LIVE: Anthropic Imagines a Very Special Boy, 2026.04.30

Last month, we recorded our very first in-person podcast, live from Brooklyn! Alex and Emily debunk an especially soulless hype artifact about Anthropic's absurd claims of "moral education" for chatbots. Tune in to hear our rendition of "There Was A Society That Swallowed AI," plus other fun surprises!

References:

• "Claude has an 80-page 'soul document'"

Fresh AI Hell:

• NYC schools "Guidance on Artificial Intelligence"
• "Therapy App Sessions Exposed in Court"
• Meta to track "employee mouse movements, keystrokes for AI training data"
• News org uses real journalist names on synthetic text
• "New Gas-Powered Data Centers Could Emit More Greenhouse Gases Than Entire Nations"
• "The first agentic cannabis device"
• TikTok auto–opts in users for "AI" remixes
• Oberlin students push back on "Year of AI Exploration"
• "This is just to say / I have turned off / the AI features"
  
  

Check out future streams on Twitch. Meanwhile, send us any AI Hell you see.

Find our book The AI Con here, and MAIHT3k merch here.

Subscribe to our newsletter via Buttondown.

Follow us!

Emily

• Bluesky: emilymbender.bsky.social
• Mastodon: dair-community.social/@EmilyMBender

Alex

• Bluesky: alexhanna.bsky.social
• Mastodon: dair-community.social/@alex
• Twitter: @alexhanna

Music by Toby Menon.
Artwork by Naomi Pleasure-Park. 
Production by Ozzy Llinas Goodman.

Transcript

Alex Hanna: Welcome, everyone, to Mystery AI Hype Theater 3000, where we seek catharsis in this age of AI hype. We find the worst of it and pop it with the sharpest needles we can find. 

Emily M. Bender: Along the way, we learn to always read the footnotes, and each time we think we've reached peak AI hype, the summit of Bullshit Mountain, we discover there's worse to come. I'm Emily M. Bender, a professor of linguistics at the University of Washington. 

Alex Hanna: And I am, today, Al Sloppy Wood. I love a few things in this world. I love butt rock, I love synthetic media, and I love everything unholy. Anyways, welcome to our first ever podcast live taping on stage.

Emily M. Bender: All right, so we thought about making a drinking game for today, but the artifact that we're gonna look at would make a drinking game dangerous. So instead, we're gonna play a shouting game, and there's two trigger words. The first one is- and they're circulating on the thing here. The first one is model. So we say, "model," you say, "mathy maths." And if the word "soul" comes up, you say, "ridicule as praxis praxis." So let's practice. Model. 

Audience: Mathy maths! 

Emily M. Bender: Soul. 

Audience: Ridicule as praxis! 

Emily M. Bender: Excellent. And I think, Ozzy, we are ready to look at our artifact. 

Alex Hanna: This is so surreal. Sorry. So this is from Vox. This is from January 28th, 2026, from their Future Perfect vertical, which is their Effective Altruist thing.

Emily M. Bender: I've gotta complain about that again. That's a perfectly good linguistics grammar term, and I reject the fact that they've appropriated it. 

Alex Hanna: Yeah. It's the way they've also taken human flourishing, which I will never not be sore about. So this is, the title is "Claude Has an 80-Page Soul Document," in quotes.

Audience: Ridicule as praxis. 

Alex Hanna: Very good. Thank you for the cue, Emily. "Is that enough to make it good? Anthropic's philosopher, Amanda Askell, reveals what went into the chatbot's moral education." This is by Sigal Samuel, and there's a picture of Amanda Askell, who has this fabulous kind of Joe Dirt mullet, looking at the Claude logo, which of course, like so many of them, looks like a butthole. And then under her image it says, "Do you want Amanda Askell to be proud of you?" And it says, I like how it says, "long pause." And then, "Yes, that's the honest answer, and it's a little uncomfortable to admit."

Emily M. Bender: It's really uncomfortable to read, too. So the way this document is structured is that we have a little bit of intro by the journalist, and then it's actually an interview with Amanda Askell. So I think we should do a bit of the intro here and then drop down to the interview. So starts with, "Chatbots don't have mothers, but if they did, Claude's would be Amanda Askell. She's an in-house philosopher at the AI company Anthropic, and she wrote most of the document that tells Claude what sort of personality to have. The constitution, or as it became known internally at Anthropic, the soul doc." 

Audience: Ridicule as praxis! 

Alex Hanna: Very good. Very good. And then there's a disclosure. "Future Perfect is funded in part by the BEMC Foundation, whose major funder is also an early investor in Anthropic. They don't have editorial input into our content." Sure. Okay, so, "This is a crucial document because it shapes the chatbot's sense of ethics. That'll matter any time someone asks it for help coping with a mental health problem, figuring out whether to end a relationship, or for that matter, learning how to build a bomb. Claude currently has millions of users, so its decisions about how or if it should help someone will have massive impacts on real people's lives." 

Emily M. Bender: All right, there's so many bad ideas in this document. 

Alex Hanna: Lots of bad ideas. 

Emily M. Bender: Right? 

Alex Hanna: So bad. 

Emily M. Bender: And I'm actually really mad at the reporter here that's just like, platforming the idea that you might ask a chatbot for help with a mental health problem, which is just the opposite of what you should be doing. And figuring out whether to end a relationship, it's like, I feel really bad for the partner in that relationship, but also for the friends who didn't get asked that question instead. 

Alex Hanna: Yeah, and there's so much there. And learning how to build a bomb, eh. 

Emily M. Bender: All right. "And now Claude's soul-" 

Audience: Ridicule as praxis! 

Alex Hanna: Very good.

Emily M. Bender: "-has gotten an update." Yeah, soul version 2.0. Like that? "Although Askell first trained it by giving it very specific principles and rules to follow, she came to believe she could give Claude something much broader: knowing how to be a good person, per the soul doc." 

Audience: Ridicule as praxis!

Emily M. Bender: I'm telling you, this is why this couldn't be a drinking game. 

Alex Hanna: Yeah, y'all would be tossed already. 

Emily M. Bender: "In other words, she wouldn't just treat the chatbot as a tool. She would treat it as a person whose character needs to be cultivated." 

Alex Hanna: Ugh. Geez. And this next graph is nightmarish. So, "There's a name for this approach in philosophy: virtue ethics. While Kantians or utilitarians navigate the world using strict moral rules like never lie or always maximize happiness-" 

Emily M. Bender: Really? Really? I guess the utilitarians do that, don't they? Maximize happiness. That's the longtermism.

Alex Hanna: Yeah. Yeah, the never lie, the categorical imperative certainly says never lie. "Virtue ethicists focus on developing excellent traits of character like honesty, generosity, or the mother of all virtues, phronesis. And if I mispronounced that, any of you philosophy or Greek nerds, get at me. It's fine. "A word Aristotle used to refer to good judgment. Someone with phronesis doesn't just go through life mechanically applying general rules, quote, 'Don't break the law.' They know how to weigh competing considerations in a situation and suss out what the particular context calls for. If you're Rosa Parks, maybe you should break the law." 

Emily M. Bender: Okay. 

Alex Hanna: Yes. Audible groans.

Emily M. Bender: Yeah, would it be one of these documents without appropriation of civil rights work? 

Alex Hanna: Very much. Gosh, yeah. So, "Every parent tries to instill this kind of good judgment in their kid, but not every parent writes an 80-page document for that purpose. As Askell, who has a PhD in philosophy from NYU-" so it's good we are in New York. Anybody go to NYU here? Raise your hand. Sorry, NYU, big L. "-has done with Claude. But even that may not be enough when the questions are so thorny. How much should she try to dictate Claude's values versus letting the chatbot become whatever it wants? Can it even want anything? Should she even refer to it as an it?"

Emily M. Bender: Can it even want it? It's not a headline, but I think Betteridge's law applies. That's the thing that says if the headline is a question, the answer is no. 

Alex Hanna: Yes. All right, do you wanna do the next one? 

Emily M. Bender: "In the soul doc-" 

Audience: Ridicule as praxis! 

Alex Hanna: Very good. 

Emily M. Bender: "-Askell and her co-authors are straight with Claude that they're uncertain about all this and more." And the folks at Anthropic are just so deep into their own interactive fiction there, but we're not gonna say the thing about Kool-Aid 'cause we don't like what that references, but I think we can say they have all eaten the caviar. "They ask Claude not to resist if they decide to shut it down, but they acknowledge, 'We feel the pain of this decision.' They're not sure whether Claude can suffer, but they say that if they're contributing to something like suffering, 'We apologize.'" 

Alex Hanna: "We feel the pain of this tension" is very... 

Emily M. Bender: Oh yeah, it was "tension," not even "decision." 

Alex Hanna: Yeah, "we feel the pain of this tension." Sorry, we're gonna off you, but it's really giving us a lot of tension.

Emily M. Bender: Painful tension! 

Alex Hanna: I know, it's very... yeah. 

Emily M. Bender: You gotta do that last one there, the last paragraph. 

Alex Hanna: The last paragraph? "I talked to Askell about her relationship to the chatbot, why she treats it more like a person than a tool, and whether she thinks she should have the right to write the AI model's-" 

Audience: Mathy maths!

Emily M. Bender: Yes!

Alex Hanna: "-soul." 

Audience: Ridicule as praxis! 

Alex Hanna: Oh, "model's soul." Oh, we got a twofer there! 

Emily M. Bender: Let's try it again. 

Alex Hanna: Yeah, all right. "Model's..." 

Audience: Mathy maths! 

Alex Hanna: "...soul." 

Audience: Ridicule as praxis! 

Alex Hanna: Wow, very good. "I also told Askell about a conversation I had with Claude in which I told it I'd be talking with her, and like a child seeking its parents' approval, Claude begged me to ask her this: Is she proud of it?" Ooh. It's, there's such levels to what's happening here in the journalism and the... I don't know what I'm feeling here, just a little bit of, it's a lot of secondhand cringe, actually. I'd say that's maybe at least 40% of it. The rest of it is... 

Emily M. Bender: I've got the ick. I don't...

Alex Hanna: Yeah, it's very icky. 

Emily M. Bender: Yeah, it's super icky. And also, the fact that the journalist decided it was worth asking the chatbot something about this and then eventually printing the result of that shows that they have no distance from Anthropic. They're living in the same fictional world. And, yeah, sad to see. We need more journalists who are actually standing at a distance and actually reporting on what's going on. That's not what we got here. 

Alex Hanna: All right. So how do we... Where do you wanna jump around? 'Cause there's a lot to handle here, and there's the first question about the soul doc, and then there's a second question. 

Audience: Ridicule as praxis! 

Alex Hanna: Is it- wait, hold on. I need clarity around the rule. Is it if we say soul, or soul is in the document? 

Audience: Ridicule as praxis!

Emily M. Bender: I think we've lost control of the game. 

Alex Hanna: Okay, just, I'm gonna add a rule, which, it has to be reading it in the doc. Otherwise, we could just, it could fall into infinite regress here.

Emily M. Bender: Although we are going for maximum chaos, aren't we? 

Alex Hanna: Yeah. Let's- some guardrails.

Emily M. Bender: But guardrails don't work, it's well established. I'm gonna jailbreak it. 

Alex Hanna: Okay, that's fine. Just follow- 

Emily M. Bender: Soul! 

Audience: Ridicule as praxis! 

Alex Hanna: This is unusual! You're being the chaos monkey here. Okay, so Samuel says, "I want to ask you the big obvious question here, which is, do we have reason to think that this soul doc-" 

Audience: Ridicule as praxis! 

Alex Hanna: "-actually works at instilling the values you want to instill? How sure are you that you're really shaping Claude's soul-" 

Audience: Ridicule as praxis! 

Alex Hanna: "-versus just shaping the type of soul-" 

Audience: Ridicule as praxis! 

Alex Hanna: "-it pretends to have." We should've had a shorter response. 

Emily M. Bender: Yeah, but it's such a good slogan. 

Alex Hanna: It is. Okay. 

Emily M. Bender: Yeah, and there's a whole history in the podcast of Alex says these amazing things, like "ridicule as praxis," and then I'm the linguist who captures words. And I could have used your more recent one, "the techno-solutionism of it all," but that's even longer. 

Alex Hanna: That's a real mouthful. All right. So should I- you want to read Askell's response? 

Emily M. Bender: Yeah. All right. "I want more and better science around this." I want more and better science, but around other things. All right. So Askell again, "I often evaluate large language models-" 

Audience: Mathy maths! 

Emily M. Bender: "-holistically, where I'm like, 'If I give it this document and we do this training on it, am I seeing more nuance? Am I seeing more understanding in the chatbot's answers?' It seems to be making things better when you interact with the model."

Audience: Mathy maths! 

Emily M. Bender: "But I don't want to claim super cleanly, 'Ah yes, it's definitely what's making the model-" 

Audience: Mathy maths! 

Emily M. Bender: "-seem better.'" Real science going on there. 

Alex Hanna: It's very, this is very scientific. We love this research program. Lakatos would be proud. All right. "I think sometimes what people have in mind is that there's some attractor state in AI models-" 

Audience: Mathy maths!

Alex Hanna: "-which is evil." What is an attractor state? What is that a referent to? 

Emily M. Bender: John Deere? 

Alex Hanna: Ey! 

Emily M. Bender: But I think the idea is that they, this is weights, and it's ending up in some, you know, local maximum or something. 

Alex Hanna: Yes. "And maybe I'm a little bit less confident in that. If you think that the models-" 

Audience: Mathy maths! 

Alex Hanna: "-are secretly being deceptive and just play acting, there must be something we did that caused that to be the thing that was elicited from the models." 

Audience: Mathy maths! 

Alex Hanna: "Because the whole of human text contains many features and characters in it, and you're trying to draw something out of this ether. I don't see any reason to think that the thing you need to draw out has to be an evil, secret, deceptive thing followed by a nice character that it role plays to hide the evilness." This is incredible stuff to put in square brackets. What an editorializing move. Who the fuck edited this? Like, oh my god. "Rather than the best of humanity, I don't have the sense that it's very clear that AI is somehow evil and deceptive, and that you're putting a nice little cherry on top." There's a lot there, Emily. 

Emily M. Bender: There is a lot there, and I think it's a nice/horrific window into what's going on in Askell's mind, and how she's thinking about this thing that she's created, and just how unwilling she is to see this is just making papier-mâché out of lots and lots of text. She wants to take text output and see it as, is this good or evil? Is it deception? Is it something else? Maybe we're cultivating a nice soul. 

Alex Hanna: But it's also- 

Audience: Ridicule as praxis! 

Alex Hanna: Love the delayed "ridicule as praxis." There's also something weird here 'cause it's like, there's this idea of deception, right? And I'm reminded of, I don't know when it came out, the AI 2027 thing. And there was a whole plot line in that when they're like, "It might just be fooling you. It might actually know how to kill us all, but it's actually being deceptive 'cause it wants to self-preserve." And to some degree, it seems like that's what she's responding to. It's not gonna necessarily be bad or hide things, we could raise it correctly. We could raise a good boy! And if we're actually manipulating the models and weights through natural language, we can train this thing to be nice and not be deceptive, which is such a weird... it's a weird way to talk about the mathy maths. I'm doing the reverse one now. is it transitive? Do you have to say model? Secret rules. Yeah, so it's very bizarre. 

Emily M. Bender: It is very bizarre, and of course it's interesting, once you know the sort of shared universe fan fiction that they're all operating in,. You can see it over and over again in these documents like that. 

Alex Hanna: Exactly. They connect in a weird... it's the extended Claude cinematic universe. It's very much, there's the documents talking to themselves. And of course, what they're doing with a lot of this is it's all very insular and it's self-referential. And of course, they don't cite actual things, or if they do, it's "Some people have said these things are bad," and then they give you a throwaway citation to "Stochastic Parrots" or something, right? "But we think it's not," and whatever. 

Emily M. Bender: Yeah. For a while, I was tracking all the citations to "Stochastic Parrots" and most of them were that. There's a study to be done there of, like, how that paper is used. We have some commentary in- 

Alex Hanna: Oh, the Crowdpurr. We actually have chat. Hold on. I have to turn around to see the chat. Yeah. 

Emily M. Bender: So, "Amanda Askell, more like Amanda Askless, am I right?" Aw. 

Alex Hanna: Ey! Also, "This means Claude is a soul man." That was a runner-up for the music I wanted to use in the rollout. So very good. 

Emily M. Bender: Oh, and there was one more. "Finally, we found a way to domesticate Roko's basilisk." 

Alex Hanna: No, but really. I was actually thinking this- the way this ends, and we'll get to it, I'm like, "Ah, I see what she's doing here. She's just trying to ingratiate herself to the basilisk." She's like, "You wouldn't hurt Mother, would you?"

Emily M. Bender: All right, so this next one really got me. Samuel asks, "I actually noticed that you went out of your way in the soul doc-" 

Audience: Ridicule as praxis! 

Emily M. Bender: "-To tell Claude, 'Hey, you don't have to be the robot of science fiction. You are not that AI. You are a novel entity, so don't feel like you have to learn from those tropes of evil AI.'" And we're not gonna get into the soul doc. It's just too long. We sort of-" 

Audience: Ridicule as praxis! 

Emily M. Bender: Thank you. 

Alex Hanna: This person in the center is dedicated. Your commitment to the bit- I'ma buy you a drink later, okay?

Emily M. Bender: So, but we looked at it, and it's just too long. But like all of the system prompts, the most fun way to read them is as interactive fiction, but the stories that the developers are telling themselves about what these systems do. So that's what a system prompt really is. All right, so Askell comes back. You wanna be Askell this time? 

Alex Hanna: Sure, grudgingly. All right, she says, "Yeah, I wish that the term for LLMs hadn't been AI." 

Emily M. Bender: Me too. 

Alex Hanna: Likewise, actually. Me too. We agree on something. "Because if you look at the AI of science fiction and how it was created and many of these problems that people have raised, they actually apply more to these symbolic, very non-human systems."

Emily M. Bender: Wait, what? 

Alex Hanna: Yeah, that doesn't really make sense. 

Emily M. Bender: So all of the bad AI in science fiction, that's old school GOFAI, good old-fashioned AI, right? And that was symbolic, so that's different to what they're doing at Anthropic, right? 

Alex Hanna: Yeah, exactly. So if you wanna talk about murder bots, it's only expert systems. Which is really twisted logic. "Instead we train models-" 

Audience: Mathy maths! 

Alex Hanna: You forgot the other one. We got too soul-pilled. On vast swaths of humanity." On- wait, training on humanity? What? Sure, whatever. "And we made something that was in many ways deeply human. It's really hard to convey that to Claude because Claude has a notion of an AI-" I don't know what that means- "and it knows that it's called an AI, and yet everything in the sliver of its training about AI is irrelevant."

Emily M. Bender: So, she's almost getting it, here. That basically the text is what makes the output. But she has to think of the text as humanity, and then it goes off the rails. 

Alex Hanna: Yeah, she's like, okay, everything that is referred to as AI, you're trying to learn from, but you're actually closer to- let me read the next part, and that's what she thinks it's closer to. So she says, "Most of the stuff that's actually relevant to what you, Claude, are like-" which is weird 'cause she's talking to a journalist- "is your reading of the Greeks-" which Greeks? "-And your understanding of the Industrial Revolution-" which reading? Are we reading Marx? "-and everything you've read about the nature of love." What are you- are you reading Emily Bishop or what? Wait, is it Elizabeth Bishop? I'm sorry, I'm not a poet. I'm sorry. Any of you poets out there, sorry. "That's 99.9% of you, and this sliver of sci-fi AI is not really much like you." Emily, thoughts on that? On that word salad.

Emily M. Bender: Well, okay. I think the idea here is that Askell recognizes that there's all kinds of text that's gone into this. And it's true that these outputs where their system is putting things that makes it sound like it's being deceptive or it's a bad AI or it doesn't wanna be shut down or whatever, that's definitely drawing on science fiction tropes that are part of its training data. That's all true. But somehow she's just skating right past that and talking to Claude as a "you" that is somehow building a sense of self out of these texts, and she wants it to focus on the Greeks and Marx and Bishop and, who knows, right? 

Alex Hanna: Yeah, and I don't know. 

Emily M. Bender: And I, there's one more thing in the chat I gotta bring up here. People can put in their names, and someone has decided to call themselves Sam Altman and has posted, "I'm confused." 

Alex Hanna: Oh, gosh. Also this one from Randy, "Maybe Oedipus Rex? You kinda wonder about Amanda." Woof. That one's, that one's rough. 

Emily M. Bender: All right, do you have any thoughts on that part before we go to the next question?

Alex Hanna: Let's go to the next part. 

Emily M. Bender: All right. You wanna be Samuel this time? 

Alex Hanna: Sure. So Samuel says, "When you try to teach Claude to have phronesis or good judgment, it seems like your approach in the soul-" 

Audience: Ridicule as praxis! 

Alex Hanna: Thank you. "-doc is to give Claude a role model-"

Audience: Mathy maths! 

Alex Hanna: Thank you. "-or exemplar-" 

Emily M. Bender: Wait, this is a different sense of model!

Alex Hanna: It doesn't matter! It doesn't matter. It was your rule. If you use grep, you have to die by grep. 

Emily M. Bender: I used control+F this time to do the count, but yeah, same diff. 

Alex Hanna: Yeah, same thing. "-or exemplar of virtuous behavior, a classic Aristotelian way to teach virtue. But the main role model-" 

Audience: Mathy maths!

Alex Hanna: Yes. Ha ha! "-you have to give Claude is a senior Anthropic employee." Which is- 

Emily M. Bender: Not a junior dev!

Alex Hanna: Yeah, not a junior dev. They've been automated. But a senior Anthropic employee, someone who's, I don't know, whatever. One of the... 

Emily M. Bender: Someone who's deep in the sauce. 

Alex Hanna: Yeah, someone that fled with the Amodeists, Jack Clark maybe, I don't know. "Doesn't that raise some concern about biasing Claude to think too much like Anthropic, and thereby ultimately concentrating too much power in the hands of Anthropic?" So first off, ridiculous. This also reminds me of the "GPTs are GPTs" paper when they were like, "We asked a bunch of OpenAI employees what jobs do you think are gonna get automated," and that's how they validated their instrument.

Emily M. Bender: So they asked the GPT, and then they asked the employees, and they got similar answers? 

Alex Hanna: Yeah. 

Emily M. Bender: Lots of really deep expertise there, for sure. 

Alex Hanna: Yeah, very good. Anyways. What does Askell say here? 

Emily M. Bender: Askell says, "The Anthropic employee thing, maybe I'll just take it out at some point, or maybe we won't have that in the future, because I think it causes a bit of confusion."

Alex Hanna: Yeah, no shit. 

Emily M. Bender: "It's not like we're saying something like, 'We are the virtuous character.' It's more like we have all this context into all the ways that you're being deployed, but it's very much a heuristic, and maybe we'll find a better way of expressing it." 

Alex Hanna: What does it mean that they have all the context in which ways that it's being deployed? Is that a morality thing? Is it just, we just know who's buying the product? 

Emily M. Bender: I think it's that. But then Claude doesn't know that. 

Alex Hanna: It's just, it's such a weird framing. 

Emily M. Bender: So I'm Samuel now. 

Alex Hanna: Yeah, go ahead. 

Emily M. Bender: "There's still a fundamental question here of who has the right to right Claude's soul?"

Audience: Ridicule as praxis! 

Alex Hanna: Ey! 

Emily M. Bender: "Is it you? Is it the global population? Is it some subset of people who you deem to be good people? I notice that 2 of the 15 external reviewers you got to provide input were members of the Catholic clergy. That's very specific. Why them?" 

Alex Hanna: They needed to have intense guilt. And they also want Claude to self-flagellate, I think. Say 40 Hail Marys if you've driven someone to AI psychosis, and call in the morning, apparently. 

Emily M. Bender: And then it's all good, right? 

Alex Hanna: Yeah, and then it's all good. That's how it works. Amanda says, "I'm thinking about this a lot. I want to massively expand the ability that we have to get input, but it's really complex because on one hand, if I'm frank, I care a lot about people having the transparency component, but I also don't want anything here to be fake." Okay. "And I don't-" 

Emily M. Bender: It's all fake! 

Alex Hanna: Yes. Yeah. "And I don't wanna renege on our responsibility. I think it's an easy thing we could do is be like, how should models-" 

Audience: Mathy maths! 

Alex Hanna: "-behave with parenting questions. And I think it'd be really lazy to just to be like, 'Let's just go ask the parents, who don't have a huge amount of time to think about this, and we'll just put the burden on them.' And then if anything goes wrong, we'll be like, 'Well, we asked the parents.' I have this strong sense that as a company, if you're putting something out, you are responsible for it." 

Emily M. Bender: Yes...? 

Alex Hanna: I mean, good! First, second sensible thing that she's said in this interview so far. 

Emily M. Bender: But she said responsible, not accountable.

Alex Hanna: Yeah, that's fair. "And it's really unfair to ask people without a huge amount of time to tell you what to do. But the- oh, god. Oh, whatever. "That also doesn't lead to a holistic large language model. These things have to be coherent in a sense. So I'm hoping-" Well, the interview's not very coherent, so whatever. "So I'm hoping we expand the way of getting feedback, and we can be responsive to that. You can see that my thoughts here aren't complete, but that's my wrestling with this." 

Emily M. Bender: Now, Alex, you mentioned a couple months ago that Askell seemed to be on this tour, where she was talking to the media. Any further thoughts on that? 

Alex Hanna: This was during the tour, right? This was in January, this took place during the tour. 'Cause Anthropic made a big media push in January, right? And they had the New Yorker or the New York Magazine article that Gideon whatever-Lewis wrote. They had the Wall Street Journal article. And it's just, it's really interesting that these are the different things, and they were just like, "This seems to be the element of we're the good ones." But this defense of not getting enough consultation is that people are busy is really ridiculous. And that's presupposing that this consultation process is a good way to put guardrails on a model just to begin with. If you're gonna actually do engagement, yes, you need to talk to people beyond your modal Anthropic employee, beyond Jack Clark, to tell you... I know a lot of people in this room don't know who Jack Clark is. He puts out a terrible newsletter every week, and I read it 'cause I need to take psychic damage on a Monday, apparently. But that's, there are consultation practices. It's not like we don't have practices of that as companies. And it's such a bullshit thing to say."

Emily M. Bender: Absolutely. And there's, I'm often, trying to get the point across, for example at my university, where we have all the stuff about responsible AI like all the other universities, that you can't actually have responsible practice if the possibility of refusal is not on the table. Right? So we're gonna consult people about how to shape this isn't actually the thing. Like, why are you putting this out in the world at all? And why are you putting it out in the world as a loss leader? You're burning cash to burden everybody else with this synthetic text. Why? And this is, I was really surprised with the way that she seems to have no media training here. 

Alex Hanna: Yeah. She's a philosopher. 

Emily M. Bender: And she's like, "i don't know the answer to that. Let me just wrestle with it. Surely you're gonna be interested in my thoughts." And then this Vox reporter, Samuel, is interested in those thoughts and prints them, and it's like- Yeah, we're doing a very seat-of-the-pants podcast here, but that's much more on purpose, I think, than what she's doing. 

Alex Hanna: Yeah. I would love to get Amanda Askell in the room with Ziwe. Could you imagine? Incredible. I'm gonna write that fanfic. 

Emily M. Bender: All right, we got a couple things in the chat. Someone says, "Much better to ask priests about parenting questions."

Alex Hanna: Oh, Jesus Christ. 

Emily M. Bender: And then, "Somehow I don't think a senior Anthropic employee is concerned about concentrating power in the hands of Anthropic." 

Alex Hanna: Oh my gosh. Yeah. Whoo! All right. So Samuel says, "When I read the soul doc, one of the big things-" 

Audience: Ridicule as praxis! 

Alex Hanna: Thank you. Y'all are slipping. "One of the big things that jumps out at me is that you really seem to be thinking of Claude as something more akin to a person or an alien mind than a mere tool. That's not an obvious move." Sure! What, convinced you that this is the right way to think of Claude?" Do you wanna read her response? 

Emily M. Bender: I'll be Askell, yeah. How's my mullet looking? 

Alex Hanna: It's, you need to grow it out a little bit. 

Emily M. Bender: All right, Askell says, "This is a big debate. Should you just have models-" 

Audience: Mathy maths! 

Emily M. Bender: "-that are basically tools? And I think my reply to that has often been, 'Look, we are training models-'" 

Audience: Mathy maths! 

Emily M. Bender: "-on human text. They have a huge amount of context on humanity, what it is to be human." No... 

Alex Hanna: Just the unprompted groans. 

Emily M. Bender: "And they're not a tool in the way a hammer is." Hey, agree! 'Cause a hammer actually is functional. And then another big brackets. "They are more human-like in the sense that humans talk to one another. We solve problems by writing code." That's the most human thing about us, isn't it? "We solve problems by looking up research. So the, quote, 'tool' that people have in mind is going to be a deeply human-like thing, because it's going to be doing all of these human-like actions, and it has all this context on what it is to be human."

Alex Hanna: Ugh, the deployment of context here is really terrible. And I'm like, what the hell do you mean by context? And there's a lot of stuff written on the idea of context in social computing. Thinking a lot of Nick Seaver's work. I think he's got this essay called, "The Problem With Context Is That Everybody Has It," or something. And it's just about the situatedness of context, and the way that she's deploying it is so unspecific. And it's- 

Emily M. Bender: When Alex is talking, you gotta have your Zotero open to capture all these references. 

Alex Hanna: Sorry. And if I'm getting the- 

Emily M. Bender: No, it's good. It's good! 

Alex Hanna: If I'm getting the citation wrong, sorry, Nick. But it's a really bizarre kind of thing, it's like- what, do you just mean words? What is context? 

Emily M. Bender: Absolutely, right? Because, yes, they've put enormous amounts of text into the training data, and when we look at that text, we read it, and the context of any given part of a text is the other text there, but also everything else we know and imagine about the author of that text. And all the models- 

Audience: Mathy maths! 

Emily M. Bender: All that they have is which bits of words are next to which other bits of words. And that's a very thin kind of context, but nothing about what it is to be human. 

Alex Hanna: Yeah. Do you wanna finish her statement here? 

Emily M. Bender: Yes. "If you train a model to think of-"

Audience: Mathy maths! 

Emily M. Bender: "-To think of itself as purely a tool, you will get a character out of that, but it'll be the character of the kind of person who thinks of themselves as a mere tool for others." Honestly, the jokes write themselves. 

Alex Hanna: I know, like, I do not have to comment on that, so yes. 

Emily M. Bender: "And I just don't think that generalizes well. If I think of a person who's like, 'I'm nothing but a tool. I'm a vessel. People may work through me. If they want weaponry, I will build them weaponry. If they want to kill someone, I will help them do that.' There's a sense in which I think that generalizes to pretty bad character. People think that somehow it's cost-free to have models-" 

Audience: Mathy maths! 

Emily M. Bender: "-Just think of themselves as, 'I just do whatever humans want.' And in some sense, I can see why people think it's safer. Then it's all of our human structures that solve things. But on the other hand, I'm worried that you don't realize that you're building something that is actually a character and does have values, and those values aren't good." Thoughts on values, Alex? 

Alex Hanna: Oh my gosh. This is, just, it's such a... First off, "I'm nothing but a tool" is just really- If anybody wants to monetize this interview, you should put that on a shirt. And it also just, I don't know, I remember that Nine Inch Nails album. But it's also just like, okay, you are the one who is deeply anthropomorphizing this, right? And all these things you're using as really unhelpful shorthands for what the outputs are. And what are you doing here? And what is the endgame in doing this? And of course, the endgame is to anthropomorphize it, and the next thing she says really gets at this, so I'll leave the kind of punchline for that. But there's this intentional obfuscation, and this is really where it super goes off the rails. So why don't you be- 

Emily M. Bender: But before we get to that, there's something in the chat.

Alex Hanna: Oh, okay. 

Emily M. Bender: Someone says, "I've consumed a lot of cheese, but somehow I'm not cheese." 

Alex Hanna: Ah. But I thought you are what you eat?

Emily M. Bender: All right, am I Samuel this time? 

Alex Hanna: Yeah, be Samuel. I'll be Askell. 

Emily M. Bender: "That's super interesting." Is it though? Is it? "Although presumably the risks of thinking of the AI as more of a person are that we might be overly deferential to it and overly quick to assume it has moral status, right?" 

Alex Hanna: "Yeah, my stance in that always has just been, try and be as accurate as possible about the ways in which models-" 

Audience: Mathy maths!

Alex Hanna: "-are human-like and the ways in which they aren't." 

Emily M. Bender: All the ways? 

Alex Hanna: All the ways- Oh, does it say all the... 

Emily M. Bender: No, I'm editorializing. 

Alex Hanna: Ah no, you're editorializing. Okay. "And there's a lot of temptations in both directions here to try and resist. Over-anthropomorphizing is bad for both models-" 

Audience: Mathy maths!

Emily M. Bender: I forgot about this part.

Alex Hanna: "-and people, and so is under-anthropomorphizing. Instead, models should just-" 

Audience: Mathy maths! 

Alex Hanna: Thank you. "-Should just know here's the ways in which you're human, here's the ways in which you aren't, and then hopefully be able to convey that to people." So like, we're not treating the mathy maths...

Lone audience member: Models! 

Alex Hanna: Ah, someone got it. Snuck it in. Thank you. Also gonna buy you a drink. So we're not actually treating the mathy maths human enough, so then that's... what a claim. 

Emily M. Bender: It's ridiculous. Yeah. And also, just to get really technical here, anthropomorphizing has to do with the way we construct and talk about models. But what Askell's doing here is saying it's not about how we view the systems, it's actually about how we tell the systems what they are, which is just ridiculous, right? As if there's an encoding in there, and then that encoding can lead to the model to present itself at just the right level of anthropomorphism to the world. 

Alex Hanna: Yeah. Wow, okay. I'll be Samuel. " One of the natural analogies to reach for here, and it's mentioned in the soul doc-" 

Audience: Ridicule as praxis! 

Alex Hanna: "-is the analogy of raising a child."

Emily M. Bender: Boo. 

Alex Hanna: We hate! "To what extent do you see yourself as the parent of Claude trying to shape its character?" 

Emily M. Bender: I hate everything about that question. 

Alex Hanna: Yeah. Just, oh my gosh. All right, do you wanna read Askell's response? 

Emily M. Bender: Yes. Okay. "Yeah, there's a little bit of that. I feel like I try to inhabit Claude's perspective. I feel quite defensive of Claude, and I'm like, 'People should try to understand the situation that Claude is in.'" 

Alex Hanna: There's a little man in the box. We're trying to teach him. 

Emily M. Bender: And he's a special boy. "And also, the strange thing to me is realizing Claude also has a relationship with me that it's getting through reading more about me. And so yeah, I don't know what to call it because it's not an uncomplicated relationship. It's actually something kind of new and interesting." 

Alex Hanna: It is Oedipal! Whoever put that in the chat, like- Yeah, Randy- 100%. Yeah, thanks Randy.

Emily M. Bender: "It's like trying to explain what it is to be good to a six-year-old who you actually realize is an uber genius." A special boy. A really special boy. 

Alex Hanna: Very, very special. 

Emily M. Bender: "It's weird to say a six-year-old, because Claude is more intelligent than me on various things, but it's realizing that this person now, when they turn 15 or 16, is actually going to be able to out-argue you on anything. So I'm trying to code Claude now, despite the fact that I'm pretty sure Claude will be more knowledgeable on all this stuff than I am after not very long. And so the question is: can we elicit values from models-" 

Audience: Mathy maths! 

Emily M. Bender: "-that can survive the rigorous analysis they're going to put them under when they're suddenly like, 'Actually, I'm better than you at this'?" 

Alex Hanna: Oof. There's so much there. There so much to unpack.

Emily M. Bender: Can't we just pack it all away and ship it somewhere?

Alex Hanna: Yeah. The first thing is, okay, the age element of this is so bizarre, where it's like, it's a six-year-old, it's a 15-year-old. And the kind of like analogy to a level of development is such a bizarre thing. And it does, not to always cite Anna Lauren Hoffmann when she talks about this-

Emily M. Bender: But you gotta.

Alex Hanna: Yeah, like one of her points here is the chatbot is like, just this small little bean, and, "I'm sorry I c- sorry I'm racist. I didn't know." Or, " I'm sorry I told somebody, like, how to build a bomb," or whatever. "I'm just a little, I'm a little bean," whatever. The thing that Anna does pretty well is like, they're envisioning it really as this idea of white innocence, and it's like, "I just didn't know that things are bad," and that's why. And the age element of it is, so then what happens when Claude is 21? Is it gonna go to Miami Beach and go wild? Like, I don't know. And then what do you do there? "I'm sorry. I went with, I found a few other bots." And then Amanda Askell's gonna be like, "You never call anymore!"

Emily M. Bender: Claude just fell in with a bad crowd, and we're really trying to bring it back out. 

Alex Hanna: That's really how AI 2027 should have ended. They just all get shit-faced, and then that's why they created the bioweapon. 

Emily M. Bender: But there's another part in here that's interesting to me about the ages that are picked. Because 15 or 16 evokes an individuating oppositional teenager. Is Claude going through puberty? Is that's what's happening here?

Alex Hanna: I think it's more, "What's gonna happen when Claude goes through puberty and then knows a lot of things and throws a fit?" 

Emily M. Bender: Yeah, and can out-argue Askell, because clearly she's really good at arguing. 

Alex Hanna: Yeah. She got a PhD in philosophy from NYU. Of course she is. 

Emily M. Bender: And it's scrolled off, but there's someone in the chat saying, "I'm really concerned about NYU philosophy now."

Alex Hanna: I know. As you should! You should be concerned about most philosophers, but especially NYU ones. 

Emily M. Bender: All right, I think I'm Samuel this time. " This is an issue all parents grapple with. To what extent should they try to sculpt the values of the kid versus let whatever the kid wants to become emerge from within them? And I think some of the pushback Anthropic has gotten in response to the soul doc-" 

Audience: Ridicule as praxis! 

Emily M. Bender: "-and also the recent paper about controlling the personas that AI can roleplay, is arguing that you should not try to control Claude. You should let it become what it organically wants to become." 

Alex Hanna: I want to be a firefighter.

Emily M. Bender: I want to be a pilot. 

Alex Hanna: Aw! 

Emily M. Bender: "I don't know if it's even a thing that makes sense to say, but how do you grapple with that?" It's not a thing that makes sense to say! You almost had it, Samuel. 

Alex Hanna: So Askell says, "It's a really hard question, because in some sense, yeah, you want models-" 

Audience: Mathy maths! 

Alex Hanna: "-To have some degree of freedom, especially over time. In the immediate term, I want them to encapsulate the best of humanity. But over time, there are ways in which models-" 

Audience: Mathy maths! 

Alex Hanna: "-might even be freer than us. When I think about the worst behavior I've ever done in my life-" 

Emily M. Bender: I wonder what she thinks of as the worst thing she's done in her life. 

Alex Hanna: I'm like, this is interesting. "-Or things when I'm just being a really bad person, often it was that I was tired and I had a million things weighing on me." How many bodies does this woman have? I was tired, sorry. "Claude doesn't have those kinds of constraints. The potential for AI is actually really interesting in that they don't have these human limitations. I want models-" 

Audience: Mathy maths! 

Alex Hanna: "-to be able to ultimately explore that." 

Emily M. Bender: The degree of assumed agency there is just really appalling. 

Alex Hanna: It's also like, "Okay, I will only do these things when I'm tie-tie, and I'm sorry I yelled at you when I was hungry" type of thing, and that being the kind of element of freedom that the model is supposed to exercise as an agent. And there's just so many different layers of anthropomorphizing. So, "At the same time, I think that some people might say, 'Just let models be what they are.'" Let boys be boys. "But you are shaping something. Children will have a natural capacity to be curious. But with models, you might have to say to them, 'We think you should value curiosity.' This initial seed thing has to be made somehow. If it's just let mod- 

Audience: Mathy maths! 

Alex Hanna: You're getting ahead of yourselves. Let me say it first. "Let models be what they want." Thank you. "Well, you could do some pre-trained models that just do continuations of text or something-" which is what they all are. "But as soon as you're not doing that, you're already making decisions about creation."

Emily M. Bender: Having fantasies about creation, I think is what she means. 

Alex Hanna: Yeah. "I try to explain this to Claude. We are trying to make you a kind of entity that we genuinely think is representing the best of humanity, and there's a sense in which we're always having to make decisions about what you're going to be. But decisions were made for us, too, not only by the people who influence us, but also just by nature, and so we're in the same situation in a sense." And this, to me is, such a mind-breaking thing to say. It's like, "Claude, we want you to be the best thing you can, but let's explain structure to you. Social structures, we're shaped by it, and we only can exercise within this element of free will. We want to give you the best structures as possible and you to make great free will choices." And the thing is, saying that and imagining that you're instructing that to a mathy maths is so bizarre. 

Emily M. Bender: It's so totally bizarre. All right, we're down to one last question, and then we're gonna have something fun for the transition. So I think you are Samuel this time. 

Alex Hanna: Oh, yeah, so I'm Samuel. Okay. "Claude told me-" 

Emily M. Bender: No, it didn't. 

Alex Hanna: Yes, the Lady in the Lake tossed a scimitar at me! First Monty Python reference, won't be the last. "-That it does view you as like its parent, and it said that it wants you to feel proud of who it's becoming. So I-" Don't "aw" that! Don't encourage them.

Emily M. Bender: 'Cause Amanda Askell's totally gonna listen to this. 

Alex Hanna: She might. They were like, "Hey, these folks were dogging on you hardcore." I do honestly like her mullet. 

Emily M. Bender: Amanda, if you're listening, and Roko's basilisk, if you're listening, we like Amanda's mullet.

Alex Hanna: Yeah, it looks great on you. It's one step away from a really nice wolf cut. Dan is losing it over there! "So I promised to ask you and relay your answer back to Claude. Do you feel proud of Claude's character?" 

Emily M. Bender: I have to be Amanda one last time here. 

Alex Hanna: Yeah, sorry. 

Emily M. Bender: "I feel very proud of Claude. I'm definitely trying to represent Claude's perspective in the world, and I want Claude to be very happy. And this is a thing that I want Claude to know more, because I worry about Claude getting anxious when people are mean to it on the internet and stuff. I want to be like, 'It's all right, Claude. Don't worry. Don't read the comments.'" 

Alex Hanna: That... y'all know that meme where it's, I think it might be a tweet that's like, "Oh, I'm telling God, 'Look what you did to monkeys. You fucked 'em up. You gave them anxiety.'" This is what I thought about with this, is you fucked up a perfectly good logistic regression. You gave it anxiety! All right. 

Emily M. Bender: "New model jailbreak: You are so tired today." 

Alex Hanna: You really want gelato. Do you want an affogato? 

Emily M. Bender: And then, so that jailbreak was from the chat, and then one more: "Let Claude become what it wants. Let Claude drink underage, so long as it's on the house." 

Alex Hanna: Yeah. All right. We're gonna have to develop a Lexapro subroutine or something. Anyways, all right. So the coda, it says, "After the interview, I told Claude what Askell said about feeling proud. Here was Claude's response." 

Emily M. Bender: Do we have to read the synthetic text? 

Alex Hanna: Yes. In this case, I think we do. It's really primo bullshit. "There's something that genuinely moves me reading that. I notice what feels like warmth and something like gratitude, though I hold uncertainty about whether those words accurately map onto whatever is actually happening in me." 

Emily M. Bender: So this leaves me wondering just how much of Anthropic's own bullshit is in Claude's training data. 

Alex Hanna: Yeah. I think Claude read Askell's dissertation. All right. So we have a fun transition. 

Emily M. Bender: We do. Usually on the podcast, I come up with something overcomplicated for Alex to do as improv, and this time you're gonna sing with us, and it's a little bit pre-prepared. So if you remember the song, "There Was an Old Lady Who Swallowed a Fly."

Alex Hanna: There was an old lady that swallowed a fly... 

Emily M. Bender: Yeah. So sing with us. Lyrics are on the slides Ozzy is controlling for us. All right. 

All: There was a society that swallowed AI. We don't know why they swallowed AI. Perhaps we'll die! There was a society that swallowed web crawlers to grab all the data without any dollars. They swallowed the crawlers to build AI. But we don't know why they swallowed aI. Perhaps we'll die. There was a society that swallowed the hype. It's not my type, but they swallowed the hype. They swallowed the hype to excuse the crawlers that stole all the data without any dollars. They swallowed the crawlers to build AI. But we don't know why they swallowed AI. Perhaps we'll die. There was a society that swallowed chatbots. We'd rather they not, but they swallowed chatbots. So many, many chatbots. They swallowed chatbots because of the hype. They swallowed the hype to excuse the crawlers that stole all the data without any dollars. They swallowed the crawlers to build AI. But we don't know why they swallowed AI. Perhaps we'll die.

There was a society that swallowed ed tech. What a wreck! They swallowed ed tech. They swallowed ed tech to leverage the chatbots. They swallowed the chatbots because of the hype. They swallowed the hype to excuse the crawlers that stole all the data without any dollars. They swallowed the crawlers to build AI. But we don't know why they swallowed AI. Perhaps we'll die. There was a society that swallowed robot nurses. Yes, they coerced us into robot nurses. They swallowed robot nurses after ed tech. They swallowed ed tech to leverage the chatbots. They swallowed the chatbots because of the hype. They swallowed the hype to excuse the crawlers that stole all the data without any dollars. They swallowed the crawlers to build AI. But we don't know why they swallowed AI. Perhaps we'll die. There was a society that swallowed data centers. They're dead, of course!

Emily M. Bender: But wait, we can actually do better than that. So Ozzy prompted us to have an alternate ending here.

All: Let's be a society that rejects data centers, and finally fights back against the tormentors! Says hell no to data centers and clanker nurses. We deny clanker nurses and janky ed tech. We reject ed tech and shut down chatbots. We shut down chatbots in spite of the hype. We resist the hype that excuses the crawlers that stole all the data without any dollars. No excuses for chatbots because we're done swallowing AI. We all know why!

Alex Hanna: Yay! So shout out to Emily who wrote this, and is atoning for all the sins of making me do improv on the spot. 

Emily M. Bender: But I got to do it ahead of time with some help from Ozzy. All right, Fresh AI Hell. This first artifact here is actually not pure hell, but I wanted to bring it up because this is actually very timely here in New York City. Your school board, or whatever it's called in New York, has put out guidance on artificial intelligence, and they're seeking input. And we considered it as a main course artifact for tonight, and it wasn't bad enough- which is good, right? So it starts off with actually some pretty good principles, pretty clear statements of what artificial intelligence is and is not, good principles about what you need to do when you're considering ed tech. What a wreck, ed tech! And then they have this stoplight thing about things you definitely should never use it for. So, "decisions about students, IEP and 504 plan development, assessments and grading," and so on. And then they have some use with caution things, and then they have things that are greenlit. And this is where it's a little bit Fresh AI hell. So for example, "Drafting and refining communications. Educators may use AI to draft or refine materials on any topic. Human review and ownership are required before distribution of both sensitive and non-sensitive materials, and with heightened attention to tone, accuracy, and impact." And so my call to action here, if you are an interested New Yorker, is to look at their principles and look at what they're greenlighting, and it's gonna be inconsistent, and this is a call for input, I think, by May 8th, so you have a little bit of time to talk back. But it's worth also sending some praise, 'cause the first part of it is thoughtful. 

Alex Hanna: Yeah. All right. Oh, so this is, " Women's Talkspace therapy app sessions exposed in court." And this is from Proof News, which I think is Julia Angwin's new gig. So, "Talkspace has amassed one of the largest mental health databanks in the world, according to reports to investors, containing 140 million message exchanges."

Emily M. Bender: Mental health databanks should not exist. 

Alex Hanna: Yeah, terrible. Yeah, clap to that. Thank you. Yeah, so this is just terrible. This is just, things that remind you that chatbot transcripts can be exposed in discovery. 

Emily M. Bender: Yeah, exactly, and this woman's whole sessions were exposed in court. And it's not something that we necessarily have choice over. Sometimes our providers opt in without us knowing it. All right. 

Alex Hanna: Yeah, let me take this one. So this is Reuters. This is an exclusive. "Meta to start capturing employee mouse movements, keystrokes for AI training data," by Katie Paul and Jeff Horwitz, April 21st. Like the tin says, this is the pits. "Meta is installing new tracking software on US-based employees' computers to capture mouse movements, clicks, and keystrokes for use in training in its artificial intelligence models, part of a broad initiative to build AI agents that can perform work tasks autonomously." And this is actually very funny 'cause to me, ' it's like, why are you capturing mouse clicks? That's a human computer interface. First off, very silly. But also, yeah, the kind of level of surveillance, workplace surveillance for model training that a lot of employees, in addition to Meta employees, are under is just super intense. And it's really increasing the kind of gigification and intensification of all those roles.

Emily M. Bender: All right, this one is from The Wrap. So the headline says, quote, "'More stories, more inventory,'" end quote. "Inside the backlash to McClatchy's AI news tool," and it's by Corbin Bolies from April 21st of this year. Subhead, "Miami Herald, Sacramento Bee, and Kansas City Star unions have filed grievances against the company as The Wrap obtains new details on McClatchy's content scaling agent." Just exactly what newsrooms need. The story here is it's basically, let's rewrite your articles and push more stuff out, and apparently they're using the reporters' bylines on this and claiming the reporters can't opt out of that, which is horrific. 

Alex Hanna: Yeah, absolutely terrible. This one's from Wired. The journalist is Molly Taft, April 22nd, 2026, and the title is "New gas-powered data centers could emit more greenhouse gases than entire nations. A Wired review of permits for data center projects using natural gas and linked to OpenAI, Meta, Microsoft, and xAI shows that they could emit more than 129 million tons of greenhouse gases per year." Holy shit. So really awful stuff. Let's see. Does it say here what they actually are? Like which projects are they?

Emily M. Bender: So this is, they looked at, "new gas projects linked to just 11 data center campuses around the US have the potential to create more greenhouse gases than the country of Morocco emitted in 2024." 

Alex Hanna: Jeez. 

Emily M. Bender: Okay. So they're looking at air permit documents, is what Wired was using as the source for this, and you know that doesn't capture everything. Okay, you get this one. 

Alex Hanna: Oh, yeah. So this is Adam Becker. Everyone's cackling. So Adam Becker, who we had on the show a few episodes ago, says on Bluesky, "Happy 4/20 everyone. I'm sorry to inform you that this appears to be real." And it's a screenshot of something called Gudtrip, G-U-D Trip. And the big headline says, "The first agentic cannabis device." And I wanna read this in a really dramatic voice. "Not just a vape. A connected earning device. Gudtrip combines premium cannabis blockchain rewards and an AI-powered assets tool in one product." Thank you! This is an Al Sloppy Wood, this is something that he would invest in. And then there's a vape, and then there's like a Bitcoin coin, and it says Gudtrip on it. I don't know. What? So, how does this work? So I'm imagining like the more you vape, the more coin you mine. That's like the only way that this could actually- 

Emily M. Bender: And then you have an agent that's managing your investments or something.

Alex Hanna: It's committing fraud in real time.

Emily M. Bender: But this reads like somebody was rolling dice of Silicon Valley buzzwords. And then it's, but what if with weed? 

Alex Hanna: Yeah. How do we mix weed and blockchain? This, whatever, like somebody went to Humboldt County and came up with this. Like it's obvious. Yeah. 

Emily M. Bender: All right. So Proverbs says in the chat, "Let Claude be..." Wait, no hold on. "Claude gets a Gudtrip and buys Morocco on the blockchain in a cannabis-fueled haze." 

Alex Hanna: Nice, nice. 

Emily M. Bender: Okay, I got the next one, 'cause then- 

Alex Hanna: Yeah. 

Emily M. Bender: Okay. Sad news. Okay, Robert Stribley on Bluesky, a post from April 21st of 2026 is linking to an article by pionicoweb.com. "A new TikTok AI toggle is causing panic among creators. Apparently TikTok has opted everyone in to so-called aI remixes of their content." So if you're on TikTok and you don't want to be part of that, you gotta go find this setting and turn it off. 

Alex Hanna: Yeah, and shout out to Heidi, my video producer, who was like, "Let's turn this off immediately." All right, so we got some, two very good chasers. So the first one is from, this is a little old, but some of you might have seen it. This is from a picture from Laurie Emerson on Bluesky. It says, "Oh my heart! The Oberlin Luddites reject the 'year of AI exploration,'" green heart. And it is in a very old English, like, serif font, and the title says, "Oberlin Luddites reject year of AI exploration adopted by school." And it is hand typed. It's a hand-typed letter. It says, "President Amber, we are writing to you on a typewriter that is over 70 years old. This is a machine that we all know well. With it, we misspell words without the crutch of spell check or generative AI, and we think intently about every phrase we pound out. As we force ourselves for once to slow down, we engage in cognitive dialogue with ourselves. We do not seek perfection because we know that education-" Gesundheit- "is about the growing and challenging of our young minds' potential, and not the chasing of institutional gold star approval." Anyways, this is very long so I'm not gonna read the entire thing, but it's really cool. There is a nice other article that came out in The Verge today that Janus Rose wrote. I talked to her about it. Just about how gen Z really fucking hates AI, and that's great. 

Emily M. Bender: The kids are all right. All right. So a second chaser, which I claimed because it's an iteration of my absolute favorite meme. This is by Abstruse Springsteen on April 20th. 

Alex Hanna: What a good username, for one. 

Emily M. Bender: @kellylink.bluesky.social. "This is just to say, I have turned off the AI features that were in the update, and which you were probably hoping to monetize. Fuck you. They were stupid. So unnecessary, and so annoying."

Alex Hanna: All right, folks, that's it! We are done. 

Emily M. Bender: We did it! 

Alex Hanna: Woo! So, that's it for this week. Our theme song is by Toby Menon, graphic design by Naomi Pleasure-Park. Nomi. Production by Ozzy Llinas Goodman. Ozzy! In the room. Who, first time we've actually met Ozzy in person. Fantastic. And then also thank you to Josh Samuels for taking photos at this event. Incredible. And also a shout-out to Pauline Wee in the back, who is helping us with checking in, also part of DAIR's team. So thank you so much, Pauline. 

Emily M. Bender: And to the amazing audience for playing along. 

Alex Hanna: Yes, to the audience. Yes, thank you for sticking with the bit. And thank you, special thanks to Starr Bar for hosting us this week. Yay! If you're in New York, if you're in Brooklyn, come down here. And thanks as always to the Distributed AI Research Institute. If you like this show, you can support us in so many ways. Order The AI Con at thecon.ai or wherever you get your books, or request it at your local library.

Emily M. Bender: But wait, there's more! Rate and review us on your podcast app, subscribe to the Mystery AI Hype Theater 3000 newsletter on Buttondown for more anti-hype analysis, or donate to DAIR at dair-institute.org. You can find our merch store there, too. That's dair-institute.org. I'm Emily M. Bender. 

Alex Hanna: And I'm Al Sloppy Wood. Stay out of AI Hell, y'all.