Interesting article in Wired about adversarial attacks on ML systems to get them to do things that they are explicitely programmed not to do such as describe how to make illegal drugs [1]. The most interesting part of this is that the attacks work on most GPT systems which is probably due to the similar data used to train them.
Vice has an interesting article about the Danish “Synthetic Party”, a political partyled by an AI [2]. Citizens can vote for candidates who will try to get laws passed that match the AI generated goals, there is no option of voting for an AI character. The policies they are advocating for are designed to appeal to the 20% of Danes who don’t vote. They are also trying to inspire similar parties in other countries. I think this has the potential to improve democracy.
Vice reports that in 2021 a man tried to assasinate the Queen of England with inspiration from Star Wars and an AI chat bot [3]. While someone who wants to be a real-life Sith is probably going to end up doing something bad we still don’t want to have chat bots encourage it.
Sam Varghese wrote an interesting article about the allegations that India is following the example of Saudi Arabia and assasinating people in other countries who disagree with their government [5]. We need to stop this.
Ian Jackson wrote an interesting blog post advocating that DKIM PRIVATE keys be rotated and PUBLISHED [6]. The idea is that if a hostile party gets access to the mailbox of someone who received private email from you then in the normal DKIM setup of keys never changing they can prove that the email is authentic when they leak it. While if you mail server publishes the old keys as Ian advocates then the hostile party can’t prove that you sent the email in question as anyone could have forged a signature. Anything that involves publishing a private key gets an immediate negative reaction but I can’t fault the logic here.