Published on October 27, 2025 12:35 PM GMT
Open on a studio apartment: Emilio and Jane are sitting next to each other on a bed, each with a laptop. Jane's laptop is plain; Emilio's is covered in stickers: a teal lightbulb, a pale-orange starfish-thing, a compass.
J: So what do you think?
E: What?
J: Did Catarina kill herself because she couldn't bear to give birth to a son and bring another patriarch into the world, even though she already gave a massive monologue about how suicide is the worst thing ever, or did Xoana murder her to keep her status and money and shit as the heir, but then how did she find out that Catarina was having a boy, and how did she stage it?
E: Uhh, I dunno. Wait isn't the playwright a professor at your college? Can't you just email him?
J: That defeats the whole point of literary analysis, like, the whole point of literary analysis.
E: Seems like the analysis is the pointless part then.
J: Why?
E: You might be wrong about what he meant. Or he might just have meant nothing, or meant it to be ambiguous.
J: Ok so, there's this idea called "death of the author" or "the death of the author" or, wait, no, it doesn't matter the original essay was in French anyway. Basically, you imagine the author is dead, or is that even what it says?
E: You're explaining this really well, Jane.
J: Shut up! Anyway, so the idea is that it doesn't matter at all what the author wanted. The text exists as its own thing, and that's what we get access to. If the author meant something, but wrote something else, then the something else is what we get. The author doesn't get to go back and correct our interpretations of the text. And if there are multiple interpretations of the text, we just have to deal with that.
E: Oh so, like, the text is under-specified and our interpretation is, like, incorrigible? I'm gonna have to go back and read some more Yudkowsky give me a moment.
J: Who?
E: Oh you'd like him you should tell your class to analyse "Plane Crash AKA Project Lawful AKA Mad Investor Chaos and the Woman of Asmodeus" next
Jane taps away on her laptop for a few seconds and turns slowly to Emilio
J: No.
Emilio has a cheeky grin
J: If you get any more like...this...at your new job I swear to god!
Jane and Emilio are on a double date with Vikram and Abby.
V: Internship, huh? You selling out on us?
E: Selling out? They're on our side!
V: Are they on our side? Really? After they released their latest model without complying with commitments 4 through 7 from the Detroit Summit commitments?
E: They have to stay in the game. The other companies broke eight of those commitments with their latest release.
A: Yeah I'm with Vikram here, this is like expecting Shell to invest in renewables
V: Uh actually I think Shell are investing in renewables at the moment
A: Why do you reserve all of your hatred for like, five companies!! God you do know you guys are the weirdest activists ever.
V: We are pretty much the first protest movement in history which is 90% dudes but isn't racist.
J: That's a...very strange brag
A: No Jane you should totally come! It was fun, even if it was there was that one guy who wanted to pave over the Amazon, you should bring Emilio!
E: I can't come. I might get kicked off the internship.
V: See, and you say they're on our side
E: They can't hire people who publicly compare their CEOs to the devil!
V: That was too good of a literary reference to pass up on. And nobody but your girlfriend even noticed it.
E: OK, but they can't hire activists who protest their own business model. How would they raise more capital? There are good reasons for all of their actions!
V: How am I supposed to figure out what those reasons are, when their actions are indistinguishable from just being another bad faith AI company? I simply do not have enough information to infer their intentions!
E: Look at what their CEO says! He's clearly on our side!
A: I'm with Vikram here, you should never trust any companies or CEOs.
The conversation continues.
Emilio is talking to his supervisor Molly.
M: How is the latest alignment run looking?
E: It looks decent, stable loss curves, it's just...
M: It's just what?
E: How do we know if we've given it enough data?
M: What do you mean?
E: We want the model to learn our values, right? How do we know we've given it enough data to narrow down those values?
M: Well the dataset is pretty big, right?
E: Yeah, and this final dataset is a bunch of moral parables and stories. I got my girlfriend to help write the initial seeds before amplification. She's a great writer.
M: OK, and how is the model responding?
E: At the moment, it responds exactly in line with how we'd expect an aligned model to behave.
M: Great! You can kick off early today, and no need to be in until ten thirty tomorrow for the demo. I hear they've got something special planned with our government collab.
E: Early? It's eight PM...
M: See you tomorrow!
Curtains
Discuss
