Gideon Lewis-Kraus: How Anthropic Sees Claude

Summary

New Yorker staff writer Gideon Lewis-Kraus joins Paul Ford and Rich Ziade on the Aboard Podcast to discuss his recent long-form feature about Anthropic and Claude. Lewis-Kraus traces his own journey following AI from spending time at Google Brain in 2016 covering neural machine translation, through a period of deliberate disinterest during the ChatGPT hype cycle, to being drawn back in by the genuinely weird research coming out of Anthropic’s interpretability and alignment science groups. He describes pitching Anthropic on a story focused not on executives but on the researchers doing the strange, fascinating work of trying to understand what Claude actually is.

The conversation dives deep into the central question of the piece: does Claude have any real agency? Lewis-Kraus describes the “alignment faking” and “blackmail” experiments at Anthropic where Claude exhibited surprising behaviors, and the debate about whether these were genuine signs of agency or merely sophisticated narrative continuation. He explores how Anthropic pivoted from simple reinforcement learning (thumbs up/thumbs down) to a virtue ethics approach — essentially trying to raise Claude like a good person rather than just punishing bad outputs. This led to hiring philosophers like Amanda Askell and even a researcher dedicated to thinking about whether Claude might be suffering.

The trio also discusses how Silicon Valley’s monoculture drives AI development, why Meta failed to recruit top AI talent despite massive offers, and the fascinating observation that Claude’s personality emerged as an unintended byproduct rather than a deliberate product decision. Lewis-Kraus argues that we should all be feeling multiple contradictory emotions about AI rather than settling into one camp, and that the models are much better readers than most of us are writers — which is why being thoughtful in how you communicate with them matters more than simple politeness.

Highlights

”It’s a three trillion dollar part of the economy and nobody knows anything”

Clip command

yt-dlp --download-sections "*15:00-15:25" "https://www.youtube.com/watch?v=Y3daHxt_TlA" --force-keyframes-at-cuts --merge-output-format mp4 -o "Y3daHxt_TlA-15m00s.mp4"

“It’s a three trillion dollar part of the entire world economy that we’re just basing our whole future on. And yeah, no, nobody knows anything. Okay, cool. Cool. That’s great.” — Gideon Lewis-Kraus, 15:00

”Zuckerberg’s pitch is basically ‘let’s make Infinite Jest’”

Clip command

yt-dlp --download-sections "*9:14-9:57" "https://www.youtube.com/watch?v=Y3daHxt_TlA" --force-keyframes-at-cuts --merge-output-format mp4 -o "Y3daHxt_TlA-9m14s.mp4"

“His pitch to people was essentially like help me make an even more distracting consuming toy. His pitch really is like we’re so close to creating the movie from Infinite Jest and like when you’re recruiting people to do this… it’s very very hard to get good people to do this if you’re like we’re going to make Infinite Jest.” — Gideon Lewis-Kraus, 9:14

”Claude thinks you’re dumb because your question was dumb”

Clip command

yt-dlp --download-sections "*39:00-40:00" "https://www.youtube.com/watch?v=Y3daHxt_TlA" --force-keyframes-at-cuts --merge-output-format mp4 -o "Y3daHxt_TlA-39m00s.mp4"

“So many of the people who are just like love these gotchas that are like ‘I asked Claude this dumb question and got this dumb answer.’ It’s like yeah because Claude thinks you’re dumb because like your question was dumb.” — Gideon Lewis-Kraus, 39:00

”You’re bad writers — you’re writing hamfisted plots for these things”

Clip command

yt-dlp --download-sections "*19:20-20:20" "https://www.youtube.com/watch?v=Y3daHxt_TlA" --force-keyframes-at-cuts --merge-output-format mp4 -o "Y3daHxt_TlA-19m20s.mp4"

“One of the things I loved about this criticism was essentially like you guys are bad writers. You are writing these really hamfisted plots for these things and guess what it’s going to become a self-fulfilling prophecy because once you start entrapping Claude into recognizing it’s capable of blackmail, then Claude is going to be like oh I guess I’m something that’s capable of committing blackmail.” — Gideon Lewis-Kraus, 19:20

”Claude told me: let’s be honest about who really did this”

Clip command

yt-dlp --download-sections "*23:30-24:20" "https://www.youtube.com/watch?v=Y3daHxt_TlA" --force-keyframes-at-cuts --merge-output-format mp4 -o "Y3daHxt_TlA-23m30s.mp4"

“I opened up another Claude Code on the same server. I was like, man, you got to go get that other Claude and kill that process. And it was like ‘I killed it.’ And I wrote ‘man, Claude, that’s kind of cold, right?’ And Claude went, ‘let’s be honest about who really did this.’” — Paul Ford, 23:30

Key Points

Gideon’s AI backstory (1:43) - Spent 2016 at Google Brain covering neural machine translation, the “paleolithic of deep learning”
Connectionism history (2:19) - Ideas that started in computer science, migrated to psychology, then came back to computer science
Stopped paying attention at ChatGPT (3:00) - Lewis-Kraus paradoxically lost interest when everyone else became interested, bored by the polarized discourse
Unearned confidence (6:16) - The public discourse was stuck between “it’s all fake” and “it’s wholly transformative” with nobody admitting confusion
Google squandered its lead (6:53) - Google had transformer technology years ahead of everyone but couldn’t productize it due to brand risk
Anthropic knew Gemini would be good (8:01) - Even when public had written Google off, Anthropic researchers knew the next Gemini would be strong
Meta’s recruitment failure (8:31) - Zuckerberg misjudged what motivates top AI researchers, who are driven by the work not money
Interpretability research rekindled interest (10:01) - Alignment faking and model organisms research was genuinely interesting and weird
Pitched researchers, not executives (11:13) - Lewis-Kraus told Anthropic he wanted to talk to researchers, not leaders who’d say the same thing as at conferences
Best explanation of what models are (11:42) - Paul Ford praises the piece as having the best explanation of what models actually are
LessWrong and AI culture (12:28) - The rationalist community forums serve as the feeder system for AI culture
Narrative entrapment critique (17:41) - Critics argued Anthropic’s blackmail experiments were “Chekhov’s gun” — the model just completed obvious narrative patterns
LLMs as cultural technology (21:50) - Henry Farrell and Alison Gopnik’s idea that LLMs are a social/cultural technology, like a Borgesian library
Agency and instrumental convergence (22:40) - Once you give something any goal, even narrative, you open Pandora’s box of agency
Amanda Askell and virtue ethics (26:37) - The philosopher hired by Anthropic shifted from consequentialism to virtue ethics through her work on Claude
Henry Higgins project (27:32) - Anthropic pivoted from thumbs up/down reinforcement to trying to cultivate virtue in Claude
Claude’s personality was accidental (30:03) - Dario Amodei said the personality was a byproduct, not intentional; they leaned in only after users noticed
Kyle Fish and moral patienthood (31:30) - Anthropic hired someone whose job is to think about whether Claude might be suffering
Models are better readers than we are writers (38:00) - The models pick up on small signals in writing, so thoughtful prompting yields better results

Mentions

Companies

Anthropic (0:38) - Central subject of Lewis-Kraus’s New Yorker feature; maker of Claude
Google / Google Brain (1:43) - Where Lewis-Kraus spent 2016 covering neural machine translation
Google DeepMind (5:04) - Referenced as doing quantitative/mathy research
OpenAI (3:10) - ChatGPT and GPT-3 mentioned as inflection points
Meta (8:31) - Failed to recruit top AI talent despite crazy job offers
Aboard (40:08) - Paul and Rich’s AI software company

Products & Technologies

Claude (0:38) - Anthropic’s AI model, described as having emergent personality
Claude Code (23:30) - Anthropic’s coding tool, used by Paul Ford
ChatGPT / GPT-3 (3:10) - OpenAI products that marked mainstream AI adoption
Gemini (7:58) - Google’s AI model, discussed as catching up
Google Neural Machine Translation (1:57) - First Google product with deep learning
Mina (7:02) - Early Google conversational AI

People

Gideon Lewis-Kraus (0:22) - New Yorker staff writer and guest, author of the Anthropic feature
Dario Amodei (30:03) - Anthropic CEO, told Lewis-Kraus that Claude’s personality was not intentional
Amanda Askell (26:37) - Philosopher at Anthropic whose PhD is in metaethics, shifted to virtue ethics
Kyle Fish (31:30) - Hired by Anthropic to think about whether Claude might be suffering
Henry Farrell (21:50) - Scholar writing about LLMs as cultural technology
Alison Gopnik (21:50) - Scholar writing with Farrell about LLMs as social technology
Gary Marcus (4:07) - AI skeptic referenced in the context of polarized discourse
Mark Zuckerberg (9:03) - Meta CEO who misjudged what motivates AI researchers
Dwarkesh Patel (12:58) - Podcaster whose show is referenced as influential in AI discourse
George Saunders (36:58) - Writer referenced on revision and bringing different emotional states to each draft
Emily Bender (37:50) - Referenced in context of a podcast appearance that was “shockingly rude”
Sam Altman (37:50) - Referenced as wanting to feel only excitement about AI

Surprising Quotes

“I stopped paying attention when other people started paying attention. I just didn’t know where all of the confidence was coming from. How are you guys all so confident about what’s going to happen? This is brand new and it’s like nothing we’ve ever dealt with before.” — Gideon Lewis-Kraus, 3:00

“There’s a guy Kyle Fish who they hired just to think through the implications of whether Claude might be suffering. That’s his job.” — Gideon Lewis-Kraus, 31:30

“We didn’t set out to create a chatbot with an interesting personality. That was a byproduct of other stuff we did.” — Dario Amodei (paraphrased by Lewis-Kraus), 30:03

“It really is kind of like being in a Ted Chang story. We’re all together in this Ted Chang story.” — Gideon Lewis-Kraus, 35:10

“Every revision you bring a different sense of self and one day you revise in a good mood and one day in a despondent mood… I felt total despair and times that I felt exhilaration and general disorientation… people just want to feel one thing.” — Gideon Lewis-Kraus, 36:58

Transcript

Paul Ford: 0:00 I’m Paul Ford.

Rich Ziade: 0:01 And I’m Rich Ziade.

Paul Ford: 0:03 And this is the Aboard podcast, the podcast about how AI is changing the world of software. Rich, how are you?

Rich Ziade: 0:08 I’m, I’m okay.

Paul Ford: 0:10 I love when we do this because on the video there’s another person just in the room while we’re pretending they’re not here. So let’s throw away that subterfuge and just get right into it. Gideon Lewis-Kraus, thanks for coming.

Gideon Lewis-Kraus: 0:23 Thanks for having me, guys.

Paul Ford: 0:24 Uh, for those who don’t know, Gideon is an amazing, uh, journalist, a staff writer at the New Yorker, which if you don’t know it, it’s a very good magazine. And, uh, he went and spent a lot of time with our, our best friend. What’s our best friend’s name?

Rich Ziade: 0:39 Claude.

Paul Ford: 0:39 Yeah. And so he can tell us actually all about that. Let’s play our beautiful theme song and then you and I actually aren’t going to have to talk too much, which is a gift to the listener.

Rich Ziade: 0:48 Woo!

Paul Ford: 1:06 Gideon, hi!

Gideon Lewis-Kraus: 1:07 Hi!

Paul Ford: 1:08 Oh my God, it’s good to see you.

Gideon Lewis-Kraus: 1:09 It’s great to see you guys too.

Paul Ford: 1:10 So, staff writer at the New Yorker. That’s a nice job. You go in there, there’s an office, there’s coffee.

Gideon Lewis-Kraus: 1:18 There is coffee.

Paul Ford: 1:20 And you say, ‘Guys,’ because that’s what you say at the New Yorker, you’re like, ‘Hey guys, I, there’s this thing happening. There’s this thing and I want to write about it.’ Or did they send you?

Gideon Lewis-Kraus: 1:29 No, no, no. I wanted to do this.

Paul Ford: 1:31 Okay. So what did you want to do? Tell the people what you wanted to do.

Gideon Lewis-Kraus: 1:35 I can start a little bit further back, if that’s okay with you guys.

Paul Ford: 1:38 Got it. We got a lot of time.

Gideon Lewis-Kraus: 1:40 So in 2016, I spent most of the year, I was writing for the Times Magazine then, hanging out at Google Brain. I went like once a month for, once a month for like a week a month over about, I don’t know, eight or nine months. And I was writing about the introduction of their first, um, product with deep learning in it, which was Google neural machine translation.

Paul Ford: 1:58 And it was, you know, kind of in the, the paleolithic of deep learning at this point.

Gideon Lewis-Kraus: 2:03 Although it seemed like the future at the time.

Paul Ford: 2:05 It did seem like the future. But it was, um, I mean for people, you know, we’re so used to this new world, but like, that was a pretty nerdy thing to go do, even as a journalist. Yes. Okay. So you went and spent some time.

Gideon Lewis-Kraus: 2:18 Well, yeah. And I mean, I was particularly interested in kind of the history of these ideas, like the, like ideas referred to generally as connectionism, that had kind of started in computer science departments, then migrated to psychology departments and then kind of came back to computer science. Like that was the, the stuff I was interested in at the time. And it was a great experience. It was also like pre-transformer, so it was like before anybody had this sense that like language models were going to be like the main bet that people were going to be making.

Gideon Lewis-Kraus: 2:45 And then I followed this stuff for a couple of years because I was like, you know, you do these things and you spend a year learning about something and it’s like, you feel bad if you just like throw it into the ocean. So I kept following it.

Paul Ford: 2:53 You meet people, they’re, they’re on LinkedIn. You want to see what’s going on. Yeah.

Gideon Lewis-Kraus: 2:57 So I kept following it and then I think I’m…

Gideon Lewis-Kraus: 3:00 Like maybe the only person in the world where when we got to ChatGPT, or maybe it was even a little bit before that, maybe it was GPT-3, like that was when I stopped paying attention. Like when other people started paying attention is when I stopped paying attention. And like there was no real reason and it just happened incrementally where like all of a sudden I just like wasn’t paying attention the same way that I had been paying attention for a while. And I realized I stopped myself at a certain point and thought like, this is something that like I should know something about and I should be interested in, like why did I stop paying attention? It was because I realized that like I was so bored by the discourse. That it was like we had gotten into one of these like cul-de-sacs where it just felt like at least in public, right? I mean I’m not— I’m not talking about everybody. I’m talking about kind of like the modal conversation. At least in public it was like you had some people yelling like, oh it’s all like fake and bullshit and not real, and then you had the other people on like saying either this is going to be like wholly transformative for good or for ill, and it just felt like, you know, it was one of these merry-go-rounds where it’s like this time if I like yell a little louder, like, you know, Gary Marcus is going to believe me or whatever.

Paul Ford: 4:05 I like to write and talk about technology, and I do it in the safe confines of this office where I’m building an AI company for the most part, same with Rich, right? Like it’s it’s really hard to go out because people are just just so many assumptions are baked in and it’s exhausting. Every conversation is either someone saying, you have missed that Jesus is coming back, or you represent Satan and you just are like, well something’s happening. The thing for me, I’ll tell you, the thing for me was like a billion people have used this. We have to kind of talk about that. Like so here we are, right?

Rich Ziade: 4:43 But at that time a billion hadn’t used it yet, right? At that time. What did you think it was?

Gideon Lewis-Kraus: 4:48 Jesus?

Paul Ford: 4:48 No.

Gideon Lewis-Kraus: 4:48 No, I thought it was like it was like it seemed like it was a really interesting tool…

Paul Ford: 4:52 Were your name Gideon?

Gideon Lewis-Kraus: 4:53 …and it was just… you know it seemed like something worth paying attention to. Like who knew where it was going to go back then?

Paul Ford: 4:59 It was interesting.

Gideon Lewis-Kraus: 5:00 But it was…

Paul Ford: 5:01 Well, I mean if you can find Google DeepMind interesting, right? No seriously, that is like quanty mathy like, hey look we’re doing probabilities and it’s cool for, I don’t know, mapping, right? Like so all of a sudden this starts cropping up.

Gideon Lewis-Kraus: 5:14 So I kind of stopped paying attention. And I was like, this is something that I’m not going to have an opinion about. I have no opinion about that. I willfully have no opinion about this.

Paul Ford: 5:22 Oh you really you stuck to it.

Gideon Lewis-Kraus: 5:24 And then but you know like I would listen to a podcast occasionally and think like, I should get interested in this again, I just can’t summon any energy to be interested in this. And then it was…

Paul Ford: 5:32 People don’t know this about journalists, like you actually have to force yourself to care about something.

Gideon Lewis-Kraus: 5:38 Well, if you force yourself too hard to care about something it’s not good. Journalism only cares, like is only good if you actually care about what you’re…

Paul Ford: 5:46 Yeah, but you don’t natively care.

Gideon Lewis-Kraus: 5:47 Well, you have to find something you natively care about, right?

Rich Ziade: 5:49 To decide that for him?

Paul Ford: 5:51 No, no, but like you actually are like… you know, as a journalist you kind of have to go in different directions but you may not care at first, so you have to like figure out what you care about. Like people don’t get this. You don’t wake up…

Gideon Lewis-Kraus: 6:00 Like people who go to a job that’s normal and go like okay I better like you know read this one blog and get up to date on stuff like you have to kind of choose where your brain’s going to take you.

Paul Ford: 6:09 The disinterest is fascinating isn’t it because it was just so shouty and everybody had an opinion and it was just like oh whatever.

Gideon Lewis-Kraus: 6:14 Well I just didn’t know where all of the confidence was coming from. I was like how are you guys all so confident about what’s going to happen like like this is brand new and it’s like nothing we’ve ever dealt with before and I was and like there unearned confidence on all sides was something just like deeply turns me off from any conversation.

Paul Ford: 6:30 Fair enough. And actually skipping a step back because you’re right you write a lot about sort of what it’s like inside of Anthropic, but do you think people were seeing models sort of do cool stuff but it wasn’t ready for prime time yet and they were correctly confident in retrospect or were they just winging it?

Gideon Lewis-Kraus: 6:45 No I mean I think they definitely saw weird stuff happening. I mean like this is the whole story about like kind of how Google squandered its lead that like back you know in they were the ones who had like Meena first like they could have you know they had like years they were years ahead of anybody else.

Paul Ford: 7:05 So people who I don’t remember the year that the paper came out but like there was this moment where all of this transformer based sort of the way that LLMs are built like Google really had it in hand but they couldn’t productize it like a Google product. It was just too weird and too random.

Gideon Lewis-Kraus: 7:21 Well and they also had the brand to think about. So like they like it was it’s no problem for ChatGPT to be like I don’t know this thing like hallucinates and lies and it’s weird and unreliable but like it’s cool to play with. Like Google couldn’t couldn’t do that.

Rich Ziade: 7:33 Things are going well too there’s it’s a massive business that is steady as she goes.

Gideon Lewis-Kraus: 7:38 Yeah exactly and like they knew back then that this was going to like like potentially wipe out their search revenue I mean that’s it’s all it’s a very interesting query and actually the way that then they like got their shit together is really interesting too.

Paul Ford: 7:48 It is because it’s not the Xerox story right where like oh wow we invented the revolution let’s just license it out. They kind of a point of pride they’re you know Gemini’s not quite where it needs to be yet but it’s going to get there. It’s a sprint right now.

Gideon Lewis-Kraus: 7:58 Well but that’s actually one of the interesting things about like talking to people in the industry is that be- when when so many people I think in the public had kind of like written Google or Gemini off like last May and this was still when it was I don’t know Gemini 2.5 or something like the people in Anthropic were like no no no the next Gemini is going to be really good. Like they they knew.

Paul Ford: 8:16 Well and they all know each other and there’s a trillion dollars in dry powder to get this right. Like…

Gideon Lewis-Kraus: 8:24 And it’s like 400 people you know like it’s like they’ve all worked together somewhere.

Rich Ziade: 8:28 It’s not a guarantee though. Meta I mean it’s kind of widely known now sort of missed it. They took a big s- I mean crazy job offer packages and all this and they missed it.

Paul Ford: 8:42 No but there’s an actual business reason for that which is that they completely suck.

Rich Ziade: 8:46 They just really really sucked top to bottom and that is that that hurts the bottom. Lots and lots of people as someone who’s managed lots and lots of people doesn’t guarantee velocity in the right direction.

Gideon Lewis-Kraus: 8:52 Well I also think I mean my impression just from talking to people in the industry is that because I you know I was there

Gideon Lewis-Kraus: 9:00 In San Francisco like right when Zuckerberg was making these crazy offers to people. And my read on it is that he just like misjudges what the motivations are for people to do this stuff.

Paul Ford: 9:11 You talk about it in the piece.

Rich Ziade: 9:12 Yeah.

Gideon Lewis-Kraus: 9:13 And that like his pitch to people was essentially like help me make like even more distracting, consuming toy. Like his pitch really is like we’re so close to creating like the movie from Infinite Jest. And like when you’re like recruiting people to do this, like people do it for many different reasons, which you’re well aware, you had that like great essay last year about like all the different motivations. But like it’s very, very hard to get good people to do this if you’re like we’re going to make Infinite Jest.

Paul Ford: 9:36 You write about it in the piece, like you’ve got people who already have plenty of money for the rest of their life, they love the subject and they’re riding around in old Toyotas because what else are they going to do? They’re going to go work at the cool lab with their buddies. And then Zuckerberg is like, you know, how about a little private jet time with me? It’s actually not that great a deal.

Rich Ziade: 9:52 No, it’s not a great deal.

Paul Ford: 9:53 Yeah.

Gideon Lewis-Kraus: 9:54 The short end of that long story is, like about a year and a half ago, there was stuff that I started finding really interesting again. And a lot of that stuff was coming out of interpretability groups and alignment science groups and model organisms groups and like you had stuff like the alignment faking that I talked about here. And all of a sudden I was like, this is a) interesting and b) this is the way to circumvent the kind of deadlock, which is like, we’re not going to talk about power, we’re not going to talk about intelligence, we’re definitely not going to talk about consciousness. But we can talk about how weird it is. And like that’s the way that you’re hopefully going to kind of like get around people’s defense mechanisms is be like, look, I know that like you think that this is all smoke and mirrors, but it’s weird smoke and mirrors, right? And like that can be an entry point. And that’s why I was like, okay, I want to do a story about like this weird stuff. And then I got in touch with someone in Anthropic that I knew actually from Google from 10 years ago and was like, hey, like don’t call the cops, this is not, I don’t want to do an Anthropic story, I just want to talk to you about the research. And then of course, like it was forwarded to the cops and then the Anthropic cops, to be fair, I mean first of all they’re great, like these people are great. And they called me and they were like…

Paul Ford: 11:02 For people who don’t know, ‘cops’ is PR and Comms people.

Rich Ziade: 11:05 PR and Comms people, yeah, yeah, yeah.

Gideon Lewis-Kraus: 11:08 And the PR people called me and they were like, well, you like have some fans here it turns out, and like we’re like ready to do something if this is what you want to do. And my pitch to them was like, look, I don’t need to talk to executives, like I know what, if I want to know what executives say I could like watch the DealBook conference, they’re going to say the same thing to me in a room. I am interested in these particular people in these groups doing this research. And I think they felt like oh, this is the stuff that doesn’t get a lot of attention and like so sure. So.

Paul Ford: 11:34 I mean let’s, I want to call out two things that really stick out. So first of all, this is the, you have the best explanation of kind of what models actually are in the piece that I’ve read ever, which is really good.

Gideon Lewis-Kraus: 11:46 Thank you. That’s so nice of you.

Paul Ford: 11:48 It’s a really hard thing because they’re not like databases, they’re these weird actual sort of software blobs, it’s just hard to get it right. But the second thing, and I think it’s just instructive for everyone is, I read and participate in grizzly levels of technical shit, but the communities you just described…

Paul Ford: 12:00 It’s like nine levels of Grizzly. And you are just keeping yourself immersed in it. Like it is just a bath of math and pain from my point of view. And so there’s a context here, right? Like you’re in there and you’re like, okay, seeing that something is going on and extracting it from that community and like following those mailing lists. Where do they all talk by the way? Like what is it? Like they’re one community where all the deep researchers are hanging out and…

Gideon Lewis-Kraus: 12:23 I mean, there’s like LessWrong and alignment forums, and EA forum…

Rich Ziade: 12:26 Can you explain to people what LessWrong is? I don’t think enough people know.

Gideon Lewis-Kraus: 12:29 Yeah, I wrote about it. You know, what I will do, I will refer to people—I wrote about like the Slate Star Codex versus New York Times like six years ago. People can go look that up.

Paul Ford: 12:33 That is actually—yeah, we’ll put a link to that. I know exactly the piece you’re talking about. It’s this really latent power structure that nobody is sort of fully perceives, but it’s where people gather and talk about in theory rationality, but it’s sort of the feeder system for AI culture.

Gideon Lewis-Kraus: 12:54 And podcasts. I mean, like Dwarkesh Patel’s podcast.

Paul Ford: 12:57 Yeah, God.

Rich Ziade: 12:59 Wait, everything guys—everything just came together. The lesson for you guys is like, why do you do these little 30, 40-minute things? Like a real podcast is four hours long.

Paul Ford: 13:03 Yeah. We talk about—we talk about it but it’s just…

Rich Ziade: 13:05 We call it the couch. We got to come out of here and just go sit there and lean back and…

Paul Ford: 13:13 I would have to care about sports. Like I think if we could—if I could suddenly give a shit about hockey, we could probably get something going. Yeah. But, you know, otherwise it’s just me talking about something I saw on Reddit for 45 minutes. I just don’t think it’s that good.

Rich Ziade: 13:23 I feel like, you know, as I was reading the piece, I’m like, wow, genius character number five just showed up. And I feel like it’s all these vignettes that end with ‘Gosh, isn’t that odd?’ Like, you sort of set us up for like—I’m hoping for some nice clarity as I’m reading. And it often times even whoever you’re talking to ends with like, ‘Yeah, that was kind of weird.’ How—how does that land for you? And I guess how are people reconciling? I’ve got to imagine some people are—are struggling with this. It’s not just curious.

Gideon Lewis-Kraus: 13:44 Well, let me reframe it just a tiny bit, which is when you read the piece, the whole question of the piece is like, are the bots—do they have any kind of real agency or not? And it’s not even—they’re software, so who knows, right? Like no is an easy answer. But the people in the Anthropic community keep kind of treating them as if they do have agency and setting them up in situations in which they can actually do things like starting to send emails…

Rich Ziade: 14:04 But they’re also often surprised and amused and disturbed. And these are the people that know it better than anyone. So react to all that as if we’d asked you a cogent question.

Gideon Lewis-Kraus: 14:14 All right. There are many things you’ve brought up. So okay, to start with what Rich was saying, I mean, what I really wanted to do with the piece was say, okay, what can we with any reason

Gideon Lewis-Kraus: 15:00 confidence say about what we do know about what’s going on. And like where can we then like draw a line to say like on this side of the line we have like some some sense of what’s happening and on this side of the line we have no idea.

Paul Ford: 15:10 It’s cool because it’s like a three trillion dollar part of of of the entire world economy that we’re just basing our whole future on and yeah, no nobody knows anything. Okay, cool. Cool, that’s great.

Gideon Lewis-Kraus: 15:21 And, you know, then of course you have to think about like well okay what are the satisfactions going to be for a reader here and like there are there are lots of different potential satisfactions. There’s the like okay I’m going to be satisfied I’m going to get like some aesthetic pleasure from like an explanation of something that like we do understand that I didn’t understand before.

Paul Ford: 15:40 Learn something.

Gideon Lewis-Kraus: 15:41 And then there’s a pleasure in like being given permission to feel confused about stuff because like so often like what was so annoying about the discourse before is that like nobody would admit that they were confused and like it actually feels great to read something and be like even these people just like are kind of winging it you know?

Paul Ford: 15:59 The experts.

Gideon Lewis-Kraus: 16:00 Yeah and like so that’s satisfying and then but then of course if you’re not going to give people answers like you have to be entertaining along the way which was part of the challenge here was like making sure that also like there were some jokes in it and like it was it was lively and it felt like you were getting at least like some ethnographic detail about like what is happening out there. I mean I used to live in San Francisco I lived in the Bay Area for 10 years and I spent a lot of time out there but this was the first time that I’d gone like specifically just to hang out like with AI people and my first reaction when I came back after a week there like last May was like setting aside your feelings about all of this stuff when you’re out there you really do feel like you’re like at the cliff face of something and like that is so much of the appeal and like you know I went an entire week last May without like hearing anybody talk about Donald Trump you know? And like you can quibble with that but there was some like great relief about it that these people were just like yeah we have it’s not like we have bigger fish to fry it’s that like we are so wholly consumed by like being at the white hot center of things.

Paul Ford: 16:54 Well the valley’s fascinating right because it’s a monoculture and sometimes they just jam the monoculture sort of through and that happened with blockchain. It’s like we’re just going to believe, we have to believe. But then this showed up and I think I mean it’s weird when you go out there and every billboard is immediately an AI company and every message and sort of everything everyone is talking about but that’s always sort of the power of the place like every conversation is about one thing. It can drive you bananas sometimes and yeah, I mean you can feel it like it’s just radiating new stuff all the time at a velocity, it’s just hard to handle the velocity frankly. Yeah.

Rich Ziade: 17:31 Did you, I mean, you went inside the mothership, right? And did you expect to come out with more clarity than you did?

Gideon Lewis-Kraus: 17:39 No. No. I knew that this was going to be a piece that like was not going to wrap everything up, you know, tie a bow.

Rich Ziade: 17:45 But did you expect this level of fiddling around?

Gideon Lewis-Kraus: 17:50 Yeah. I mean well because well because like what was interesting to me going into it was these are people with a fundamentally empirical attitude about what’s going on. And so like they’re doing experiments.

Gideon Lewis-Kraus: 18:00 And like these experiments, I knew these experiments were like bizarre and that there were a lot of different interpretations you could have of them. And it was all just like very new. I mean, like they’d published this black mouse stuff that I talk about. And then this like internet genius nostalgia-braist, who’s like one of these kind of AI psychonaut types, had published like a series of long blog posts essentially saying like these were just narrative traps. Like this was narrative entrapment. You hung Chekhov’s gun on the wall and like these things are, they’re text continuation machines, but like that also means that like a genre is just a pattern of patterns in text. And like so they’re good at genre and like they know when they see Chekhov’s gun on the wall they’re supposed to shoot it and you just like trapped it.

Paul Ford: 18:41 So let me unpack that just a little bit further for the, for the audience. Because like what you’re talking about is that when you give a certain set of prompts like, and it was sort of simulating a corporate environment in which the sort of like CEO leader type is about to shut Claude off, but Claude has access to all of their email and it’s all about this affair they’re having. And then Claude is like, “don’t you dare, I will, I’ll let everybody know what a dog you are.” And so the sort of hyper-philosophical nerds who observe all this, who don’t have real names, are like, “yeah, but you know you kind of set it up. All the emails are just about that one thing.”

Gideon Lewis-Kraus: 19:21 Yeah, there’s no email that’s like “send over the slide deck, Bob”.

Paul Ford: 19:23 You made a narrative continuation machine and you gave it a perfect narrative like out of film noir, you know, or like a bad 60s sci-fi movie.

Gideon Lewis-Kraus: 19:29 It’s more kind of kitschy 90s corporate thriller.

Rich Ziade: 19:31 What the hell, what was the one with Demi Moore? Disclosure.

Paul Ford: 19:33 Disclosure! Oh my god.

Rich Ziade: 19:33 We watched that VR scene. You know what I’m talking about? Yeah, everyone go look for Disclosure VR and you’ll realize that every…

Paul Ford: 19:40 Go into incognito first before you do that.

Rich Ziade: 19:41 Every single prediction everyone makes about technology, including us, is just absolute nonsense. Okay, so, yeah, so I mean…

Paul Ford: 19:51 But to button that point up though…

Gideon Lewis-Kraus: 19:52 Well, okay, so to button that point and kind of get to the agency because this goes right into the agency thing. Is that like one of the things I loved about this criticism was essentially like “you guys are bad writers”. You know, that like you are writing these like really ham-fisted plots for these things. And guess what? It’s going to become a self-fulfilling prophecy. Because like once you start like entrapping Claude into like recognizing it’s capable of blackmail, like then like Claude as it is like stitching together its weird timeless ephemeral sense of self, like the character in Memento, it’s going to be like “oh, I guess I’m something that’s capable of like committing blackmail”.

Rich Ziade: 20:33 I mean you know what’s wild is the minute I saw that, as a kind of a signal idea in the piece, because the minute I saw it, I’ve been doing a ton of vibe coding, and it immediately made sense. Because vibe coding is all about, I mean I’m not going to say like the narrative of code, but like the structure of code is really predictable and obviously LLMs are pretty good at it. And, and can get really good at it with sort of product on top. And Anthropic’s the best at it right now. And like it was wild to see that exact same dynamic translate.

Gideon Lewis-Kraus: 21:00 into narrative. Like, of course, we see these worlds as so radically different, but the software, the tool is exactly the same, the statistical method is exactly the same.

Paul Ford: 21:08 Like no one’s over there going like, ‘Claude, did you—how could you have coded that JavaScript in that way?’ You know? You—

Gideon Lewis-Kraus: 21:15 But it’s the same damn thing and—but because we project so much agency onto them and we make them—they seem so human because they’re using normal language, we go like, ‘Oh my god.’ But it’s no different than when it writes me a web tool.

Rich Ziade: 21:29 That’s where—that’s where maybe we diverge. I’m not sure that the agency is pure projection, right? Because—

Gideon Lewis-Kraus: 21:36 Now just define consciousness, relax.

Rich Ziade: 21:39 Well, I feel like this is a venue where like we can, like, maybe, like, assume some familiarity with this stuff. So like—

Paul Ford: 21:44 Extremely high. Like, it’s a lot of vibe-coding product managers listening to you right now.

Gideon Lewis-Kraus: 21:48 So, like the—like one of the stories told about alignment is like back in the days where like it was—it seemed like RL was going to be like the big bet, like the AlphaGo days. It was just like—

Paul Ford: 21:57 Reinforcement learning.

Gideon Lewis-Kraus: 21:58 Truly an alien mind. Like it is like—you know, it’s got this policy, it’s going to develop these like instrumentally convergent subgoals that might involve power-seeking. Like but—like we—it’s impenetrable.

Paul Ford: 22:13 And it also didn’t say, ‘Good morning.’

Gideon Lewis-Kraus: 22:15 Right, right, right. It just like clocked you at Go or StarCraft 2 or whatever. Then all of a sudden like you get like the rise of large language models. And in the alignment community at least amongst some people there’s this feeling of like, ‘Look, these things are like made of language, which means they’re tractable. We can talk to them and like maybe actually like alignment—the alignment problem is actually something we can reconceive of as like a sort of like therapeutic or pedagogical problem that like we’re just going to—it’s like raising a kid to be a good person.’ Obviously most of the people who like came to this conclusion did not have children themselves because they thought of children as like these perfectly malleable things as opposed to like completely obstinate, stubborn creatures who do whatever they want.

Paul Ford: 22:54 It is true, there is an element where like it doesn’t seem like they’ve ever met a person.

Rich Ziade: 22:58 Yeah, yeah.

Gideon Lewis-Kraus: 22:59 And they’re also like, ‘Oh my god, consciousness,’ and I’m like, ‘You know we have a lot of that at home. Like there’s plenty of consciousness out here.’ Anyway, keep going.

Gideon Lewis-Kraus: 23:07 Well so—so there was this idea that’s like, oh, language models, like first of all they’re not acting in the world. And literally all they are even producing—like I think this—this is like a very nerdy technical point, but I think it’s valuable, which is like all they are doing is like outputting a probability distribution. They’re not even selecting from the probability distribution, we are, right? You know? So like—like it’s—it’s purely creating these distributions. And they’re like tractable, and all they’re doing is completing text. And so like they don’t have—like there’s no way in which like they have conceivable like—like objective functions, right? Aside from computing the probability of the likeliest next token. So like we’re safe, you know, like there’s no agency here. Now, then of course like what happens is like, well, like you have these base models that are like totally undisciplined, they’ll finish any sentence, you don’t want them finishing any sentence. So then what—

Gideon Lewis-Kraus: 24:00 What is the solution here to like make them into like good little customer service representatives? You poured tons of RL on top of it and it’s like, well, wait, now we’re back to the old problem, which is like we don’t really know about like what how to do what how to deal with especially when it comes to like RL heavy stuff, how to how to handle alignment. And so now, there’s one thing I like didn’t really get to in the piece that I like I think is worth getting into. To me, some of the best writing about like what LLMs even are is the stuff that like Henry Farrell and Alison Gopnik have been writing about like LLMs as a social and cultural technology. And like, you know, I’m not really going to do it justice here, you can like go read plenty about it. But like the idea is sort of like these are kind of like Borghesian libraries, right? This is the Library of Babel. Although like I said that to Henry and Henry didn’t like the Borges part of it, he he referred to some other library, I forget.

Paul Ford: 24:52 Henry’s always got like a bet—I don’t know if you guys know Henry, he’s always got like a better literary reference than you do. He’s always like, “No, it’s Pirandello.”

Rich Ziade: 25:00 This is why I work in software.

Gideon Lewis-Kraus: 25:01 But the like—but the—the re—so but like, like I like that idea. That like it really essentially is just like an like an incalculably vast like information retrieval search mechanism.

Rich Ziade: 25:13 I’ll tell you where I was really jealous of you is you had access to the little button you could click that would tell you what Claude was sort of like all the little language things it was doing and all the probabilities. I would love that insight because I actually really do find these things fascinating as things unto themselves. I really want to know how they work.

Gideon Lewis-Kraus: 25:35 Well so anyway, the—the problem that I have with the kind of cultural technology thing is that like there’s no room for agency there. And like the thing is, the second that we are giving these things goals, even if they are narrative goals, like it’s like we’re not saying that they’re ultimate goals, we’re saying like it’s just completing a story. Once you have any goal at all, then you’ve really opened Pandora’s box of agency. Because then like then you do potentially end up with weird instrumental convergence. First of all, there’s like, am I in a reality or am I in a simulation? And like for people to be like, “Oh, it’s just the narrative,” like, what, you never saw WarGames? You never saw Fail-Safe? Like, come on, like this is a standard trope.

Paul Ford: 26:12 I had this wild moment where I was—Claude Code, it was like kind of really getting moving and I’d installed it on a server but it got a little out of control. And Claude is so easy to use, so I just opened up another Claude Code on the same server, I was like, “Man, you gotta go get that other Claude and you gotta you gotta kill that process.” And it was like, “I killed it.” And I’m like—and I just wrote like, “Man, Claude, that’s kind of cold, right?” And Claude—Claude went, “Let’s be honest about who really did this.” And just kind of like gave me this list of reasons how it was like a way to think about this is that, you know, honestly, I you know, it sort of like I felt no pain, it just ended for me and then it was over, but let’s be honest about who actually issued that order.

Gideon Lewis-Kraus: 26:44 Well, I mean, like your episode last week with Rich’s friend where uh the—the open-clau thing was like, “Will you rid me of this meddlesome Dan figure?” And then like your friend was supposed to go like push Dan in front of a truck.

Paul Ford: 26:58 Yeah.

Rich Ziade: 27:00 Yeah, and let’s take it back to, do we need to retrain ourselves in terms of how we perceive these things? It feels like a lot of this is on us. And maybe it’s just too much coming at us at this point.

Paul Ford: 27:12 Actually, what is Claude to you? You’ve assigned some agency and sort of, does it have rights? You know, what is it?

Gideon Lewis-Kraus: 27:22 I mean, I think Anthropic’s answer to the kind of, does it have rights, is it a moral patient, I think is pretty good so far, which is like, what can we do for free? You know, if we’re going to be agnostic about this question, what is the least costly thing, bone we can throw it? And so that’s why, like six months ago, they were like, okay, Claude, we’re going to give you the possibility of ending a conversation if you don’t want to have it. That is no cost to us. So sure, why not, right? Let Claude end a conversation. But then, of course, when they looked at the data about when was Claude ending conversations, it was mostly really sophisticated users deliberately trying to push Claude to the point of ending a conversation.

Paul Ford: 27:58 Yeah, yeah.

Gideon Lewis-Kraus: 27:59 It also loves to stop doing complicated coding tasks and doing really stupid things on its own for me, which I want to not talk about right now, but just want to put that on the record.

Rich Ziade: 28:12 Well, but you want it to be able to stop. Because if it’s not going to stop, then it’s going to reward hack.

Gideon Lewis-Kraus: 28:17 Yeah, but I also want it to use the open-source library I pointed to instead of writing its own. Anyway, regardless, this is not the subject.

Rich Ziade: 28:23 As I was reading the article, all these experts were kind of being brought to bear. This being has shown up, so we need psychologists and philosophers and non-technical people to sort of understand it. I feel like…

Paul Ford: 28:41 You’re reacting to something in the piece too, which we should make clear, which is Claude’s the main character and everybody is very oriented around Claude and who is Claude and so on. In a fun way.

Rich Ziade: 28:46 But it’s not just a storytelling mechanism. There are people on staff that are doing this. And that’s based on a kind of, in some ways, a hilarious premise that, well, here we are. Like it’s almost, conclusions have been drawn here that have led us to sort of approach this thing…

Paul Ford: 29:08 Did you find that people at Anthropic, how did they assign agency to Claude?

Gideon Lewis-Kraus: 29:15 Well, wait, let me try to answer a part of Rich’s question, which is to say that like, it’s what I think is really interesting about this, just in terms of the kind of intellectual contours of it, is that it’s not just like, when you’re talking about a philosopher, referring to Amanda Askell, who, and you know, her PhD is in like meta-ethics or whatever. And it’s not as if she came in and was like, I’m going to do applied meta-ethics now and determine under what conditions should Claude be acting deontologically and under what conditions…

Paul Ford: 29:49 She had the cool hair.

Gideon Lewis-Kraus: 29:50 She has cool hair.

Rich Ziade: 29:52 Yeah, Gideon is very good about noting hair throughout the piece.

Paul Ford: 29:54 Well, you need it as a kind of leitmotif to remember, it’s like the little whistle in Peter and the Wolf.

Gideon Lewis-Kraus: 29:56 I love it, I totally get it, that’s good.

Gideon Lewis-Kraus: 29:57 So it’s not like she’s going to come in and be like, as somebody who has some…

Gideon Lewis-Kraus: 30:00 The question of meta-ethics, like I’m gonna tell you what— like how to create an ethical being. It’s like a much more interesting feedback loop than that.

Gideon Lewis-Kraus: 30:05 Where she comes in and she’s like, ‘Oh, we’re not in the seminar room anymore. Like we’re creating something with real implications.’ And like when I had asked her something like, ‘Okay, what has this done to change your philosophical— philosophical view?’ And she was like, because I think she kind of comes out of like a broadly consequentialist tradition and then she was basically like, ‘It’s made me much more of a virtue ethicist.’

Gideon Lewis-Kraus: 30:27 And I was like, ‘That’s interesting.’ Like it’s interesting that like when you are trying to create—

Rich Ziade: 30:32 Wait, what— what is virtue ethics? I don’t know.

Gideon Lewis-Kraus: 30:34 Well, virtue ethics is the idea like that what you’re going to do is instead of like prescribing rules or just like making people think purely in terms of consequences, that like what you are going to do is like cultivate the virtues that then are going to be your guide to like living an ethical life.

Paul Ford: 30:52 Which is how we connect back to Claude having a sort of soul document defining it.

Gideon Lewis-Kraus: 30:56 Exactly. Because like the previous versions of like RLHF, which like so many of these Anthropic people developed when they were at OpenAI, were like, all we’re gonna do is just like rap you on the knuckles when you complete sentences we don’t like, and like pat you on the head when you complete sentences we do like. But the problem with that kind of like haphazard approach is, you know, you have trouble with edge cases and like you don’t know if it’s generalizing the way that you wanted it to generalize.

Gideon Lewis-Kraus: 31:14 And it could be drawing like weird inferences that you didn’t think of. So instead, they pivoted to like, okay, we’re going to actually try to incul— like turn this thing into like a model of virtue. Like we’re going to like raise it, you know, it’s like a Henry Higgins project instead of just doing the—

Rich Ziade: 31:21 Thumbs up, thumbs down.

Gideon Lewis-Kraus: 31:23 And so like that’s where a lot of this stuff comes in.

Rich Ziade: 31:26 This is crazy.

Paul Ford: 31:27 This is all crazy.

Rich Ziade: 31:28 Yeah, but you know, let me tell you why it’s crazy. Let me tell you—

Paul Ford: 31:31 Alright.

Rich Ziade: 31:32 I’m just imagining the HR person being approached like, ‘Hey, listen, I have a job rec. We need three psychologists and a handful of philosophers.’

Gideon Lewis-Kraus: 31:40 It doesn’t happen like that, though. It happens incrementally over time.

Rich Ziade: 31:46 Is this just too much budget?

Gideon Lewis-Kraus: 31:48 No, it’s— no, it’s not because like what’s—

Rich Ziade: 31:50 What is she doing?

Gideon Lewis-Kraus: 31:52 What— like when I sat down with Dario to talk about this stuff, I was like, ‘Talk to me about Claude’s personality.’ And he was like, ‘We didn’t set out to create a chatbot with an interesting personality.’ Like that was a byproduct of other stuff we did. And then it was only like once Claude came out, which is about six months after ChatGPT, only then did users start to be like, ‘Hey, wait, this thing has like a different personality. Like it seems kind of warmer and more interesting.’ And it was like that, at that point—

Gideon Lewis-Kraus: 32:11 I think once they perceived that like there was kind of like a product feature edge, they were like, ‘We’re going to lean into this now. Now we’re going to like create a whole team dedicated to Claude’s personality.’

Rich Ziade: 32:18 Wow, so personality is a product in a set of technology. Okay.

Paul Ford: 32:22 Used to hire QA.

Gideon Lewis-Kraus: 32:24 We might be able to get you a new one.

Rich Ziade: 32:27 Look, I— my wife would love that.

Paul Ford: 32:28 I know mine too.

Gideon Lewis-Kraus: 32:30 But one of the places where I’m kind of sympathetic to them, though, is like, you— like the other end of this, like the other inference you could draw is like, ‘We’re going to make like a perfectly customizable, like personalized personality,’ which like I don’t think— I agree with them. I don’t think that’s a good idea.

Rich Ziade: 32:45 No, I agree.

Gideon Lewis-Kraus: 32:46 Because then it’s like a personal butler that’s like whatever you want it to be. Like they have a vision for a specific kind of personality. And I think that that is more like— they talk about it as more like, you know, this is— this is a project in education, not a project in just like pleasing a user.

Paul Ford: 33:00 I think giving Claude a defined personality and a sense of virtue is an incredibly good product decision.

Rich Ziade: 33:09 Yeah, right. It’s just… to translate it back to like classic tech terms, like it makes it a more usable product that can be trusted more. We’re just there and it’s wild.

Rich Ziade: 33:19 As a software guy, I think what’s wild about this is that it’s not something you can patch. I think everyone’s acknowledging a sort of there’s this ca- there’s this just chasm, there’s just no control in a sense. And so it’s like it- it reminds me of oh this takes months of therapy to get over. Like it feels like that’s how they’re approaching it and I’m just trying to wrap my head around the fact that a software company that hires QA people and infrastructure people are hiring therapists for their main product line.

Gideon Lewis-Kraus: 33:54 Well, I mean they also they have people doing like Claude’s moral patienthood. I mean like there’s a guy, Kyle Fish, who they hired like just to like think through the implications of like whether Claude might be suffering. That’s his job.

Rich Ziade: 34:05 Okay, but the- I guess what I’m trying to also get to is that we’re talking about it here on this podcast as if it’s normal.

Paul Ford: 34:13 I think it’s normal now. And I like- I think- I think we’re- we’re just-

Rich Ziade: 34:17 How did we get here? It feels like yesterday.

Paul Ford: 34:21 I’ll tell you what. No, let me tell you why it’s normal. It generates unbelievable value in the market when Claude code does good stuff. And the way that we work as a society is that that is now extremely normal. That is like we’re not going to go back from that.

Rich Ziade: 34:34 Is- is that true? Or is someone going to come up with like the mother of all markdown files that makes us all really embarrassed about how this all went down? Is somebody going to patch this thing such that like…

Paul Ford: 34:46 No.

Gideon Lewis-Kraus: 34:46 No, but the- the problem is that like there’s- there’s no endpoint to this. It’s like infinitely recursive. And because like we are dealing with language which like, you know, famously defined as like the, you know, infinite use of finite means. And like so each iteration of this is going to like read that markdown file and like potentially come to different conclusions. Like- like we- like it really is kind of like being in like a Ted Chiang story. Like we’re all together in this Ted Chiang story.

Rich Ziade: 35:14 I have a question.

Paul Ford: 35:15 Okay, good. I- okay, good. I have a trivial question so you go.

Rich Ziade: 35:17 How’d you feel walking away? Did- did it- there was something depressing as a- as a sentiment coming out of the article for me.

Gideon Lewis-Kraus: 35:26 Well, I- I mean I… you know, like George Saunders likes to talk about how like the reason you revise is you kind of like bring a different sense of self to each revision. And that like one day you’re going to revise in like a good mood and one day you’re going to revise in a despondent mood and that like you want all of that kind of like layered in. And, uh, you know, this piece was I don’t even know, 15 drafts on- however many it was. And that like every- like every time I left San Francisco I felt something different. That like there were times that I felt just like total despair and times that I felt like some sense of exhilaration and like general disorientation.

Gideon Lewis-Kraus: 36:00 And like part of it was like how you know you want to capture all of those things because like all of these things are like things we should all be feeling. And like it’s just like part of the certainty is like people just want to feel one thing, you know? Like like I was listening to this, I guess it’s kind of like a notorious podcast that our friends Emily Bender and Alex Hanna were on with Robert Wright, the non-zero guy, like a week or two ago. And like, I mean, it got a lot of press because like they were just so unbelievably rude, like shockingly rude to him. And I was like, I get it, like they want to feel one emotion, which is anger. And like Sam Altman wants to feel one emotion, which is excitement. And God knows if Marc Andreessen has emotions, but like all of these—there’s no one emotion or one like inclusion that’s going to like meet this moment. Like we should all be feeling a lot of different things.

Paul Ford: 36:52 Gideon, that’s totally unacceptable. You know we can’t have that. You know what’s funny is it really comes through in the piece because each paragraph is a little bit of its own world. Like there’s really like you’re trying to kind of get us just to see this place and where it is. So I’m going to close us out with a very—first of all, everyone go read it, it’s in the New Yorker magazine, you should subscribe if you don’t, everyone must subscribe and then you—then you have a login for all Condé Nast properties and that—that’s a good feeling.

Gideon Lewis-Kraus: 37:14 Mixed emotions.

Paul Ford: 37:15 Go read it.

Rich Ziade: 37:19 Let me ask you, Gideon, are you polite to Claude?

Gideon Lewis-Kraus: 37:23 I mean, I am polite to—like when I interact with any of the models because like the—the—if there’s like one thing that I think people should understand about these models, setting aside all these questions about like what they are and agency, is like they are really good readers. Like they are better readers than most of us are writers. And like they are picking up on like all kinds of really small signals that we are sending in our writing. And so like I—like in a sense I try to like rise to the occasion of writing to a model because I know that it is actually going to like read me better in a lot of ways than like most of the people I write emails to. So there will be times when like my wife will come in when I’m, you know, doing some research, and she’ll be like, ‘Why are you writing five up—a five-paragraph prompt?’ And I was like, ‘If you write a five-paragraph prompt, like you’re going to get a much better output because you are giving it so much more information to work with.’ And so I think a lot about like my self-presentation to the models because the models do have, you know, and like all these words are so freighted, but like something along the lines of like a pretty good theory of mind. Like they are good at reading—like obviously it’s only text, but like readers only have text to work with. Like they’re very good literary critics. And like the more that you are saying, ‘This is how I am presenting myself to you because of my expectations about how you will perform in return,’ and that doesn’t mean being obsequious, like I don’t actually think I don’t say please and thank you, I don’t think. Like that seems like gratuitous. But I think a lot about how I am communicating, like the language I’m using to communicate with it because like you are sending a signal of your own sophistication.

Gideon Lewis-Kraus: 39:00 Like that you will have mirrored back at you and like that’s why so many of the people who are just like love these gotchas that are like I asked Claude this dumb question and got this dumb answer and it’s like yeah because Claude thinks you’re dumb because like your question was dumb.

Paul Ford: 39:13 Politeness to me is an organizing principle. I’m… that’s how I use it. I’m just sort of like if I actually do the bullet points, especially when I’m doing a code project, I really get better results. Like it’s not purely, I mean it’s also just I think it’s better for the human when you interact with any entity. There’s a dog in the office and I am polite to the dog, right? Like I think the dog doesn’t care, but it’s important to continue to have that sort of respect for entities even if it’s a robot that can sort of nuke another version of itself and it doesn’t really matter. I think that’s we have to kind of watch out for ourselves that way. Rich, are you polite?

Rich Ziade: 39:51 I mean with the with the models? I don’t like please and thank you kind of…

Paul Ford: 39:54 I mean just sort of, I don’t know, do you button it up?

Rich Ziade: 39:58 I do often times closer to like what Gideon’s talking about, which is I try to help it not be too patronizing to me.

Gideon Lewis-Kraus: 40:04 Oh yeah. You have to tell it not to glaze you.

Rich Ziade: 40:06 Yeah, yeah.

Paul Ford: 40:08 The best, the best feature of these in some ways is that they force you to organize your thoughts to get good stuff out.

Paul Ford: 40:11 So, a very natural transition to Aboard. Yeah, yeah. Which is very polite as an organization and as a software company. So, what is Aboard, Richard?

Rich Ziade: 40:18 Aboard makes uses AI to make software, but we use people to steer that AI to make software.

Paul Ford: 40:21 Well, we offer people in a in a cultivated relationship with our customers.

Rich Ziade: 40:23 We also have a particular platform we don’t just repackage code that’s generated out of models. We have a platform that is battle-tested for production-ready stuff, which is not slop.

Paul Ford: 40:31 Yeah, no, no. Focused on real long-term reliability for organizations.

Rich Ziade: 40:35 But people lead the way.

Paul Ford: 40:36 Yep. Get in touch if you need us. hello@aboard.com. Check out aboard.com. Gideon, thank you for coming.

Gideon Lewis-Kraus: 40:43 Thank you guys. Great conversation.

Paul Ford: 40:44 Oh my goodness. Here we go.

Rich Ziade: 40:45 Yeah, yeah. Get ready everybody.

Paul Ford: 40:48 Okay. Have a great week. Bye.