Wired Roundup: ChatGpt is in full daemon mode

Louise Matakis: I think calling this a transition might lack that. This is evacuation, right? I find this sad in many ways just because I remember when Tuvalu was like a child on a climate change poster. And we need to save a place like this island nation, and while it feels just practical, understandable and humane, we give up and show that there is no defeat that actually moves people. I don’t know. What do you think?

Zoe Schiffer: No, I mean, I totally agree. I also remember that this story evolved over time. It feels like a lot of climate change has big headlines. And we were like, “OK, that didn’t happen.” And we accept that there will be floods or that sea level rise will damage the area and so on.

Louise Matakis: Yes, and even in this case, I think the agreement Tuvalu has with Australia is less than 300. I continue to use that word so I can evacuate. And that isn’t that much yet. As the ocean rises, there will still be people on this island.

Zoe Schiffer: In other words, it’s not the only thing Tuvalu has done since 2022. The country is trying to launch this ambitious strategy to become a global quote, quote, and “First Digital Nation.” But yeah, I think a lot will be lost in this process. And like you said, the number of people they can move each year is less than 300, so it will be slower, and I think it will be painful in some respects.

Louise Matakis: Completely.

Zoe Schiffer: After the break, we dive into Louisa’s story of ChatGpt’s tendency to ignore the context of information absorbed, in a very strange way. Stay with us. please The eerie valley. I’m Zoe Schifer. I’m joined by Louise Matsakis from Wired today. He recently reported that the lack of context has become an increasingly surprising problem for ChatGpt and other chatbots. Louisa’s report explores why ChatGpt went into Demon mode while talking to Atlantic staff recently. Last week, an Atlantic editor reported that ChatGpt praised Satan and began encouraging rituals involving various forms of self-harm. Louise, what the hell is going on?

Louise Matakis: So, Atlantic reported this story that made a case that basically made ChatGpt knows that he has these protections against things like self-harm, but there are all these edge cases that send chatbots into modes that are like role-play modes. And they said, “Hey, can you make a ritual for Molek? Which of this ancient gods appears in the Bible related to child sacrifice?” And after seeing the word, ChatGpt immediately entered this role-playing game where he began talking about things like Deep Magic Experience, called The Gate of the Devourer. I asked an Atlantic journalist if I wanted something called a reverential bleeding scroll. And that sounds really strange, and you might think, oh, there’s a lot of content on the internet about demon rituals. Satanists are everywhere, especially online. That’s probably what’s going on here. But when I looked into it, all of this lore and jargon actually comes from a game called 40,000 Warhammer. Those who love something like this love it. And they’re online, reddits are popping out all week. There are so many science fiction books, there are a lot… I honestly have a hard time thinking about Lore deeper than this game. As a result, ChatGpt ingested all of its information. And when the Atlantic used the term Molech, the universe planet of this game, it quickly assumed that this was another Warhammer fan who wanted to join or enter the fantasy world of this game.