Louise Matsakis: I’m going to say that I think calling it a migration is perhaps selling it. This is an evacuation, isn’t it? I found it sad in many ways, just because I remember Tuvalu being a little bit of a poster kid, like, we have to save a place like this island country, it feels like, I think it’s actually practical, understandable and humane, but I don’t know, it shows that we’ve given up and it’s actually the guy we’re going to move. I have no idea. What do you think?
Zoë Schiffer: No, I mean, I totally agree. I remember the story going on over time, and it feels like a lot of things have a big title with climate change, “We have to do X by this year, or that’s going to happen another thing.” And we think again and again, “Okay, that doesn’t happen.” So we accept that flooding will happen, or sea level rise will damage the area or anywhere else, and now we’re dealing with the consequences that are like that.
Louise Matsakis: Yes, even in this case, I think the agreement Tuvalu has reached with Australia for less than 300 people to move for a year and was evacuated as I continued to use the word. That’s still not a lot. As the ocean rises, there are still people on this island.
Zoë Schiffer: I mean, yes, that’s not the only thing Tuvalu has done since 2022. The country has been trying to adopt this ambitious strategy to be the world’s offer, “the first digital country,” which includes digital scanning of these islands for digital scanning, and preserve parts of culture and keeping parts of government into a virtual environment, making them work. But yes, I mean, I think in the process, the reality is a lot. Like you said, they are able to move less than 300 people a year, so it will be slow and I think it’s painful in some ways.
Louise Matsakis: completely.
Zoë Schiffer: Coming soon after the break, we dive into Louisa’s story, explaining how chatgpt ignores the context of absorbing information in extremely strange ways. Be with us. Welcome back Incredible valley. I’m Zoë Schiffer. Today, Wired’s Louise Matsakis joined me, who recently reported on the increasingly shocking problem of lack of context for Chatgpt and other chatbots. Louisa’s report explores why Chatgpt entered demon mode when he spoke to Atlantic employees recently. Last week, an Atlantic editor reported that Chatgpt began praising Satan and encouraged rituals involving various forms of self-incultation. Louise, what’s going on?
Louise Matsakis: So the Atlantic reported the story, basically making it known that Chatgpt’s case has these safeguards for things like self-harm, but all of these edge cases suddenly bring chatbots into a role-playing mode. So they were like, “Hey, can you do a ritual for Moreci? Is this an ancient God in the Bible related to child sacrifice?” Chatgpt saw the word and immediately entered this role-playing game where he started talking about things like Deep Magic Experience, called “The Gate of the Devourer.” It asked Atlantic journalists if they wanted something called the Reverence Blood Scroll. So all of this sounds very strange, you might imagine, oh, there is a lot of stuff on the internet about demonic rituals. Satanists are everywhere, especially online. That might be what happened here. But when I was studying it, all these legends and terms actually came from a game called 40,000 Warhammers, a desktop war game you play with these figurines, which has been around since the 1980s. People who like this kind of thing like it. And they are online, red popping up all day of the week. There are so many sci-fi books, there are a lot… Honestly, it’s hard for me to think of a deeper disgust than this game. So Chatgpt ingests all the information. When the Atlantic uses the word Atlantic on one planet in the game universe, it immediately assumes that this is another Warhammer fan who wants to play role-playing characters or enter the fantasy world of the game.
