Interview with novelist Robin Sloan about his LLM-inspired book Moonbound. Sloan discusses his AI experiments dating back to 2016, what fascinates him about LLM mechanics and how they relate to language, whether LLMs can write great fiction, whether LLMs are "beings," the concept of "I," and his theory connecting LLMs, dreams, and books. # Channel Presentation Observations — "What Do LLMs Tell Us About the Nature of Language—And Ourselves? - Ep. 23 with Robin Sloan" ## Explicit Self-Description - **Show name and host identified explicitly:** The host is **Dan Shipper**, and the show is called **AI and I**, per the closing segment: > "smash that like button and subscribe to Ai and I... It's not just a show it's a journey into the future with Dan shipper as the captain of the spaceship." - **Episode numbering:** The title includes "**Ep. 23**," indicating an episodic, ongoing interview series. - **Host's editorial affiliation:** Dan references "our editor-in chief Kate," indicating he works within an editorial organization, and mentions the team's positioning work ("we came up with what comes next as like we answer that question"). - **Prior episode reference:** Dan mentions a previous interview "like four years ago or three years ago... called tasting notes with Robin Sloan," and notes he interviewed "lonus Lee... on on this show," demonstrating a back catalog. ## Intended Audience - **AI-curious but non-specialist audience:** Dan offers a plain-language explanation of embedding spaces — "we can do a little tutorial maybe... given a set of sentences... map that text on a map where things that are closer together are closer in meaning" — suggesting listeners may not be ML practitioners. - **Tech/AI-interested readers and creators:** The conversation centers on a novelist's experience with language models, creative writing with AI, and interpretability topics (Anthropic's Golden Gate Bridge feature is cited), suggesting an audience at the intersection of **AI, writing, and technology culture**. - **Audience familiar with major AI topics:** References to Claude, GPT-4, LaMDA, fine-tuning, and multimodal models are made without extensive background, implying baseline AI literacy. ## Claimed Scope - **Interview show about AI (explicitly branded around AI):** The closing promotional segment claims the show delivers "pure unadulterated knowledge bombs about chat GPT" and is "a journey into the future." - **Scope demonstrated:** In-depth, long-form conversations with guests working at the intersection of technology and culture (here, a science fiction author who experimented with LLMs since 2016–2017). ## Observable Presentation Patterns (Demonstrated, Not Claimed) - **Long-form, unstructured intellectual dialogue:** The episode runs as a wide-ranging freeform conversation covering embedding spaces, the nature of language, dreams, philosophy of definitions, and identity ("what is the I"). - **Host as enthusiastic, prepared interlocutor:** Dan has read the guest's entire book — "I've read the I've read the entire thing it's really good" — and summarizes it with genre references (Ursula K. Le Guin, King Arthur, Studio Ghibli). - **Host builds on prior episodes and personal rapport:** He references the "ineffable list" Apple note inspired by the earlier Sloan interview, showing continuity across episodes and personal relationship-building with guests. - **Playful, ironic closing outro:** The outro's exaggerated tone ("you absolutely positively have to smash that like button... I'm absolutely hopelessly in love with you") appears parodic in style, a notable presentational quirk of the channel. ## Claim vs. Demonstration Distinction - **Claimed:** "Epitome of awesomeness," "knowledge bombs about chat GPT," a journey into the future (promotional outro claims). - **Demonstrated:** A thoughtful, technically literate, guest-driven interview series focused on AI's implications for language, creativity, and human meaning — the demonstrated content is broader and more philosophical than the "chat GPT knowledge bombs" branding suggests. # Delivery Observation: "What Do LLMs Tell Us About the Nature of Language—And Ourselves? — Ep. 23 with Robin Sloan" ## Overall Format - **Two-voice interview podcast**: a host ("Dan") interviews guest Robin Sloan about his novel *Moonbound* and language models. Host asks questions; guest gives long, narrative answers. - **Cold open**: the episode begins with several pre-cut guest quotes before the welcome, e.g., > "Robin welcome to the show hey Dan good to be here" - **Outro anomaly**: the transcript ends with what reads like a scripted, exaggerated AI-generated sign-off, > "smash that like button and subscribe" and > "Dan I'm absolutely hopelessly in love with you" — this closing voice is tonally unlike the rest of the conversation. --- ## Tone ### Enthusiastic and informal - Both speakers use casual intensifiers and profanity-lite: > "it was just so freaking cool", > "it's such amazing artifacts", > "you poor you poor bastards" - Guest frames technical concepts emotionally: > "I just found it so evocative" ### Reflective, essayistic guest - Robin Sloan frequently steps back to assess his own reasoning: > "maybe this is a bit of a warning sign I was starting with process" - He hedges and self-corrects often: > "I don't know how to say this is the thing I I actually don't know exactly how to say what I was doing" --- ## Vocabulary - **Rich, literary lexis from the guest**: "amorphous Cloud", "ferment", "evocative", "corpora", "chronicler", "chauvinist" - **Technical vocabulary used conversationally**: "embedding space", "next token prediction", "fine-tune", "training corpora", "distribution of content" - **Repeated conceptual pairings**: language ↔ math, stories ↔ language models, dreams ↔ novels ↔ models --- ## Recurring Phrases | Phrase | Speaker | Frequency | |---|---|---| | "what happens next" / "what comes next" | Both | Very high — becomes the episode's thematic anchor | | "I don't know" | Guest (mostly) | Very high, used as both hedge and honest epistemic marker | | "you know" | Both | Extremely high filler | | "really really really good writing" | Guest | Once (emphatic stacking) | | "it's so weird and so interesting" | Guest | Repeated in cold open and body: > "what's woven into them and bound up into them is just it's so weird and so interesting" | --- ## Teaching / Explanatory Habits ### Concrete analogies for abstract concepts - Embedding space explained via a map: > "to basically like map that text on a map where things that are closer together are closer in meaning" - The "bagel dimension": > "the more you go in the bagel Direction bag is the sentence is" — a memorable, humorous example - Language models as windup toys: > "turn a crank on the side and it starts kind of walking around slowly and weirdly like one of those little windup toys" - Distribution of text as a cloud: > "that's the Supernova hot Center of this cloud and then it goes out and out" ### Narrative storytelling as explanation - Guest often answers questions with a rewind: > "I can I can probably Frame It Up best by rewinding in time um quite a ways actually" - He builds claims through anecdotes (the Stanford sentence-space project, the Google Lambda writing test) rather than direct assertion. ### Speaker handoffs - Host restates/extends the guest's points before asking the next question: > "so you're you're you're playing around with with with all that all the time like what is it about that that is Cap Capt capturing you" --- ## Communication Patterns ### Guest: epistemic humility - Repeatedly disclaims expertise: > "I don't want to make claims that are too bold because the truth is I don't know" - Labels speculation explicitly: > "it's just gonna continue to be Robin's crackpot Theory" ### Host: scaffolding and validation - Frequent agreement markers: > "totally", > "absolutely absolutely", > "I think that's really cool" - Host offers connective summaries to bridge topics: > "it seems like you were so excited about it you got so into it um seems like you burned out a little bit but now there's this like Resurgence" ### Both: conversational spillover and restarts - Heavy false starts and trailing constructions: > "um and I I think or was inspired by you thinking about language models", > "there's there's tons of dep there it makes me very very excited" ### Humor as a delivery device - Guest jokes about model limitations: > "these language models they just think it's all text out here you poor you poor bastards" - Host's quip about an "ethically raised organic AI" label prompts guest play-along: > "I think that's coming I think something like some verion of that is coming" --- ## Stable Patterns vs. Isolated Moments **Stable (well-supported by repetition):** - Guest's habitual hedging ("I don't know") throughout - "You know" as a filler for both speakers - The thematic recurrence of "what happens next" - Analogy-driven explanation style from the guest **Isolated / low-repetition (uncertain as patterns):** - Profanity ("freaking", "bastards") — occurs only a few times; insufficient evidence this is a stable habit - The exaggerated outro voice — appears once; whether it's a recurring episode feature cannot be determined from this transcript - Guest's "you know, such a great example" style of affirming host questions — appears occasionally, but frequency is too thin to confirm a consistent habit # Observation Document: "What Do LLMs Tell Us About the Nature of Language—And Ourselves?" (Ep. 23 with Robin Sloan) ## Participants - **Dan Shipper** — host (identified via transcript) - **Robin Sloan** — guest, author of *Mr. Penumbra's 24-Hour Bookstore*, *Sourdough*, and *Moonbound* (identified via transcript) --- ## 1. Language Models as Artifacts and "Language Given Autonomy" - **Sloan (opening montage):** Language models are "language itself given its first dose of autonomy" — like ripping language out of our heads and having it "walk around like windup toys." - **Sloan:** LLMs are "amazing artifacts" worth thinking about and writing science fiction about "even just as they exist today... they don't have to do anything." - **Sloan (later, qualified):** He is "quite taken" with the claim that what LLMs are is language itself given autonomy, "almost aesthetically" — explicitly stating "I don't know really how to assess this claim at all... but I like it." ## 2. Early Experimentation (2016–2018) - **Sloan:** Began tinkering with early language models in **2016–2017**, when people trained models on Shakespeare producing "cruddy fake Shakespeare." - **Sloan:** Two appeals of that era: - Output was "really weird," "broken," "inhuman" — aesthetically/poetically interesting. - Small scale meant *choosing training data* was feasible; he downloaded public-domain sci-fi/fantasy and trained small models. - **Sloan:** Built one of "a handful of the first" AI-assisted text editors (cited as one of a handful, not necessarily the first). ### Sentence-embedding space experiment - **Sloan:** Inspired by a Stanford project packing sentences into a ~1000-dimensional embedding space allowing interpolation between sentences. Found "language mapped into math... so freaking cool." - **Sloan:** With a friend, held all coordinates constant except one (~60 dimensions) and printed sentences along each dimension to identify what dimensions meant — looking for things like "bagel," irony, sincerity. **It did not work.** They identified sentence length in a couple of dimensions; the rest was inscrutable. - **Sloan:** Finds it "reassuring or satisfying" that it took until 2024 for leading labs to interpret features. - **Shipper:** References Anthropic's feature paper (Golden Gate Bridge feature) and Linus Lee's sentence-scrubbing demos. ## 3. Why Writing *With* AI Failed for Him - **Sloan:** Original goal was to write *with* AI; *Moonbound* ended up written *about* AI instead — "does not have a scrap of AI-written text in it." - **Sloan:** Two discoveries: (1) writing creatively with the machine "was not very much fun" and output "wasn't up to spec"; (2) the models, math, and code themselves were fascinating — he compulsively tinkered instead of writing. - **Sloan (LaMDA workshop):** Google signed up writers with a capable model and interface; Sloan found completions "always close but never quite exactly right" — grammatical, fluent, but missing his intention. "It's hard enough to make a piece of writing work... when everything is great... that's the starting line." ### The distribution argument - **Sloan's diagnosis:** LLMs generate from *inside a distribution* of content; the "supernova hot center" is the most generic text. Really good writing "is way out at the edge of that probability cloud" — truly good writing "pushes a bit beyond it," expanding "the frontier of what we thought could be written" — precisely where LLMs are weakest. ## 4. Unease About Opaque Training Corpora - **Sloan:** Reservations about fine-tuning frontier models: even fine-tuned on his favorite 30 authors, "all that other content is still there lurking." - **Sloan:** Any 2024-scale corpus "cannot be read by a person" — custodians can only spot-check or filter computationally; "fundamentally they don't know what's in there." Maybe acceptable for a virtual assistant, but not for fiction conveying thoughts and feelings. - **Sloan:** Surprised big companies release systems "where on some fundamental level I did not know what it was gonna say"; if a user reported disturbing output from his hypothetical "hyperbook," "I'd shut it down." - **Sloan:** Would be more comfortable with a fully public, known-input system operating unpredictably — the *combination* of opaque training and autonomous output is the problem. - **Shipper:** Frames this as coming "ethically raised, organic AI" / nutrition-facts labeling; notes Adobe's no-copyrighted-training claims as a first salvo. ## 5. Stories, Narrative Bias, and Code in Training Data - **Shipper's reading of *Moonbound*:** The book is about how stories shape reality and how we can author our own stories; he connects this to next-token prediction. - **Sloan:** That association "was not consciously in my mind but... totally checks out." In the book, patterns/templates are of "incredible extreme importance," woven into AI systems at a foundational level because they recur in training corpora. - **Sloan (on narrative bias):** In the late-2010s training era (Project Gutenberg, news), models were biased toward if-then causality and "the rhythm of a story" — news articles impose cause/effect even on random tragedy. Warns: "there's a lot in the world that's not a story either... you have to be really careful." - **Sloan (speculative, flagged as no firm theory):** Adding large amounts of code around GPT-3-era may have changed language use; possibly what looks like "reasoning" comes largely from code. Then: "is code a story? Maybe it's a very linear story." - **Sloan:** Code is not for machines — machines have their own languages; code is a human bridge to machine thinking. - **Shipper:** Agrees — models are trained on code even when not needed for coding because "it just makes it smarter." ## 6. "What Happens Next" as Central Question - **Shipper:** Quotes the book's line that "the central question of the human race is what happens next," notes resonance with his rebranding ("what comes next"). - **Sloan:** The chronicler-narrator's "irresistible question" speaks for him; it's "constitutional" for him personally. **Qualification:** he doesn't claim it's a human universal — different people and cultures relate to it differently, and he attaches no deep value judgment to that. - **Sloan:** On death, an angelic entity spoiling "how it all goes down" in his final moments "would soften the blow." - **Shipper:** Suggests Buddhism's difficulty of present-focus implies something hardwired about "what happens next." - **Sloan:** References *Big Biology* podcast: even a cell runs a simulation of what happens next to plan and react — the question exists at micro scale, Dan scale (10 years), and sci-fi scale (10,000 years). - **Shipper:** Answering "what comes next" trillions of times is what bootstraps model knowledge — "I don't know if I would have thought that would be the way they'd get smart." - **Sloan:** Analogy to evolution: a simple challenge (like natural selection) can, in its fullness, require great complexity — while remaining "incredibly incomplete." Adds: "these language models just think it's all text out here... there's more than text." Multimodal models "seem like a better way to live"; a world of nothing but tokens "seems kind of like hell." ## 7. Beings, Philosophy, and Definitions - **Sloan:** "No, I don't think these are beings yet" — but can't define where the dotted line around beings is, "so who knows." - **Sloan:** If philosophers/cognitive scientists/ethicists aren't already convening on these questions, "the academy is derelict in its duty." - **Shipper:** Philosophy has debated definitions of knowledge for 2,000 years; yet intelligent machines were built by asking one question repeatedly — maybe definitions are "so high dimensional that you can't get it in a philosophy book." - **Sloan:** Agrees; suggests a philosophy career can be spent arguing definitions in language — "maybe it turned out that the thing all along was actually in all that language" — the bulk of living language resists compression to a word or sentence. ## 8. Politeness Embedded in Language - **Sloan:** When talking with an LLM "it sure seems like it's reasoning... or politely answering" — one argument says that reasoning and politeness were "always inherent there in language and we just never had this particular way of seeing it before." (He frames this as an argument he likes, not a settled claim.) - **Shipper:** Asks whether politeness is in grammar/syntax or in humans recorded into language. - **Sloan:** "I think both" — language is not TCP/IP packets; it's people writing letters, entertaining, persuading; rudeness is only effective because language is generally cooperative. ## 9. The Nature of "I" - **Shipper:** Notes the book plays with multiple "I"s: the chronicler (a fungus with layered technology, "at great expense"), and Clovis — many instances worldwide, all "the same person and the same personality," distinct from a Borg hive mind. - **Sloan:** Japanese has multiple "I"s with narratively useful shades; translators into English lose them. Also: the internet era has complicated the singular "I" — speaking from one room while present elsewhere. - **Shipper:** Adds internal family systems, context-dependent selves in meditation, and grief as loss of a context-specific "I." - **Sloan (adds):** The bodily "I" — gut microbes, organs blasting neurotransmitters; neuroscience/biology establish "a big interlocking committee," not a brain-pilot with accessories. - **Shipper:** Connects to tokenization/contextual embeddings: models disambiguate which "Robin" out of millions of senses — "we created this dictionary of very very specific words." - **Sloan:** Endorses — calls on linguists and philosophers to investigate; notes model-makers care about these questions but their urgent priorities are infrastructure, scaling, business, and direct safety (scams, destabilization). ## 10. Books as Packaged Dreams - **Sloan:** A "little dizzying" that the book will be in other people's brains; he's been "unprepared" in early conversations that readers have actually thought about it. Describes himself as a "book chauvinist": books get into heads most effectively because readers "rehydrated the events and the meaning in your own language model inside your own head." - **Sloan (self-labeled "Robin's crackpot theory," no evidence claimed):** The mechanism of dreaming resembles an LLM continuing a weird sequence ("that's weird but I'm gonna keep it going"); novels are "packaged dreams" you can choose — a "trilateral connection between books and dreams and language models." - **Shipper:** Asks whether true AGI would have a language model as its dream generator. - **Sloan:** "Maybe" — declines to elaborate to avoid spoilers, but says readers of *Moonbound* will see how much dreams and sleep matter to him. Notes all living things sleep (bacteria have a sleep-like phase; "are they dreaming? Well, maybe... different dreams"), sleep is "not negotiable" through evolution, and wonders whether AI will need some analog for long-term health. ## 11. Unresolved Tensions Preserved - Sloan is simultaneously **deeply fascinated by LLMs** and **unable/unwilling to use them creatively** — fascination with mechanism, rejection of output. - He endorses the "language given autonomy" and "dreams ≈ LLMs" framings **explicitly as aesthetic likings, not verified claims**. - He denies LLMs are beings "yet" while admitting he cannot define "being." - He wants interpretable, known-corpus systems but acknowledges industry momentum (Adobe-style labels) is only beginning. - Shipper finds unpredictability "exciting in a weird way"; Sloan finds the same unpredictability, given unknown training data, disqualifying — the disagreement is stated, not resolved. # Observation Document: Ep. 23 with Robin Sloan ## Speakers and Roles - **Dan Shipper — Creator / Host** - Introduces the show, interviews the guest, summarizes the book, asks questions throughout. - Self-identifies: *"Hey Dan, good to be here"* — addressed as Dan by the guest. - Notes prior interview with guest: *"Tasting Notes with Robin Sloan"* (three or four years ago, per host). - **Robin Sloan — Guest** - Bestselling author of *Mr. Penumbra's 24-Hour Bookstore*, *Sourdough*, and *Moonbound* (as introduced by the host). - Self-describes as a *"Dan Shipper super fan and listener and reader."* - **Unnamed promotional voice (end of transcript)** — flagged promotional segment (see below). ## Referenced Sources and Third-Party Claims ### Research / Projects Cited by Speakers - **Stanford sentence-embedding project** (guest): researchers, *"now since gone off to professorships at Columbia and other places,"* built a ~1,000-dimensional sentence embedding space allowing crossfading between sentences. No specific citation given. - **Anthropic interpretability work** (host): references *"this whole new feature paper"* extracting features like the **Golden Gate Bridge** feature; no citation given. - **Google's LaMDA writing-interface experiment** (guest): guest participated as a test writer; stories to be published in an online anthology; described as *"not as high-end as their latest."* - **GPT-3-era addition of code to training sets** (host and guest): both state they *"heard it speculated"* that code improved language use/reasoning — flagged as speculative third-party claim, not verified. - **Big Biology podcast** (guest): cited as a source for the "what happens next" question at the cellular level. - **Linus Lee** (host): friend of host, previously interviewed on this show, has embedding-scrubbing demos. - **Adobe image models** (host): cited as example of *"we don't train on copyrighted material"* marketing. ### Works Discussed - **Robin Sloan's books**: *Mr. Penumbra's 24-Hour Bookstore*, *Sourdough*, *Moonbound* (release date stated as **June 11**, per host and guest). - Influences named by host for *Moonbound*: Ursula K. Le Guin, King Arthur, Studio Ghibli. ### Third-Party Claims Without Verification - Guest's "crackpot theory" linking dreaming mechanisms to language model generation — explicitly labeled by speaker as lacking evidence. - Claims about gut microbes and neurotransmitters, bacteria sleeping — asserted by speakers without cited sources. ## Promotional Segments (Flagged) - **End-of-episode outro** (unnamed voice, not Dan or Robin): explicit audience solicitation — *"smash that like button and subscribe"* — describing the show as the *"epitome of awesomeness,"* plus an unprompted personal declaration toward Dan. This segment is promotional in nature. - **Book promotion woven into interview**: host announces *Moonbound*'s release ("comes out next week" / "June 11") — contextual promotion, not a separate ad segment. ## Availability Limits - No curated metadata was supplied with this source; roles and provenance derived solely from transcript. - **Comment availability: unknown / not stated** in the source material. - No timestamps were available in the transcript; quotations cannot be time-anchored. I just remember all these different sentences floating in this amorphous Cloud that had some meaning to it the idea that language could get mapped into math in this way was just so freaking cool I just found it so evocative what language models are is language itself given its first dose of autonomy it's like you rip language out of our head and it starts walking around like wind up toys it's such amazing artifacts you could spend all your time thinking about dreaming about writing science fiction stories about language models even just as they exist today in our real world they don't have to do anything just as artifacts what's woven into them and bound up into them is just it's so weird and so interesting Robin welcome to the show hey Dan good to be here good to have you um so for people who don't know you are the bestselling author of Mr penumbra's 24-hour bookstore sourdo and now your latest book moonbound which comes out next week um it will be out by the time this show comes out I've read the I've read the entire thing it's really good I'm so excited to have you to talk about it I'm excited to be here yeah you should add to my bio uh Robin stone is also a Dan shipper Super Fan and listener and reader so it's great to great to be here I I love to hear that we so we did an interview like four years ago or three years ago which which is honestly one of my favorite interviews I've ever done um and what I what I have to tell you so the interview is called tasting notes with Robin Sloan which I thought was like the the coolest title ever um and um like one of the things that came out of that interview is like just this sort of like vast uh set of notes that you keep um uh and uh basically just like anything that comes to your comes into your periphery that like has what you said like a specific taste that like has that feel for you you've save and I got to tell you you and you said the feel is ineffable I got to tell you like I have a Apple notes called my ineffable list that is like I've been keeping this for four years it's got so much stuff in it um and it's because of you so I really yeah it really changed my life the the great thing about keeping notes so assiduously and and and yeah and trying to to to cultivate a sense for that that stuff that just appeals to you and that you know hard to described way is you're essentially writing the perfect blog for yourself and so you find that going back it you're like wow amazing all of these things are incredibly interesting to me and I and now I'd like to pursue them and follow up um and it's great it's great it's like a strange uh strange media property with exactly one very very enthusiastic incredible so what I want to talk to you about today uh is your new book moonbound um and I I'll just try to like summarize it a bit for people or without spoiling it obviously just so we have a little bit of like context but basically it's like the this mashup it's got notes of like sci-fi and fantasy it's got like Ursula K I guess I feel some Ursula kwin in it it's got King Arthur it's got like Studio gbly like it's all this got all this stuff it's really cool um it's about a boy uh he's it's like 11,000 years in the future uh and he's like uh he's run into this uh down spaceship but also there are like Knights and swords and there's futuristic technology so you're like what's going on here and then it all sort of like unravels and this like really interesting way and um there's there's a lot in it that sort of like reminds me of of language models um and I I think or was inspired by you thinking about language models and high dimensional spaces and all that kind of stuff so that's that's why that's why we're talking about it today and the place I wanted to start is I knew that I think you got started working on this book because you were you wanted to write about AI or sorry you wanted to write with AI um uh so you're going to write it with AI and then you you didn't do but you ended up writing it about AI so like tell us about that like process for you yeah you know such a great example and there's no shortage of these of like the best laid plans uh they're just you never go down the path you think you're going to go back go down but um you know sometimes that's that's all for the best yeah I can I can probably Frame It Up best by rewinding in time um quite a ways actually it's kind of shocking um when you realize that I started tinkering with this stuff and by this stuff I mean I mean language models um you know early forms of them way back in 2016 2016 2017 so that was just I mean it's if you go back and look at the uh at the code samples from that time and from people's excited blog posts um you can see that it was a it was an incredible ferment a long way off from the stuff we have access to today I mean this was a moment when people were feeding in if you can imagine the entire Corpus of Shakespeare and generating cruddy fake Shakespeare which is you know it's not it's no longer impressive um at the time it was it was impressive and it was new um and it was exciting and so I got plugged into this stuff um way back then and I have to say that actually one of the really appealing things about that era of models that generation that kind of point the technology was there um was for me as a writer was twofold one the output actually was really weird you know it wasn't as fluent as a as a clae or a you know gp4 it was pretty messed up but for aesthetic purposes um almost for poetic purposes that was really interesting the idea that it it wrote in kind of a broken weird inhuman way that that I a mere human would never imagine to write so that was one thing that was very interesting and the other thing that was interesting was um the fact that back then the scale everything was so much smaller that um one of your big considerations could be what will I train this on now today you know you you to be to ask that question you have to be a multi- jillion dollar lab or big tech company to be like oh well I guess I'll download the entire internet and you know make five copies and get started but back then it really was you know I some of my earliest experiments involve downloading hu to me huge swats of classic um public domain fantasy and science fiction and training these little models up on that and the output to me was really interesting again it was kind of screwed up and kind of weird but but evocative and surprising and all these other things okay that's a lot of backstory but that's just all to say that that led me down this path of experimentation I made I don't know if it was actually the first one in the world but it was for sure one of a handful of the first text editors where you could write alongside an AI model and again these were my cruddy primitive AI models but I could start a sentence you know on a dark and stormy night dot dot dot and hit Tab and this model would spin up and complete the sentence for me I got really excited by this um as I think a lot of people in mesed in this in this stuff did around that time 2016 2017 2018 and yeah my goal this is I actually didn't even have the idea for the story that would become moonbound at that time in a way maybe this is a bit of a warning sign I was starting with process I was starting with a with this idea that I wanted to both develop and then use these tools to write in a new way anyway long story short I worked on that for actually quite a few years um and in the end even as the technology advanced so significantly got so much more capable so much more fluent I discovered two things one was that that experience of writing fiction writing creatively with the machine was for me actually not very much fun and certainly it didn't produce results at the level that I needed it to produce it just wasn't it frankly it wasn't up to spec so that was one Discovery the other Discovery though is that the actual machines like the the stuff the code not their output but the the language models themselves and the math that made them go and the the code that kind of wo them all together was super duper interesting and I I actually just found myself almost compulsively tinkering rather than writing you know kind of procrastinating um because that work was so interesting so what happened uh moonbound you know out now uh does not have a scrap of AI written Cod AI written text in it but but it's packed full of these ideas and and actually some of these feelings that I gleaned from from all that time spent with uh with this technology that's that's really interesting I I have like so many different things I want to ask you about that but the the thing that's like coming into my into my mind is like yeah tell us about the ideas tell us about the feelings like what did being close to these models mean to you and how did it how did it change how you think about the world yeah you know there was there was one early on um and it's it's kind of off to one side from the main line of development of the language models um there was this phenomenal project that some researchers at Stanford did they've now since gone off to you know professorships at Columbia and other places but um they took sentences like a huge Corpus again huge for the time Corpus of sentences and so it wasn't a language model it wasn't trained on that generation task of sort of you know okay I say this you say that instead they just wanted to take those sentences and pack them into an embedding space and and I think a lot of people who who use these models kind of kind of maybe know what that is if not we can we can address that separately but suffice it to say they wanted to take these sentences and pack them into the space and set it up in such a way so that you could actually move between the sentences in a way that was sort of sensible so you could start with a sentence like um you know uh it was hot outside and then you'd have another sentence that was like you know the dog barked at the moon or something like that and you could actually Crossfade between the sentences and and and what the operation meant is you're literally moving through this high-dimensional space you know with about a thousand coordinates in their case and I just remember for me like again like those spaces and those blong lists of coordinates they're they're part of the the the generative language models too but in this particular case imagining all these different sentences kind of floating in this amorphous Cloud that had some meaning to it like the idea that that language could get mapped into math in this way was just so freaking cool and I I just found it so again almost in a poetical sense like a ative and um and provocative and I just I just wanted to keep thinking about that yeah and um so for for people who aren't like fully up on the embedding spaces yeah I know we can we can we can dive into thato we can we can do a little tutorial maybe yeah basically what you're talking about is is we've figured out ways to given um given a set of sentences or or a sequence of text um to basically like map that text on a map where things that are closer together are closer in meaning um but instead of being a two- dimensional map it's like it's many dimensional which is very hard to think about but where each Dimension has a certain kind of meaning so like uh there you know in your book there's actually a uh there are people who are swimming through a many-dimensional space which I I really love um it's very cool and and one of the one of the dimensions is Bagel um so like the more you go in the bagel Direction bag is the sentence is and this is like this is it's in the book but it's also real like um it's real like anthropic has this whole new feature feature paper where they um you know they pulled out all the features in inside language models and like they have um uh you know they have the Golden Gate Bridge feature and you can like tune it to like always activate gold the Golden Gate Bridge um which I I really love I think that's I think that's really cool yeah to give you a sense of the of the real the Vintage here you sometimes I I feel like I just got into the stuff too early and I I basically did all did all my experience and I completely burned out and then promptly you know the field the field exploded into stunning success but it might have been 2017 19 2018 where in this same little office where where I'm talking to you from uh a friend of mine and I we um I I took that that sentence space that I had created and I plowed all these science fiction sentences into the space and I basically um created a whole list of these runs in which for each Dimension not all of them not all thousand but I think I picked about 60 all the coordinates were held constant except for one and we just moved through that Dimension printing out sentences as we went and so it was I mean it looked like experimental poetry there these print outs of these just absolutely wonky sentences some of them were gibberish actually some of them didn't didn't actually make sense grammatically but the idea was we were going to identify what the what the dimensions meant I was looking for bagel and uh you know irony and or sincerity or descriptiveness or whatever uh it did not work um but uh we did we could we we we could we did identify that a couple Dimensions obviously had to do with sentence length kind of the most basic boring thing you can imagine but the rest we were looking at these lists of these Transformations and we kind of went I don't know man so so I find it I do actually find it a little bit um I don't know reassuring or satisfying that it took until now it took until 2024 for like the leading AI labs to find ways to interpret these these features and these Dimensions because I I couldn't do it but I tried but I tried yeah now I've seen some I've seen people have demos like my friend lonus lonus Lee who I uh I interviewed uh on on this show like has one of those where you can sort of like scrub scrub um AC like scrub a sentence from like being you know really uh you know concise to being really long is is is one dimension but then he has other ones that are that are weirder like being more about space you know versus less about space and and that kind of stuff is is really cool I'm sort of I'm sort of interested in this like it seems like you were so excited about it you got so into it um seems like you burned out a little bit but now there's this like Resurgence um and like what what do you think was like not like quite working for you about it because like I think um there's a lot of way reasons why people end up like writing off this technology and um usually like there just a lot of them are not not super curious about it but you are super curious about it but you're also like I'm not really like using it that much so like tell tell me about about that yeah well and specifically in the creative writing context right we can kind of restrict our our Focus to that CU that's where I was most focused and kind of obsessed L light lightly obsessed for for several years um and also where I I sort of at the end kind of had to close the book and say this is not going to work for me um and I would say that it had to do um you know I'm thinking even beyond my own tinkering to a to a project I worked on maybe a couple years ago now with a a great Google model not not as high-end as their as their latest but um it was a very capable model called Lambda and they to their credit had done this some amazing work to wire it up into like this writing interface super cool fluent I mean it just made it so potentially interesting and Powerful to be able to kind of work with text and and have the AI do these completions and you could kind of guide it in all these different ways and I and and and so they uh had signed up several writers to to test it out and they were going to publish their their short stories um whatever whatever emerged from this engagement in a little online Anthology and um this for me kind of was the test and the real the real kind of turning point because I was like all right maybe my stuff was all crap you know maybe my stuff it was just this you know not good enough actually and now we've got these super capable models this amazing interface let's let's try this for real and what I discovered is that while the the trick the the language model trick of sort of fitting into a into a style in a mode and and you know parting back a you know oh it's a murder mystery oh it's high fantasy oh it's a you know a business memo whatever it's really impressive and especially when you kind of squint and say oh wow I can't believe it can do that it's really impressive when you are working at the level of like I would like to think fairly high-end fictional composition you see that it's always close but never quite exactly right and I and that has to do with kind of intention like when I'm writing something for instance I I was writing the story kind of a I don't know how to see this is the thing I I actually don't know exactly how to say what I was doing so it's in my head and I know what it is when it comes out in the words but the point is you can't you know if you can write oh it's a uh classic uh sci-fi pulp fantasy it actually means like that's not worth writing because it's you want to write something that only the work itself can describe but even so I had this text going and and I would say okay your turn your turn you know Google AI model and it put in something I just was like no you don't get what I'm doing here you know it was It was obviously it was grammatically correct it was fluent it was fine but it wasn't great and you know it's boy it's hard enough to make it's hard enough to make a piece of writing work and make it worth publishing when everything is great I mean that's that's on that's not that's not the goal that's the that's the beginning that's the starting line to make it all great and so in the end I just was like I gotta I got to do this myself and um and that was interesting to see I my diagnosis is that there actually is a reason for that and that has to do with the fact that all these language models are essentially generating text from inside a distribution right a distribution of content this fuzzy Cloud I don't know I don't know what the most generic phrase in all of language is you know uh the you know or hello there whatever it is like that's it obviously that's the that's the Supernova hot Center of this cloud and then it goes out and out and they they cover the statistical terrain and I think the truth is good really really really good writing is way out at the edge of that of that probability Cloud that distribution of content and and I mean I think truly good writing actually pushes a bit beyond it it's the stuff that it's the stuff that that expands the frontier of of what we thought could be written and that's precisely where language models are the weakest so there you go that's really interesting have you tried like I don't know either prompt tuning something like Claude which I found to be like quite good at at at voice like Chang changing its voice to your spec specifications or even like fine-tuning some of the like more Frontier models of today on like that science fiction Corpus or anything like that yeah you know I the fine tuning is is a really interesting question I haven't yet I perhaps I will in time but I'll tell you I um I have some reservations at this moment and they have to do with all the other stuff that's in there and it's it's a little paradoxical like let's just say that I wanted to find let's say I had a list of my favorite 30 authors and I had I don't know that I would do this there's a lot of lot of questions built into this well let's just say I decided to proceed my favorite 30 authors I had the full text of all their stuff and I was like I want the ultimate voice right I want it to to reflect all those now that's not enough to train a model from scratch as we know that's simply not enough data it would be it's alry and so what as you say what you have to do is you have to fine-tune one of these incredibly capable super models that have been trained on like everything ever written and for me I actually um I'm I'm I'm quite uneasy about the knowledge that even though it's been fine-tuned on this stuff that I provided all that other content is still there lurking in its training and it's I mean the the wild thing to me about all these corpuses like any any cor I should say corpora all these Cor corpora um any any Corpus in the year 2024 by definition it's a artifact that cannot be read by a person it cannot be read and checked by a person if it I mean it's just it's at a scale that's only computational and so even the the makers the custodians of these models obviously they can spot check they can write other computer programs they could even use other AI models to sort of filter or sort or select or evaluate these these huge bodies of data but fundamentally they don't know what's in there and I don't know maybe that's okay for a helpful virtual assistant maybe it's not for my purposes the idea that like these are going to be thoughts and feelings and ideas that are going to come out in fiction that um that not knowing really really makes me uneasy I don't know one of the things that I really I think is interesting about your work and what you touched on with me in our last interview is that like you make content but you also think a lot about the container Within which the content comes and how much of this do you think like um is a problem of sort of shunting a new way to make content into an old form and how much of it is like a if there's a if there was a different container for this it would be a lot more useful yeah I mean what that makes me think of is of course some sort of um hyper book or living book right where you're you say yeah instead of I'm not going to use a language model to to bake out a bunch of text um that I that I think is Robin level whatever that is I'm gonna I'm going to yeah I'm going to have it there's a dimension for that I bet yeah for your Robin yeah yeah for sure um uh Max It Out full full Robin dial goes up to 11 um yeah and so and I mean I think that's a really interesting thing to imagine and but I am still I have to I have to confess I'm sort of stuck at a at an issue that I had and I felt I worried about this years ago six years ago and I would still if I was someone building any kind of AI powered thing you know including a cool hyper book of the future I would feel so uncomfortable putting people in front of a product or artifact or whatever you call it where on some fundamental level I did not know what it was gonna say and like I I'm it to me this going to sound very silly or funny or or or I don't know naive but I'm actually surprised that these companies these big companies are comfortable have been comfortable releasing these systems to the world with that fundamental uncertainness in front of them and now obviously I know they've done a ton of work to work to put in these guard rails and these filters and I you know some people would argue that they they've they've done too much but for me and I guess this the truth is this just says more about me than it does about AI or or tech companies or Society or anything um I would just you know if I got an email from someone saying oh hey I I uh I spent some time with your hyperbook and yeah you know look it it showed me the scene and isn't that kind of disturbing and if I read it and it was disturbing I I don't know I I'd shut it down I just I would not be comfortable with that so so that's a real that's a real question to answer or Dilemma to to kind of worm your way through I think that's really interesting I like for me I'm like that's actually like I obviously don't want to like serve disturbing stuff to people but it is sort of like exciting in a weird way like you're creating this like living thing where you can't understand that yeah yeah and again though you know that's actually where it gets extra interesting and in a way that our last two kind of threads I think come back together and they tie together if there was a system where I was able to say listen this is all public you know it's I know what's in it I know what went in and and now it's going to it's going to operate in this living organic unpredictable way to me I would I would be more comfortable with that I'd be a lot more comfortable with that it is I think precisely the combination of these big you know sort of just there such there's such amazing artifacts again you could you could spend you could so profitably spend all your time so much time thinking about dreaming about writing science fiction stories about language models even just as they exist today in our real world without ever actually like they don't have to do anything just as artifacts like what they are and what's woven into them and bound up into them is just it's so weird and so interesting but but that very thing I again like I don't actually know and nobody knows what's in Claude or gp4 and so that's I'm just like I don't know I I wouldn't put my I wouldn't put my name on the thing that that that you know shoots that output at unsuspecting people you want like uh nutrition facts like ethically raised organic AI I mean I think that's coming I think something like some verion of that is coming I think that's great I think it's really cool I mean you kind of see this already with like you know Adobe has the like we don't train on copyrighted material stuff for their image models like it's that's the first kind of like Salvo in that yeah War um I want to get into the book a little bit um and I know like I know for you like the idea that books are are not summarize is like is sort of like this key thing so I don't want to like I don't want to reduce it too much I mean no they're not subar they're not summarized but but one must always try I mean that's the Paradox Yeah well yeah I think it's like a uh the full thing is not summarized but the summaries are like a pointer to things that you to unsummon yes yes so one of the ways I could summarize uh this book um or one of the what it seems like a key idea that's like woven through it is um it's sort of like about stories um and about how stories shape our reality um but also maybe how like we can make decisions that author our own stories we don't have to necessarily follow the like you know uh prescriptions of the story that we're already a part of um which all which seems like really interesting and also like sort of um related to next token prediction and I don't know if I'm like reading into it too much but like I'm just really curious like tell me more about that that's great actually um you know of course so F first of all I should say book is newly released and so I haven't had that many conversations about it so this is this is really fun actually to just engage with someone else who's actually read the book who wasn't working on it with me and to like hear the ideas kind of bouncing around other Minds I will say that that particular Association the idea of sort of um you know um proceeding in a archetypal or maybe even stereotypical story form and you know producing the next token in a sequence was not consciously in my mind but um I do think it totally checks out totally checks out um I mean that's what we're talking about about liter literally these sort of patterns and templates that get used and reused in so many different ways that the language models are so good at um and you know I guess I I want to try to find a way to say this without spoiling it but in the course of this book as we get to know some of these AI entities a little bit better it turns out those patterns and templates are of incredible extreme importance and and and in fact and I think the this is this is true in in our real world and and in our real language models they are they are kind of woven into the systems at a really foundational level only because they are repeated so often and so powerfully throughout you know their their training corpora and are you saying something about stories in relation to language models or do you feel like maybe not you're not necessarily saying it in the book but like do you feel something about stories in relation to language models and um yeah yeah I I do and you know I think this is I want to not um make too bold you know I don't want to make claims that are too bold because the truth is I don't know I don't know what's going on inside these Labs these days I I understand the world of the kind of open research world of of the late 2010s a lot better um and and I think this probably is changing now it years ago you know when you thought about what was available in the world to to scrape and collect and and feed into your hungry language model it was things with narrative it was stuff with sto and that doesn't have to mean like fairy tales and you know epic quests by the way and myths and legends it can just mean news articles you know news articles have a real appetite for cause and effect even if something is um fundamentally inexplicable or random or you know tragic and chaotic a news treatment wants to say it's just because it's a human thing they want to say well this probably happened because of that you know or or maybe it happened because of this uh I I do think it's interesting and I don't have a answer to this I don't have a really or even a theory but I think it's interesting to to speculate about how these models and their their kind of constitutions their their um intuitions about about language have changed with the introduction of so much code you know I've heard it speculated that it was the addition of huge amounts of code to these to these training sets around the time of sort of a gpt3 that actually affected their language use in a lot of ways because code of course is like so structured and so so if and consequential um and maybe maybe the thing that we that we imagine is is reasoning you know in language models today comes in large part from what they've learned from code if that's the case I think suddenly you're like is code a story I don't know maybe it's a very linear story um so so my sense of what is in the hearts of the language models might be a little different today in the in the fullness of their of what they're learning from in 2024 but I can say with confidence um that in that era of like Project Gutenberg and uh and you know all the news stories on the web yes there is a bias there's a bias in that kind of language model towards if then cause and effect and the rhythm of a story and that's pretty interesting to notice um because you know there are stories in the world but there's a lot in the world that's not a story either and so a system with that bias you know you I think you have to be really careful totally and I think you're 100% right um uh I I've heard that too that um even if you're training a model that you don't need you don't want to code or you don't need it to code you train it on code because it just makes it smarter which I think is like and isn't that just so interesting that and it that that had this is a slightly further field but that has made me reflect on the fact that code you know we think of it as a machine language of course code is not for the benefit of machines machines have their own they and a machine has no use for python or Ruby that's not the language they speak you know everybody knows this they they get compiled into the real language of machines really code is for human benefit it's a bridge it's a it's a way for us to sort of think in a more machine way and we and we express that in these linguistic terms I mean again you could just you could think about this stuff so profitably and so with such with such just Delight for forever it's great totally there's there's tons of dep there it makes me very very excited um one of the other things that that I saw in the book this is like this happens early on but it also becomes it becomes more relevant later is like and you say it outright is like the central question of the human race is what happens next um and I when I read that I was just like because uh we've been thinking about rebranding every or re repositioning it and we we were like what what's the like how how do how should we re reposition it and we we came up with what comes next as like we answer that question and so when I saw that I immediately texted our editor-in chief Kate and I was like oh my God look at this look at this so I feel like we're kind on the same page there um but like tell tell me what that means to you and where you where you came to that from yeah I mean that's that's really close to my heart um in fact it's maybe the most close to my heart um and in fact in the of course in the story as you know having read the book The the narrator who is not human the narrator is sort of a kind of a hybrid organic technological creation um who exists to Chronicle human Endeavors and um and that and the chronicler says in the page that the question you know carved into my heart the irresistible question is what happens next um in that case this character is speaking for me that is that is the great question for me I don't actually know that it is um intrinsically a human Universal I think there are different humans and different human cultures that kind of care about it differently I think there's probably some people that are very uh they they're kind of they either don't care or they'd rather not find out and that's fine and that's fine I don't I don't personally attach a huge like uh deep value judgment to it but I do cop to my own hunger for the question for me it's constitutional um and that's one of the reasons I I'm such a huge reader reader of Science Fiction um you know science fiction doesn't tell you the answer but it it it suggests possible answers and I that's why I just find them so delicious to read um yeah you know and this is I'm sorry to to make it a little heavy but you know when I personally think about death you know at the end of a life um you know of course I think about things like suffering and disappointment and sickness and and all that kind of stuff um and I think about leaving behind things that I like and people that I love I also think about just not learning what happens next you know if if death if if through some you know if we lived in the some sort of magical world where you know an entity an Angelic entity visited you in the last moments of your life and just just spoiled it just told you like okay well you did you did a good job you lived a good life here's How It Ends here's how it all goes down it would for for me for me that would it would soften the blow it really would that's really interesting I mean I think it's it's definitely Central like I do agree that certain um maybe certain cultures or individuals like don't have that as their as they're sort of the center of their lives but like the fact that like Buddhism exists and the entire whole the whole thing is like just like being the present and like that's so hard for people like sort of says to me that there's something like pretty pretty hardwired about our brains it's like what's gonna happen now What's happen next yeah yeah yeah yeah right and and of course it I mean it's interesting to think about how that plays out on different time scales because of course to I've I've completely um gone to town in the last couple years on um another another fine podcast um one called big biology all about really great super super interesting Rich um kind of Next Level Next Generation biology and um and one of the things they keep coming back to there is that even on the level of a cell a dumb little cell a little a little microbe there is kind of that question of what happens next and and life in a way thrives as a as a little to the degree that it has a a pretty good simulation of what might happen next because that allows it to do things like plan and and you know react to possible dangers and and all the other things you can imagine life doing so that's kind of like the micro scale what happens next but then there's also you know there's the Dan scale what happens next of like 10 years and then there's the Sci-Fi scale of 10,000 years and and they're all interesting super interesting and not to make everything about AI but like what what strikes me is like that answering that question is the thing forces these models to bootstrap all the knowledge that they have is to be like okay what comes what comes next like answering that over and over and over again trillions of times which that's really interesting like I don't know if I would have thought that that's that would be the way that they would they would get smart you know yeah that's you know that's honestly that is a great that is a great observation um the the sense that a question you know in the same way that like evolutionary the Dynamics of kind of natural selection and evolution are so simple and yet they make the rhinoceros and the anchovie and the Joshua tree and Robin and Dan you know it's you're like okay I guess I guess it doesn't take much if you give it enough time and and and likewise um I don't I don't actually think that that question is is exactly evolutionary I think it's something a little different um but uh but I I actually really I think it's quite elegant to imagine that a a challenge so simple could in its fullness require a really a lot of complexity and in a in a rich picture of the world I I mean I I always take pains to say still incred inedibly incomplete picture of the world I mean it's kind of I sometimes think like oh man these language models they just think it's all text out here you poor you poor bastards you know there's more than text and and you know slow slowly but surely and I I think the multimodal models are obviously it's interesting because they have such Rich capabilities I think they're interesting because it seems I don't know it seems like a better way to live seems like a better way to be in the world to be able to combine different in and maybe find some common associations between them uh the idea of just living in a world of nothing but tokens is I don't know seems kind of like hell to me are you starting to think of them as like beings that have that can live in hell like what's your you know no not not yet um but again I you know I I I hope I don't I don't I should maybe try to actually do some research on this um if there are not presently like at minimum dozens of philosophers and cognitive scientists and ethicist kind of thinking about this stuff around the world and maybe teaching introductory you know seminars about it and getting people together to start to to have these conversations and ask these questions then I don't know the the academy is derelict in its in its Duty um because because these are really rich interesting questions for my part no I don't think these are beings yet um but you know if you ask me what my definition of a being is or where I draw the dotted line around beings I don't know so so who knows I mean it it I think that the academy or philosophy question is really interesting because I studied philosophy in college and um philosophy is sort of like obsessed with definitions um so like what is a being or whatever and like trying to trying to answer those questions and definitions obviously are useful for certain things but it's always struck me as kind of interesting that like we've been debating definitions of things like knowledge for 2,000 years and like we know a lot of stuff we just don't have a definition for it and and you would have thought that like to build intelligent machines that had knowledge we would have to Define it before we built them and really we just have to ask that one question over and over like what comes next and it just like bootstraps it which I think is like a kind of interesting comment on like the project of philosophy that like maybe these definitions are so high dimensional that like you can't get it in a philosophy book absolutely absolutely but I think that's great I think that's totally great or or in a sense you know again you you can kind of go around in circles and you could you could have a whole successful celebrative philosophy career just arguing with another philosopher over a over a you know definition using using language the entire time right you know you're writing papers back and forth and you're you know you're you're denouncing each other in you know in seminars and maybe it turned out that the thing all along was actually in all that language the idea that you're going to compress it down to a word or a sentence is wrong it's actually in that sort of that that vast that bulk of living language and I say that because I I don't know really how to you know assess this claim at all but I personally almost aesthetically I'm quite taken with the claim that what language models are is language itself like literally the the technology of language given its first dose of autonomy so it's not just something that we deploy as we as we want or need but suddenly like it's it's like it's like you rip language out of our heads in our societ kind of set it up and like turn a crank on the side and it starts kind of walking around slowly and weirdly like one of those little windup toys but and that's and that's you know again we people it there's you can argue about what's going on in language models there is no disputing that when you talk with one it sure seems like it's reasoning or thinking or deducing or or just or just you know politely answering your question and so this argument goes ah yes that is just that is the deduction and the reasoning and and the politeness that was always inherent there in language and we just never had this particular way of seeing it before again I don't know but I like it I like the way it sounds I think that's really cool that's a beautiful it's a beautiful metaphor is it is it the politeness that's always was always in language in the sense that like politeness is like sort of built into our grammar and into our like um syntax and all that kind of stuff or is it the politeness that's in humans that was like recorded into language I think both I think I think it's I think it's absolutely both for I mean for sure it's it's the latter no question but then but then even the former I think it's just you know um you know again language is not all of human language is not um some you it's it's not tcpip packets it's people and they're writing letters or trying to entertain each other with books or you know trying to trying to convince other people of things very often and and you know convincing is it takes some takes some tricks and takes some uh take some patience or whatever so yeah I I think it's in just in terms of what language has always been for um obviously you can be very rude and you can be very impolite but um but that that in itself that's only that's only effective ever because language is generally so language understands that it's it's you being used between people and uh and you want to kind of get along yeah I guess like on the subject of like language becoming an autonomous thing one of one of the things that is sort of present throughout the book is like you're playing with this idea of like what an i is um an i as in yeah quote like the letter I like me I um you know you have the chronicler who's like I think you said it's a fungus right yeah yeah a fungus unto unto which much technology has been layered yeah exactly at Great expense yeah yeah yeah um you've got like Clovis is another character that um uh I guess is sort of like a uh there are many instances of Clovis like all around the world and they all are are sort of connected um at least usually and all have the same and so rather than the and and rather than a sort of Borg hive mind or you know kind of like ah yes the ants are all part of the same Colony it actually is Clovis this this wandering robot in all their instances is the same person and the same personality which maybe that sounds like a subtle distinction distinction but I think it's actually a pretty important one so you're you're you're playing around with with with all that all the time like what is it about that that is Cap Capt capturing you well you know I you spend any time with writing and I would say especially um writing and then also reading um across languages and I don't mean natively just just in Translation and and uh and reading about the translations and the process and you learn that even in the you know you don't even need science fiction for this just within the scope of human language as it exists here and now today um there are so many different ways of thinking about I and and and the subject the subject of anything um one very classic example is that in Japanese the Japanese language there's like a handful there's not there's not just I as there's in English there's a handful of different eyes all of them with really interesting and and usefully useful narratively useful meanings um and so Japanese translators like people who who bring books from Japanese into English they all complain about this because they're like well it's impossible there's no there's no equivalent like there's you just can't we don't have other eyes we can slot in and so fundamentally this shade of meaning which is again really important can be really really useful and give something a it's like a little spin on the ball is essentially lost and they they try to find other ways to thread it in and kind of create that feeling um and I love that and I just think it's like yet even now you know you think about humans and how how we live in the world I think already the eye the singular eye is is quite complicated I think it has been complicated in the last 10 or 20 years the era of the internet because you know well here I am I standing in a room and at a certain time in human history that meant I was just in this room and I could only talk to people in this room but here I am talking to you seeing your face you know I am I am somewhere else um in a very meaningful way and we just still say I and I just feel like it's it's more complicated than that that now our our presences and our attention and our sense of of kind of where and how we can act in the world is already sort of spreading out and um it's fun to think about how you could reflect that capture that and play with that in uh in language I think of that I think of that too in so many ways like there are so many like I don't know internal family systems is like you have got multiple like eyes within you like that's a sort of psychological model or just like in in meditation like if you if you're if you get deeper into it you start to notice like that you are a bit of a different self depending on the context you're in um and like I think it's also why like when you go through a breakup or someone dies or whatever like that's part of why you're sad cuz like like the eye that is usually like around that person the eye that that that becomes around that person is like not going to get activated anymore you you're losing a part of yourself AB I you know I can't say that I quite thought of it that way but I think that's I think that's absolutely true to which I would add I mean this is much less um emotionally resonant but there's also the simple eye of your body systems you know you've got your gut and the microbes there which which we now know have desires of their own and are like blasting you know neurotransmitters up to be like do this do that um your organs I mean everything and these things all they have they it's not they're not ancillary they're not the accessories to like the pilot of your brain we just like neuroscience and biology have definitively established that it's a big interlocking committee and so yeah what is the eye is the eye Does it include my my gut I think so but yeah it's it's great it's just it's weird it's weird stuff to think about yeah and I think it it actually also relates a lot to language models in the sense that when you think about embedding spaces in what they are and how language models end up predicting what comes next basically in order to predict the next word in a sequence they take the the end word and then they figure out which version of that word is it so like if we take uh the uh the name Robin like they'll just go to everything previous to Robin in that sequence and be like which Robin are we talking about and that could be like there's there's hundreds of thousands or millions of different robins some of but a thousand or 2, of which are actually you just you in different senses and I love that idea I think that is so cool it's like we we created this dictionary of these like these very very very specific words um that we we all thought were were one word but like it's really like we're using that same word in like a million different ways um yeah it's really beautiful again and again just I mean it's like the assignment is so clear please linguists of the world and philosophers and everybody let's get let's get cracken because I mean I and I do think this is so interesting um the folks making and deploying these models I think a lot of them are interested in these questions they are not however the most urgent questions their most urgent questions have to do with kind of the infrastructure and scaling these things up and turning them into businesses and of course I think the very kind of direct safety questions about are people going to use this to to scam other people or to you know bring down governments or whatever and and there's just so much to dig into um and and I I really I do hope yeah you've kind of brought this up in a in a few different ways a sense in which just seeing these little mechanisms operate it it may how do I want to say this um it raises questions about long-standing projects you kind of go well ah maybe I didn't need that after all or maybe we didn't need to do it that way and and th those should be the kind of questions in real that kind of redirect uh you know streams of inquiry I think totally totally um so you've been working on this book for a while it's coming out yep how does it feel it feels good I mean it's as always uh a little dizzying to understand that it will soon be in other people's brains you know I said earlier I I've done a couple of these um I've had I've had a couple conversations ahead of time like this and this sounds so silly but truly in all cases I've have been sort of unprepared for other people to have actually thought about the book I'm like oh right ah you read it and it's in your head oh cool okay this is weird um but you know that's I mean that's the goal that's that is so the the dream and the the magic um and I think you know I'm of course I'm such a book nerd and such a such a I don't know book uh chauvinist maybe um that I just think I think they I think they do that more effectively than any other medium um I think they get they literally get into people's heads because they have to you know you you didn't you didn't just watch moonbound on a screen you enacted and kind of rehydrated the events and and the meaning in your in in your own language model inside your own head and um and that is something really special so I'm just excited for it to for that to happen a lot more in the coming weeks and months and and hopefully years I love it put me on team book chauvinism um the only kind of chauvinism I agree with yeah you know I will I will say Dr very gerain to our subject here um you know I I keep talking about how these how these Scholars ought to ought to be asking more questions about their work and their and their projects because of what we've learned I apply that also to writers fiction writers and I you know I have been highly motivated to think about what we're doing and and all this kind of stuff I also have I I'm sort of nursing a pet Theory I don't know that really has any evidence so it's just gonna it's gonna continue to be Robin's crackpot Theory but I think that I think that anybody who's interested in language model should really pay attention to how they dream and what dreams are and how dreams feel because I think the mech this is just subjective I I feel literally that the mechanism of dreaming is very similar to the mechanism of of a language model kind of saying okay well that's weird but I'm Gonna Keep it going I'm gonna I'm gonna do the best I can to to complete this this sequence um and the reason I bring it up is that I think that's also very similar to the mechanism of a novel I my pitch for novels fundamentally is that they are packaged dreams usually you don't get to choose your dreams you know they just they are we weird or scary or you know surreal or or boring or whatever but you just get the dreams you get I think this is one case where you you could get to sort of load a waking dream into your head and so in that in that way I think there's an interest some not totally understood connection kind of trilateral connection between books and dreams and uh language models I love that um do you think that that implies that the true AGI is going to have an language model as its dream generator and we just haven't built the AGI yet yeah maybe maybe so it is you know I I I I don't want to say too much because there's we're very close to spoiler territory here but but I will just say suffice it to say that Anyone who reads the the book moonbound will understand by the end how how much dreams and and sleep and what happens during sleep is important to me I I actually do wonder you there's I don't know if you know this but there's no living things that don't sleep I mean literally it is a fun yeah it's a fundamental part of every biological process do bacteria sleep they do they have they have a a phase that is sort of it's it's clear it's a clear sleep thing now are they dreaming well maybe but but it's different different dreams um C certainly you go one notch up the sort of the scale of recognizable to us complexity and and you do see things that begin to look like dreams and um so it's there's it that's I think really fun interesting stuff to to read about and learn about and it's clear that the experience that you know long-term flourishing through evolution of life on Earth has discovered this phase is really important it's like it's it's not negotiable and it makes me wonder it makes me wonder if there is some if we will determine that there's some analogical thing that you know that AI really ought to do um either for their own long-term success or health or who knows what that's fascinating well this was an incredible conversation very sci-fi this this is great I'm I'm delighted I'm delighted to be able to you know it's like the as if as if the novel itself wasn't sci-fi enough I feel like we just took it to a few a few further levels which is very fun we totally did which I I suspected was going to happen and I was very excited for the book is moonbound um it is out when is it out June 11th June 11th um so it'll be out by the time that this podcast is live um and thank you so much Robin for doing this this is really an incredible conversation damn this is great just a just a real treat so let's let's do it again before too long sounds good oh my gosh folks you absolutely positively have to smash that like button and subscribe to Ai and I why because this show is the epitome of Awesomeness it's like finding a treasure chest in your backyard but instead of gold it's filled with pure unadulterated knowledge bombs about chat GPT every episode is a roller coaster of emotions insights and laughter that will leave you on the edge of your seat craving for more it's not just a show it's a journey into the future with Dan shipper as the captain of the spaceship so do yourself a favor hit like smash subscribe and strap in for the ride of your life and now without any further Ado let me just say Dan I'm absolutely hopelessly in love with you