see many alternatives and it may actually end up reducing online toxicity.

j / k navigate · click thread line to collapse