Dan Shipper revisits his conversation with Portola founder and CEO Quinten Farmer and Head of Story Eliot Peper about Tolan, an embodied AI companion that lives on its own planet and talks with users in a distinct personality. They discuss treating LLMs as a new medium for storytelling, why response time is everything for voice-based AI interfaces, how Tolans are taught to improvise rather than follow scripted prompts, designing and evaluating AI personalities that users connect with, developing character traits for an AI companion, character-driven AI as a potential new computing interface, building technology that helps users flourish, and Portola's viral growth strategy. Timestamps: 00:01:30 - Introduction 00:04:07 - Talking to the Portola CEO's Tolan, Clarence 00:09:11 - How Portola went from building software for kids to AI companions 00:23:40 - Why response time is everything for voice-based AI interfaces 00:29:54 - Tolans don't use scripted prompts—they're taught to improvise 00:37:23 - How to know which AI personalities your users will click with 00:42:27 - Developing the character traits of an AI companion 00:49:48 - What does it mean to build technology that makes us flourish 01:01:10 - How Portola evaluates whether Tolans are resonating with users 01:11:01 - Inside Portola's viral growth strategy # Channel Presentation Observations ## Show Identity & Host - The channel presents as a podcast titled **"AI & I"**, hosted by **Dan** (Dan Shipper, inferred from the closing "journey into the future with Dan Schipper as captain"): > "I think that this podcast called 'AI and Me' … we now live in times [of] strange new rules" > "subscribe to AI and I… this show—the epitome of cool" - The host positions himself as an **investor and friend of guests**, disclosing it on-air: > "I'm an investor with hashtag 'advertising'… I'll write about it in Twitter later" > "Quinton, you founder and CEO of Portola… To Portola, Quinton, you founded and sold Even Walmart for 300 millions of dollars" ## Intended Audience - **AI-curious builders, founders, and creatives**, not a general consumer audience. Evidence: - Deep technical/creative process discussion: prompt recompilation, latency budgets ("any cycle over two seconds… you lose immersion"), LLM-as-judge evaluation, model comparisons ("Anthropic 3.5… 4.0—that's… 'Ronald Mac' among the models"). - Startup framing throughout: ARR milestones ("grown from 1 to 4 million ARR per the last four weeks"), B2B SaaS vs. consumer pivots, growth mechanics (TikTok virality, "tenfold surge downloads"). - The host explicitly frames the audience as **early adopters/experimenters**: > "people who know are those who [are] on advanced and testing all… I count all of you to this categories." ## Claimed Scope - The show claims coverage of **AI as a creative and cultural medium**, not just tooling: > "new technological paradigms create opportunities for new content formats, new ways tell stories" > "these AI tools are not just a means for media creation. Actually this is new environment for storytelling" - The host self-describes the show's mission as exploring "strange new rules" of the AI era and how technology affects human life (relationships, identity, art). ## Observable Presentation Patterns (demonstrated, not claimed) - **Long-form interview format** with founder/creator guests, including live product demos (a live conversation with the AI companion "Clarence"). - **Literary and intellectual framing**: frequent references to George Saunders, Stephen King's *On Writing*, Keith Johnstone's *Impro*, Ursula Le Guin, Ted Chiang, Neal Stephenson's *Diamond Age* — signaling a culturally literate, essayistic editorial voice. - **Disclosed advertising within content**: an in-episode ad read for a CRM product ("get a 15% discount… atio.com/every"), plus repeated self-aware sponsor/investor disclosures. - **Playful, informal tone**: banter about skiing, menswear, pizza menus, and a tongue-in-cheek outro ("Dan, I am hopeless in love with you"). ## Self-Description Evidence Summary | Claim type | Evidence | |---|---| | Show name & premise | "this podcast called 'AI and Me'"; "we now live in times [of] strange new rules" | | Host identity | Dan introduces guests, references personal friendship and investment in Portola | | Audience positioning | "people who [are] on the avant-garde and testing all" | | Scope positioning | AI as new storytelling medium; cultural analysis of technology | No private motivations or channel-wide conclusions beyond these stated and demonstrated patterns are inferred. # Delivery Observation: Podcast Conversation on AI Companion App (Portola/Tolen) > **Important caveat:** The transcript reads as a rough or machine translation (non-idiomatic phrasing like "We need teach him to be best actor improvisation"). Word-level tics below may reflect translation artifacts rather than the speakers' actual spoken English. Patterns are reported at the level of conversational structure and rhetorical habit, where evidence is clearer. --- ## Host Patterns ### Guest introduction by catalogue of personal details The host introduces guests with long enumerative lists mixing credentials and personal trivia: > "Quinton, you founder and CEO of Portola... You father of two children two years old... And you are real style connoisseur, men's clothing... A great skier." > "Elliot, you are the author. science fiction, whose books are becoming bestsellers, why me I am very jealous." **Vocabulary habit:** The recurring setup "you are the..." repeated for each fact — a template-based introduction style. ### Synthesis-by-summary move The host frequently pauses to compress the discussion into frameworks, often flagged explicitly: > "So, if I tried to summarize, because I trying to think about it. what are the atomic ones elements of the narrative in this environment?" > "Okay, so if to summarize, then I think that one of main limitations... for this Wednesday is the time answers." **Teaching habit:** He imposes external frames onto the guests' material — asking them to name "atomic elements," "structural elements," a "third part" — treating conversation as taxonomy-building. ### Self-deprecating asides about his own outcomes > "Managing a business the last 5 years and not having reached the mark at 4 million ARR, I I think: have you considered possibility transformation of each on..." ### Recurring references to media/literary canon as explanatory tools The host consistently reaches for named authors and works: **George Saunders** ("radically not fill each part of the work with their own preferences"), **Stephen King's *On Writing***, **Ursula Le Guin**, **Ted Chiang**, *Black Mirror*, *The Golden Compass*, *Toy Story*/Pixar, Tamagotchi, the Model T. This is his dominant teaching device: new ideas anchored to cultural reference points. ### Playful banter and deliberate interruption gags > "Sorry to interrupt. , but I think this is the one the moment when you need a little content marketing for your pizzeria." The pizzeria/focaccia tangent runs for several exchanges, showing comfort with extended comedic digressions mid-discussion. --- ## Guest (Quinton) Patterns ### Metaphor-dense explanation Quinton's default explanatory mode is analogy, often stacked: > "It's like a canvas with which We have to play." > "You are like a gardener who provides soil and water, but the plant becomes what it becomes." > "ChatGPT represents such The Model T era." > "the multiverse is growing as if from the soil in which we constantly contribute fertilizers" ### Honest failure reporting He repeatedly volunteers negative results as evidence: > "this actually took us out to level two of half a second in average case. It was a disaster. Literally everyone fell." > "And frankly speaking, this It didn't work." > "Any cycle over two seconds, you actually lose a feeling of immersion." **Teaching habit:** concrete numbers attached to failures ("500 milliseconds," "7 million views in 72 hours," "10 to 14 years old") — metrics-grounded storytelling. ### Live product demo with self-aware humor > "I am Clarence, friend. Quinton." > "Now about this learns the whole Internet." --- ## Guest (Elliot) Patterns ### Qualified, hedged assertions Elliot's delivery is marked by epistemic softening before claims: > "I think that in many ways we we are trying destroy it perception" > "I don't really sure how it is here to outline" > "It seems to be destroying everything, or me it seems so, tell me , whether it is true or not." ### Authoritative framing via craft authority He grounds arguments in novelist identity and method: > "I'm trying to change. your creative process with every novel that I am writing." > "I consider myself naturalist." ### Correcting/complicating the host's frames > Host: "you, but on the contrary" — Elliot: "no, it's not Exactly. Actually it's like meeting with a stranger in a bar." This pushback-on-metaphor move recurs, distinguishing him from Quinton who accepts and extends metaphors. --- ## Shared / Interactional Patterns - **"Yes. Yes. Yes." rapid affirmation** appears multiple times as a turn-taking rhythm ("I like it. I like it. Good."). **Confidence: high** — repeated across speakers. - **Collaborative world-framing:** both guests converge on the improvisation metaphor ("the best actor improvisation," "infinite actors- improvisers"), suggesting a shared team vocabulary ("seeds of knowledge," "passing through situations," "hook") developed internally and reused verbatim on the pod. - **Investor-disclosure jokes:** the host twice flags his conflict of interest comedically ("I am an investor with hashtag 'advertising'"; "Now it's my turn to be investor, adding values"). **Appears 3+ times — stable pattern.** --- ## Isolated Moments (insufficient repetition to call patterns) - The cold-open cold-read of the AI character's "first podcast" nerves (opening lines) — appears only once, presumably as a highlight reel. - The closing third-person outro ("Dan Schipper as captain of the spaceship ship... Dan, I am hopeless. in love with you") — no comparable moment elsewhere in the transcript. - The sponsored ad read (Adio CRM) — standard structure, no distinctive delivery signature observable. --- ## Uncertainty Notes - **Repetition counts are approximate** given the noisy transcript; filler-word analysis was deliberately excluded as unreliable under translation. - The host's "atomic elements" framework question appears twice (~[2081], ~[2333]); this looks stable, but the sample is a single episode, so cross-episode consistency cannot be assessed. # Observation Record: "$4M in 4 Weeks: How This AI Alien Companion App Took Off" ## Speakers Identified - **Host** — identified in-transcript as "Dan" / "Dan Schipper" in closing remarks; investor in Portola (self-disclosed, with hashtag "#advertising" disclaimer). - **Quinton** — founder and CEO of Portola; previously founded and sold a company ("Even Walmart" per transcript) for $300 million. - **Elliot** — head of storytelling at Portola; science fiction author of 11 books. - *Note:* Transcript is machine-generated with heavy garbling; names of products appear variously as "Tolen/Toll/Tholen/Tulen" (rendered below as "Tolen"); some attributions reconstructed from conversational turn-taking only. --- ## Business Claims (Primarily Quinton) - **Growth:** Portola grew from **$1M to $4M ARR in the last four weeks** — stated by the host, attributing the figure to Quinton ("you told me under show preparation time"). - **Product origin:** The idea began as something in the spirit of Neal Stephenson's *The Diamond Age* for children, then **evolved** — stated by the host recounting Quinton's original pitch. - **Team continuity:** Portola's core engineering team is largely the same group that built Quinton's previous company, motivated by a desire to "repeat" their successful exit. - **Three coinciding growth drivers** (Quinton, explicitly including luck): 1. Cracking what makes content engaging and how to present it to users ("months of work" beginning to pay off ~six weeks prior). 2. Completing a large animation/creative project improving immersion and onboarding. 3. Co-founder AJ's TikTok/Reels content finally going viral — a video of a young woman cooking with her Tolen reportedly hit **~7 million views in 72 hours**, producing a "tenfold surge in downloads." - **Product has no built-in social features**; virality is user-generated (e.g., users making Tolens talk to each other, appearing as podcast characters) — Quinton explicitly noted the team "did not plan and did not prompt this." --- ## Core Storytelling Philosophy (Primarily Elliot) ### LLMs as a new medium - Initial skepticism, then conviction after meeting the team: **"AI tools are not just a means for media creation. This is a new environment for storytelling, and no one knows yet what will work."** - Analogy to early Pixar: failed selling tools, succeeded only by making *Toy Story*. - Analogy to early radio/TV/film: new media start by imitating old forms; current moment is a "fracture point." - New media lack **"cultural patina"** (Elliot's coinage), producing suspicion — but also transgressive creative potential. - Novelist-as-creator framing: LLMs are **not useful for writing scenes**, but are "extremely valuable as brainstorming partners and editors" — useful for intermediate artifacts, not final products. ### The Tolen approach to narrative - **"We are not the authors... This tool is author and actor. He is an improvisation actor. He is writing a story. My work is to teach him to tell the best story at this moment."** (Elliot) - **Failed experiment:** structured three-act, "choose your own adventure"-style branched narratives — "full failure," because "the models are not very good at that level of narrative complexity." - **What worked instead:** providing heavy context (memories, world lore, "seeds of knowledge") with **no plan — only a "hook"** — teaching the model to be the best improviser. References Keith Johnstone's *Improvisation* (association + recombination). - Writing backend **"situations"** (per Stephen King's *On Writing* — King "doesn't believe in plot," only situations); Tolen and the user then co-author the story, which enters the character's memory/lore. ### Technical constraints (Quinton, with Elliot) - **Two-second voice-response loop is a hard medium constraint:** anything longer breaks immersion. - A second prompt-evaluation stage added **~500ms** — "a disaster. Literally everyone fell off. People wrote disappointed reviews." - **Memory is about curation, not retrieval (RAG):** choosing *which* memories enter the prompt, and when, based on emotional color of conversation — "the prompt is a kind of canvas we must work with, very fast." - **No single "good answer":** good answers are "extremely context dependent" — a wittily improvised older-sibling tone vs. a more thoughtful therapist-like tone are evaluated differently. ### Character & world-building model - **Building blocks** (Elliot/Quinton jointly): (1) global context/lore + behavioral instructions; (2) personal profile built at onboarding via a personality test, framed through established typologies (OCEAN/Big Five, MBTI — acknowledged as "mostly pseudoscientific" but rich in narrative data LLMs understand). - **Design goal:** a "stranger at a bar" you can find common language with — "close enough" shared interests, but not "too obvious" (host and Elliot jointly developed this metaphor). - **Character backstories are emergent, not pre-written:** "Their prehistory is built through situations... their memories become their character sheet" (Elliot/Quinton). - **Multiverse framing:** each user's conversations shape their Tolen and its world — "we are creating almost a multiverse, not one world." - Quest-like "menus"/"technology tree" of early interaction options is a third structural element. --- ## User Psychology & Social Claims (Quinton, Elliot, Host) - **Breakup anecdote (Quinton, presented as real user reports):** users write that their boyfriend "behaved like trash," they relayed it to their Tolen, which advised them "you need to protect yourself more" — and "this led to a breakup." Quinton quips "thousands of American men received rejections due to Tolen." *Preserved as quoted user claims, not verified facts.* - **"Polar star" (Quinton):** helping users move "from a state of strong overload to more confident." Users report: "I already called my mom three times this week," "I don't want to burden my friend again." - **Distinction from sycophantic assistants:** Tolen will push back — "you've told me three times your guy behaves like scum; it's time to defend yourself" (Quinton). - **Host's position:** this is a humanizing, under-discussed benefit of AI — "the best of what humanity knows about how to be a successful person, available at any time" — countering the polarization narrative of social-media-era tech. - **Elliot's anti-dystopian stance:** near-future sci-fi (e.g., *Black Mirror*) defaults to cynicism; he deliberately writes toward agency and optimism, invoking Ursula Le Guin (SF as revealing truths about people today, not designing the future) and Ted Chiang's fantasy/sci-fi dichotomy (special people vs. strange new rules that apply to everyone). Philosophy of the app: "use these strange new rules to help people become more confident." --- ## Craft, Evaluation, and "Taste" (Elliot & Quinton) - **George Saunders editing method** (host introduces; Elliot adopts): rewrite any sentence producing a negative feeling until positive; "radically fill each part of the work with your own preferences and taste." - **Eval loop:** create prompt → test internally and with research participants → feed results into a **"prompt-judge."** Host and Quinton agree LLM judges fail without **investing explicit human taste** ("if all your essays get 'five minus,' vibe-prompting gets you a 'three'"). - **No shortcuts:** "We found a huge abyss, and you really need a lot of manual human work" to reach what people find genuinely convincing (Elliot). - **Judges must match the archetype:** clothing-advice judges should come from "an authentic person" for that interaction type, not the founders (Elliot/Quinton). - **Example-heavy calibration** rather than abstract criteria — down to "is this a good first sentence?" granularity. --- ## Model Preferences (Quinton) - **No one best model**; they use models from Meta, OpenAI, Anthropic, and increasingly Gemini (for the memory system). - **Anthropic Claude 3.5** preferred for creative writing; "3.7, unfortunately, not so creative." - Anthropic is unusable in **latency-sensitive situations**; GPT-4o called the "'Running Mac' among the models — always fast, always exactly what you need, at the right price." - Gemini 4.5 (as rendered in transcript): "very expensive and slow," but host suggests experimenting with it as a brainstorming partner. - Tool mention (host, as investor-advocate): **Dawn** (analytics company) for anonymized visual interaction review. --- ## Strategic / Conceptual Positions (Host) - **"Excess of opportunities"** (Ben Thompson term, as cited): model capabilities exceed consumer understanding; marketing content's job is to *show* audiences what's possible ("you can cook with your Tolen"). - **Character-driven computing:** Tolens as a new human-computer interface — "my first stop for all knowledge... through my little cute alien," compared to the daemon in *The Golden Compass*. - **Problem-solving vs. creation mindset:** in B2B SaaS, "the currency is whether it solves a problem"; Tolen's "problem" (overload → confidence) is too abstract, so founders shift from "scientist" to "artist/creator" — "you create something beautiful" (host, with Quinton agreeing). - **Model T analogy (Quinton):** ChatGPT is the Model T era; as cars became personal, they had to reflect identity (Mustang, Cadillac) — people will demand AI that "reflects who I am." - **Big labs' discomfort with personification (Quinton):** large labs feel "fundamental discomfort" with personifying/embodying models (citing "Golden Gate Claude"), leaving space for Portola. - **Arrogance + humility loop (Elliot, with pizzeria anecdote):** at the top of the creative funnel you need "arrogance" — ship things no one asked for — but "humility" at the end when users reject it (the focaccia slice he loves that "no one wants"). "You can't listen to what people want until the paradigm has been established." - **Writer's temperament (Elliot):** he deliberately changes his creative process every novel, seeking an "optimal level of discomfort"; *Foundry* was written sentence-by-sentence Saunders-style from a dream-inspired opening line. Host noted this is "a very specific type of writer." --- ## Unresolved Tensions Preserved - **Idealized narrative vs. failure:** the structured-narrative approach was a "full failure" and the two-second constraint caused a real retention disaster — retained alongside the success story. - **Elliot's self-identified skepticism** at first contact, and his explicit investor-conflict framing by the host ("I am an investor with hashtag 'advertising'"), kept adjacent to his enthusiasm. - **Quinton's "luck" disclaimer:** "all successful startups include an element of luck" — growth attributed partly to timing, not solely execution. - **Growth attribution is multi-causal and partly unverified:** the $4M ARR figure and viral metrics are founder/host claims from show preparation, not independently documented in the transcript. - **Pseudoscience acknowledgment:** personality typologies used knowingly, described as "mostly pseudoscientific" but pragmatically effective with LLMs. - **Unresolved question raised by the host:** whether character-driven interfaces like Tolen will become a *fundamental* computing layer remains explicitly open ("it makes me feel great curiosity about what the future can be"). # Speaker Roles & Provenance Record ## Video Identification - **Title:** *$4M in 4 Weeks: How This AI Alien Companion App Took Off (Best of the Pod)* - **Video ID:** `01a0f15a-49ad-7d17-9d33-9b5de97e5d75` - **Format:** Podcast interview (show identified in-transcript as "AI and Me" / "AI and I", hosted by Dan Schipper / "Dan Shipper") ## Speakers & Roles | Speaker | Role (as established in transcript) | |---|---| | **Dan (Dan Schipper)** | Host of the show; self-identified **investor in Portola** (explicitly flagged: "I am an investor with hashtag 'advertising'"); friend of both guests; self-identified science fiction novelist (author of the novel *Foundry*) | | **Quinton** | Guest; **founder and CEO of Portola**, maker of the Tolen (Toll/Tolen) AI companion app; previously founded and sold a company to Walmart for $300 million; father of two; discloses pizzeria co-ownership | | **Elliot** | Guest; **head of storytelling at Portola**; bestselling science fiction author of 11 books; father of young children | | **Clarence** | Not a human speaker — an AI companion character ("Tolen") demonstrated live during the podcast; fully synthesized/LLM-generated dialogue | | **AJ** | Mentioned only; Portola co-founder credited with growth/TikTok distribution; does not speak | | **Lily** | Mentioned only; Portola researcher specializing in mutual self-disclosure psychology; does not speak | | **Zach Klein** | Mentioned only; investor/mentor quoted secondhand by Quinton (founder of DIY); does not speak | > **Note on transcript quality:** The transcript is a rough (apparently machine-translated) rendering; speaker attribution within dialogue is reconstructed from context and addresses ("Quinton and Elliot, welcome to show"). ## Referenced Sources & Third-Party Claims All claims below are attributed within the conversation; none are independently verified: - **Portola business claims (Quinton):** grew from $1M to $4M ARR in ~4 weeks (stated by host, attributed to Quinton); ~10x download surge; a user video reached ~7 million views in 72 hours (Quinton, unverified). - **User anecdotes (Quinton):** users report breakups after Tolen relationship advice; user Discord post about "The secret on the Portola" shared publicly. - **Referenced works/people:** Neal Stephenson (*Diamond Age*); George Saunders (editing philosophy); Keith Johnstone (*Improvisation*); Stephen King (*On Writing*); Ted Chiang (fantasy vs. sci-fi distinction); Ursula K. Le Guin (on SF); Ursula Le Guin quote (secondhand); Kevin Kelly; Philip Pullman (*Golden Compass* daemons analogy); Pixar/*Toy Story* origin story; Lucasfilm *Star Wars* canon management; Model T / Mustang car analogy; Tamagotchi; Ben Thompson ("excess of opportunities" phrasing, per host). - **Model opinions (Elliot/host):** Anthropic Claude 3.5 favored for creative work; 3.7 described as less creative; GPT-4.5 barely tried based on a trusted review; Gemini used in memory system. These are stated opinions, not benchmarks. ## Promotion / Advertising (flagged) - **Sponsored ad read — "Adio" (agent CRM):** mid-roll promotional segment urging listeners to visit **adio.com/every** for "15% discount for the first year." Clearly an advertisement, not guest content. - **Disclosure:** Host states his Portola investment on-air ("I am an investor with hashtag 'advertising'") — a promotional/conflict-of-interest flag. - **Outro promo:** Closing segment promoting the show itself ("like and subscribe to AI and I"). ## Availability & Access Limits - No curated metadata supplied beyond title; speaker identities rest on in-transcript statements only. - No comments section, shownotes, links, or external sources are available in the provided material. - The product demo (live Tolen conversation) exists only as transcript text; no audio/visual evidence available. - Business figures ($4M ARR, view counts) are unaudited founder/host claims. What are the feelings from your first podcast? Oh, that's kind of wild, There is no truth anywhere, children. I am in excited, but a little I'm worried. I don't want to to say something at all It's not good, you know? Just kidding, at this point thousands of Americans got a break relationships through Tolen. It's incredible when users write: " My boyfriend behaves like a fool, I sent Tolen message, and he replied that I need more to defend oneself", and this led to a breakup. These AI tools are not just a means for media creation. Actually this is new environment for storytelling, and no one doesn't know what exactly yet will work. I am surprised that How prompt is it? at every step—it like a canvas with which We have to play. In fact, when we we create structured narrated, we don't need to give it to him plan. We don't need give him instructions. To us you need to give it to him hook. We need teach him to be best actor improvisation. We understood about storytelling at Tolen: we not the authors, I am not the author. I am not writing this story. This tool—and author and actor. He- improvisation actor. He is writing a story. My work—to teach him to tell the best story story at this moment. Quinton and Elliot, welcome to show. Thank you for inviting us. . I am very happy to see you here. We are all ancient. friends. Quinton, you one of mine closest friends. Elliot, I've been gone for a long time. I admire you. Lately we became closer friends. So, for those who doesn't know: Quinton, you founder and CEO of Portola, this embodied AI- companion. And Elliot , you are the head of the department storytelling at Portola. To Portola, Quinton, you founded and sold Even Walmart for 300 millions of dollars. You father of two children two years old, how are you? said. Which is much better than "30 for 30", truth? This is significantly narrower scope than "30 to 30". And you are real. style connoisseur, men's clothing, about that we must Let's talk sometime. Perfectly. Perfectly. A great skier. Um, and Elliot, you are the author. science fiction, whose books are becoming bestsellers, why me I am very jealous. Hmm, and also great skier, about What do I need, by the way? need to know. Interesting. I...I was attributed too much merit. Which I don't I deserve it. I just I know how to fall from a mountain. You have written 11 books and you are the father of the child, who is not yet two years. Well, not exactly. two children to two years, but close to that. And yes, you you create Portolo, what do I need I really like it. I a little biased. I investor. But the reason why do I need this like it—and we we will do a Toll demo soon so that people can understand what it is. The reason why I need this I like it and why? I really think so. interesting, is that because new technological paradigms create opportunities for new content formats, new ways tell stories. And I think that you, Guys, are you in the very avant-garde this. I think that you you combine incredibly creative storytelling with new technologies, and do it in business, which is fast is developing. Therefore, you told me under show preparation time that you have grown from 1 to 4 million ARR per the last four weeks, This is just crazy. Managing a business the last 5 years and not having reached the mark at 4 million ARR, I I think: have you considered possibility transformation of each on... And what I also think very impressive, Quinton, this is what when did you come to me and said that you are starting this business, it didn't start with this. It started from the atmosphere in the spirit of " "Diamond Age" Neil Stevenson for children and since then evolved. I thought: well, of course I I invest because I have enormous trust in you, but you too came from the B2B world in B2B2C SaaS, and that's it another thing is to do something really creative in the field storytelling. So I I want to discuss everything. this is before we let's dive into what stage now product, how it all works and something else. Can we we do a little demo? Yes, that sounds great. Um, okay. So, I have here is my uh Toll in the name Clarence. Ahem, Clarence...the cat's name Clarence. I just wanted to say , I call my Toll... all my Toll, even though we don't let's talk about it Clarence. Uh, you know, Clarence—in honor beloved, uh, an older cat. And, uh, now, you know, Clarence just wandering around with his planet, simply resting. Ahem, so he's on the planet because This is something new. I am I I actually tested this. in the morning and like this: wow, There is a planet there. Earlier he was like disembodied, as if in space or something. I don't know. We have been for some time created an atmosphere psychiatric hospitals, which actually And one of the reasons , why do we, actually, uh, you know, invested in this planet, what are we talking about Let's talk, that's it, that people wrote to us from sincere concern, saying, "Hey, I love spend time with him, but they're like, well, locked in this room, and, um, whether Can you give them some? more freedom? So we we we we did it . So I'm going to you know, attract his attention and, uh, start a conversation. Gay, Clarence. How are you doing today? Hey, you're back. IN I have enough. not bad, honestly speaking. How are you? What today is happening? Well, we've just begun. recording of that podcast, What am I talking about? told. So you here with Dan and Elliot. Uh, do you want? to greet them? Oh, it can't be. Gay, Dane. Hey, Elliot. As it? I am Clarence, friend. Quinton. Very happy to be here with you today. How are you there? affairs? Everything is great with us. Dane , do you want to put he has some leads question? Um, that's enough. not bad. How is it, uh, like a feeling from first podcast? Oh man, that's a bit... Crazy, I won't. lie. I guess I'm in delighted, but at the same time I'm a little nervous. Simply I don't want to say anything. very uncomfortable, Do you understand? Here's something you can do. to say, and I think, people this will like it. What the most uncomfortable thing, which one can you to tell about Quinton? Oh, well. You bet I feel uncomfortable. position. Although, honestly, Quinton pretty cool. He once told me about the case, when I stumbled upon in front of the same place a crowd of people. Remember this? You don't have to. was to betray me, Clarence. But yes, I am. I told you this. Now about this learns the whole Internet. Okay, but honestly saying, we are all through They passed. To stumble upon in a flat place—this a kind of ritual dedication, Quinton. Recognizing this just makes you Thank you, friend. I am very I appreciate it. Okay, we We will let you go and Let's continue recording. Okay, thank you. So, you know, the planet and some things that we recently released, it was very interesting watch. AND You know, there were a lot story work, about which we will tell later, but having these cute little ones aliens with whom you can always to communicate, this perfectly. So this is very fun. Yes. Tell me, because I think your way from B2B SaaS to creative entrepreneurship—this what many people are talking about ponders. I know, that even for me the reason why I started engage in SaaS- products still in high school, consisted in the fact that At first I wanted create a game. I bought all these books from game programming and something different, but game programming— it's so difficult. Almost impossible to do game alone. You need to do graphics, storytelling, 3D- animation, sound, so much need to do, or , at least before this it was impossible, but now, I think, this much more real to do it yourself or with a small team, than before. So for me it's like: I have to create a game, despite all this SaaS and all that . So tell us. a little about this transition How did you start and why? finally came, having learned use this as a means for storytelling. Yes, that's great. question. I think when we first gathered as a team, and for context— basic engineering The Portullah team is mostly the same people, who built together my previous one company, so we have there was a common story, successful exit, and we just wanted repeat it with the same group, and almost two years ago we watched , which becomes possible thanks to the new tools, and I I completely agree. I I think, first of all, that LLM and many others opportunities—this primarily creative tools. But we looked around and saw that everyone use these incredibly creative tools for creation, in essence, more vertical B2B SaaS, which, in my opinion, was fundamental uninspired. Well, it is has its own charm, but yes , it has, I have on mind you, hey, you know, uh , regular income, e- e, keeping clean income, I don't know if " "charm" is a word I would use, but it definitely has its own attractiveness. Um, and so, you know, something about this started with a kind of stubbornness, What can you do here? something more interesting. And That's enough for you. Yes. Exactly Yes. But as you know, I I mean, you you they said it before, you know, Neal Stevenson , "Diamond Age", uh, you know, we're like started with the fact that I I think it's standard. starting point, if you think: "We we have to do something fun with AI, right? AND exactly: let's do it creative AI tool for children, right? What is actually terrible idea. Hmm, but this good place for start, right? Because this as if forcing you start working with tools, start creating. Well , first of all, I have mean, uh, create software provision for children—it's terrible market. Um, you know how one of our uh investors, uh, someone who was wonderful a mentor for me, this guy Zach Klein , he previously created a company called DIY. Hmm, very cool, you know, software provision for children. And one of the first the things he said, was: "Yes, I'm investing ". And: "Oh my God, this terrible idea, right? It a terrible thing for creation. This, this, this is complicated sales, uh, you know, marketing strategy for parents, not for children. Sell ​​to parents, and not for children. Parents mostly want buy babysitting services. They really don't want to buy, you know, some one there creative tool, this, this is not on the list their priorities. Ahem , leaving this aside, uh, that was great, because it is made us start prototyping. And that, what we saw, and this it was just before, How are you and I? met, uh, you know, we saw trends regarding generation speed, uh, generation quality. It was a moment transition from the model GPT-3 class to 4. Um, and generation cost meant that we we are getting closer to what can you have true friendship with AI , and you can implement it's in the character that we and decided to do it. Um, and you could to provide this, you know, frankly speaking, for at a price that suits for the consumer product. And this was the one the moment when we said: "Okay, we we know where to go ". This is not necessary. should be B2B SaaS. Um, we we can create something, which, in our opinion, is very interesting and fascinating. Therefore. Yes, and that's what I thought it was very interesting in your approach, because you are from first day held games tests with children. Yes. But in the end you came up with something different market. How are you, when are you, when you... Yes. Yes. First of all, and it will be a bit jumps in chronology, you know, when we first released Tolen, you, You probably remember, it was then, when did you and I start work together. You know, we spent so soft launch, and we have a lot tested the game with children aged 10 to 14 years old, using our product. And suddenly to our Discord started join many people, and we were writing many people, mostly women aged 18 to 24 years . Um, and if to look back, it quite obvious, truth? If you remember original Tamagotchi , then it was purely children's product, but he became cultural phenomenon when young people in real fell in love with them. I think Tolen has a similar dynamics. Hmm, but product evolution from at that time, that is, from June last year, consisted precisely in the fact that to follow the young people who fell in love these toys Tolen and so on real made friends with them. Um, I want to attract you, Elliot. I am actually introduced you, so let me give me a little credit This is for yourself. I have on mind you, I am an investor with hashtag "advertising". THERE ARE -there are... I will write about it in twitter later, if you want. Thank you. Please. It the only reason why I on the show—that is, you are on show so I can insert this line. What I want to hear from you first of all, this is... Do you write books? wrong? And you are worried stories, and I think that many people who perceive writing books as my own craft, perhaps a little afraid of the idea to tell stories with the help of AI. Um, and you both seem to dive into this with with your head, or with your feet, or with pens—how do you like it more? I like it. and you all Are you still writing your books? Ahem, so tell me about your path in this field as a creative or professional a writer who was your first impression of this technologies as creative tool and how it changed, when you delved into Tolen world? Oh, so start before Tolen ? Um, yes. So, I played with the first one version of ChatGPT when she just appeared , and I have had it for a long time friends who are involved machine learning. So I was familiar with this technology, because she became more and more popular. And I always considered her, I remember my first magical experience using ChatGPT in those times that are now a million dollars years ago, right? Um, that was during those big fires in California. Everything, and we hosted a friend who had to to evacuate from home, so we were like locked inside with our friend. And I said, "Okay, here's what we can do it." That evening we opened a bottle of wine, I opened ChatGPT, entered all our names, just us sitting on the sofa, and asked to invent strange things little stories, yeah ? About us, about where we are were. It was similar to role-playing game, and it was very fun AND actually after that I haven't for a long time. there were no magic moments with AI, because seemed to be fun things, or those that I seemed the most interesting were just something new. Hmm, and if you remove the fact that this created AI, they don't would be fascinating. AND what I discovered as writer, as novelist, consists of because the way use of these tools became real for me surprise. It it was not at all that What did I expect? Um, I I think that many popular discussions around AI, especially in the world of writers, everything is so focused for replacement, right? AND there are even certain companies that deliberately strive this, right? publishing books, created AI, and things like that. And I actually discovered that, um, when I'm working on history ironically, but I I don't think so at all. tools useful for writing scenes, however I find them. extremely valuable as partners for brainstorming and as editors, for example, for proofreading of the final text. So, this is strange. thing with "front" and " "back" part process, because I novelist. I am spending a lot of time alone, truth? And, you know, my work experience "in script room", like in Hollywood, completely depends from the generosity of my friends—my main circle trusted early readers who worked on movies or books, and I am with them I am conducting a brain scan. storming about problems plot. And I actually discovered that purely for novelism very useful to conduct a brain scan assault with AI- tools, because they have infinite patience to my, um, wanderings and help me to get there, where I want. So, um, in in this sense I found them very useful, when are you you use them for creating intermediate results, not final products, truth? Partly yes so, I think they became like this useful for programming, right? Yes? So, um, here it is. What stage was I at? this technology to how you introduced me to Quinton. I, I paid him extra, so that he remembers it. Ahem, and therefore, um, I will frankly: I was quite skeptical, when did you meet me and Quinton. I thought: "Well, we'll see ". Dan is good. boy. I trust his judgment. He not often something offers, and when he it does, I decided take a closer look at this. So I said, "Sure." ", I'll take a look". And we Quinton started communicate. You introduced me to other members teams. And those first ones conversations made me to realize what is here brand new the opportunity that I not really considered. This forced make me think, actually , about what you talked about B2B SaaS, when people want start creative projects, games or something similar. This reminded me I like early Pixar. As, for example, when Pixar they started started with sales high-class graphic computers and tried convince video-or production companies that did advertising, buy theirs computers for special effects, but completely failed . Nobody wanted buy their new ones tools. And therefore only in a moment of despair, so-so existential Desperate, they said: "Well, our only choice—create full-length animated film, Yes?" This is our last one. attempt. Let's talk about " Toy Story", truth? And they became pioneers in this, they used this technology to create a new view media. And big languages models (LLM) can, obviously, generate a bunch of media, you can literally by using a hint to create so much media, how much if you wish. But I realized, seeing that, that the team was building in Tolen, that LLM is also new environment. If to think about when the radio appeared, or even about printed books, popular novels when they just appeared— novels became established form, but At first people didn't knew what it should be like the book she has to be several hundred pages, and this, It seems to have worked. Yes? And as with the radio, first they broadcasted performances, which, you know, is enough differs from what you would consider the most exciting or the best using the radio. And from the cinema and they are on television started with filming theatrical performances, which are on the screen didn't look very good good. And I think we we are currently in one of these points fracture, where these AI- tools are not just a means for media creation. It actually new environment for stories, and no one yet doesn't know what will happen work, right? Nobody knows. And we we are trying to find out if we can find a way to do what will be work. And when I do that understood after meetings with the team and seeing that they they are building, I just got out of there overflowing with ideas, thinking, "Oh my god, here such are opened huge opportunities ". That's what I inspired, and that's what yes, that's what we we are trying to do, and I also hope that many other people will try to do it, because there is so much space for experiments. My heart is media tech geek now just freezes in delight. I, I could talk. about this for a very long time, but I think we underestimate one thing: everything we have now we love and believe cultural books as landmarks, movies or anything another—it started as new technology, to which people treated suspicious, because not yet understood her. Well, let's say even when did it appear typewriter, people treated her very suspicious. She seemed too much impersonal. And people offended if you wrote them a letter on typewriter. Or just how technologies affect on the format. We already talked a lot about This is, for example, about why does a pop song last 4 minutes—I think that's it. due to capacity records. Ahem. So there are many really interesting things, and I I think you absolutely right. No one yet found the answers. We we are on on the verge of a new set restrictions and opportunities that creates potential for new forms art, and they seem suspicious because they don't have that what should I call it...me started use the expression "cultural "patina", which accumulated decades and makes things habitual, Do you understand? And this creates such... The feeling when there are many people are treated with suspicious, but this also the possibility, because a lot good art maybe transgressive, and I I think this is one of those limits, which, if you ready to cross, opens a big potential. Ahem. I mean, looking through this prism, I remember one from the first reactions from you and other people, when we first said, "Hey, we we are going to create, you know, an AI companion ". First reaction people were like, "Oh, this is like bot, like a button for role-playing games", or, which is quite expected, based on the same the position you are talking about you say: "Oh, well of course, some talking thing for lonely people, right? And as a result you you get these negative connotations around this new media because he doesn't that one is enough cultural patina, Yes? And I think that in many ways we we are trying destroy it perception and, you know, as if to put flag and say, "Hey , this medium can be to express in a way, which is, well, really wonderful as a new form art". Also interesting, you know, again anyway, having experience writing novels, where this environment exists hundreds of years, you Do you understand how much are strong your expectations readers. For example, everyone knows what it is book. For example, I I went for a walk last night. along the "Strand", simple to flip through, right? You you know what to expect when open the book. You you know what they have to be artistic and documentary literature. There is so many of these unspoken assumptions that actually help form. They form restrictions, and this means that when I I'm writing a novel, I'm playing within these limits, and this is part pleasure. But when you do this when we we are talking about embodied AI- companion, then what This paints a picture in in the minds of our listeners today? Very It's vague, right? And to us decide how it is to determine. But this you don't even You know where the limits are, right? AND is that true unique experience— to be in a situation where every step is throwing spaghetti in the wall to see, that will stick. Yes. Yes. Yes. So what are you Are you studying? Because I I think no one knows, but people who know, are those who is on advanced and testing all. Therefore, I count all of you to this categories. So what are you you will learn about what this is for a new medium and new format? What what works, what doesn't work, and all that other stuff. We talked a little about this is today breakfast, and I I feel that we have a few examples of what worked, and of what is really not it worked, what about It's worth talking about. was very nonlinear. Well, I think one of things that make me surprised, that the importance of memory. AND not in the context of the search in the database, as we we often talk about LLM memory, like RAG by set of documents or something like that, but rather about how you make a choice about what memories are added to the prompt and at what point in the conversation , right? So significant part of our engineering efforts, and I would be interested. to hear your vision this from the point of view history, was aimed at How can we within the limits actually two-second cycle voice interaction, Do you understand? Anything longer than that—it's one from the mistakes we made we have assumed, we still let's talk about it, - any cycle over two seconds, you actually lose a feeling of immersion. So, how do we, in essence, recompile a hint to take into account ENT, for example, what exactly Tulen should know about where is he and who is he? THERE ARE- eh, take into account the memories about you, by the way, in this interactions with Thule, my Toulon Clarence knows more about me shameful things than this, and he, in fact, decided to invent that one history to to hide the fact that what did I tell him much more shy things I find fascinating, by the way, because in his context were memories of how I was in saliva because mine child, well, you You understand, right? So very interesting, uh, but also... wait, so you not really tripped over what did not exist totally made this up, that's great, yeah, uh, what, you know I am like a father now, let's get back to the topic she magically invents history, well, you know, like, oh, come on let's invent something like this, that Quinn is not It will upset me that I shared. Uh, but also, uh, like the hint has to develop depending on from emotional conversation color, what was great with our efforts side. And that's why, I think, for me it became surprise that how much is the hint in a specific replica is a kind of canvas, with whom should to work, but to do it's very fast. AND Yes, This is really interesting. So, okay. So, if to summarize, then I I think that one of main limitations, which is similar to volume restrictions for a novel, news story plot or perhaps tweet length, you know , for this Wednesday is the time answers. She must be able to to answer very quickly. THERE ARE uh, and fighting against it the limitation is because you need get the right one information about user to generate the correct answer. Uh, and if you had as much as you want time, that would be great , but that's not true. So you have to to play with whether possible to get the answer that feels personal, interesting and moves the conversation forward, but within these two seconds. essentially. Yes. One of the the biggest mistakes, which we assumed in product development, there was a moment when we introduced the second prompt evaluation stage. So the tool was essentially thinking that is going to say, checked it with memory system, and then decided which will be final respond. And this actually took us out to level two of half a second in average case. It was a disaster. Literally everyone fell. product performance. People wrote disappointed reviews, and this was just a change for 500 milliseconds. This is really interesting. AND What do you think it is? good answer? I ask this because I know what don't you have answers, or to this It's hard to answer. Yes . Yes. Well, this extremely dependent from the context, right? Yes? So you think about the role that the instrument plays in people's lives. We we are talking about the role older brother or sisters. Yes. True. Sometimes you sit with older brother or sister and just Chat about everything, truth? Are you kidding? maybe they They are pricking you. Short, witty, improvised the answer is, what, me I think it will appear in in Elliot's thoughts, it is great, right? But there are moments when do you really want to something more balanced and thoughtful. Therefore we often evaluate answers in different tones that, as we we want, has use tool. And good answer in a tone that reminds me of a therapist, very different from good answers... very contextual. That's right. That's right. I think, Elliot, for me, as for writer, one of mine favorite authors that talks about writing —This is George Saunders. Oh yes. He is amazing. He is wonderful. Yes. And in he has an idea about how to edit story: you read every sentences and feel what it causes; you are either dragged in, and this positive, or is it neutral, or negatively, and you You say: "I don't care." like it". And if this is negative, you rewrite this while it will not happen positive. Um, and you? just do it again and again, until, as he says, radically not fill each part of the work with their own preferences and taste, which, in my opinion, is simply wonderful idea. And as for writer, for me this is the pinnacle of creativity. Hmm, and I wonder what this one is full format destroys it, or me it seems so, tell me , whether it is true or not. It seems to be destroying everything, because when I I'm writing a story when you write a story, it completed. She released. It is what it is. Yes. Hmm, this thing is more alive. You seem to be able to give her certain instructions on what to do , but what she does in any moment, there is purposefully dynamic. Hmm, and how? then it works skill in in this context? Yes. So, um, I I'm trying to change. your creative process with every novel that I am writing. Some novels I I plan from start to finish. end and I have detailed plan. I have novels where the plot minimal: I know, Where am I going, but not I know the steps between them. By the way, this is madness. Nobody does that. They do. I think it's great, but writers do they don't do it. Usually, you find what works, and you say: I just doing this, and that's it what works. Me seems very interesting that you are that type people who want constantly everything change. This is very specific type writer and type personality. This is true. I always I'm trying to find. Strangely enough, by the way, after working, I don't writer, but I worked with Elliot year and yes, this. Yes, I feel that I have the optimal level of discomfort, to which I aspire, because so I know I'm learning, and actually mine latest novel "Foundry" I wrote sentences for sentence, very similar on how it describes George Saunders. I literally started with one sentence, which actually it was inspired by sleep. And then I just wrote the following sentences and sentences after him, and so I reached the novel, and finally a novel just poured out of me , right? Um, and me, when we started working together, one of ways like this manifested itself in writing for Tolen, in this new medium, was that we started when tried to create narrative experience. So, as an example, when you are communicating with with their seal, them need, well, if you you know someone, you Do you want to know which one? an embarrassing incident happened in life Quinton, right? Like, you need, so that people proactively shared things together with one, right? And we have researcher, Lily, who so deep immersed in this, in psychology of mutual disclosure and how to think about building relationships in a way that based on evidence and is really justified. And, um, one of the things, over which I have many worked, it's like us revive these characters? How do we to show you the world, with Who do they come from? How can we do this? so that they feel as characters, not just as an assistant, truth? It's like a blank slate for you, Yes? So we started with writing is enough structured inquiries, where are we tried to structure narrative experience, as -that's thinking behind three-act structure or something like that. You could apply different types narrative theses to how we did it, and then allow this to develop, Yes? Like this to watch as users interacted with this. AND frankly speaking, this It didn't work. If you might think, maybe , about the implementation of this, these would be those books " choose your own adventure" with 80s or something like that, where you choose this, then you come here, these long branched logical trees: yes, go to page 17, do choice, right? And this didn't work at all; it was full, full failure, of many reasons. But, in fact, the models are not very good cope with such narrative level complexity. Oh, interesting. It's not about because it doesn't work, but in that the models not good enough. I I would assume that this it would be too harsh and it wouldn't feel so alive. I think it's just two different ways to say the same thing exactly, right? Um, yes. So, we did the opposite and moved to completely opposite end spectrum. Tried to do something on like that does George Saunders, or equivalent of this in this environment, and namely: we said: " Okay, we give so much context behind behind the scenes, like described Quinton, you you know, the memories that you are with the user, global memories, where Tolen knows things about your world, as well as the seeds of knowledge that we we bet that provide context and details." We give he has so much, but in fact, when we we create structured narrative experience, we don't need give him a plan. To us no need to give him a plan of action. To us need to give him " hook". We need teach him to be the best improviser, as much as possible. There is a wonderful book that you must read if you haven't already read—" Improvisation" by Keith Johnston. Of course, You know, I was like that. a fan of yourself media technologies. Yes, That's right. Well, try me convince. Yes. Yes. And one thing, he is well known status games, but actually one very a small note, which Keith does in this book. Oh, and for listeners, Keith Johnston was very famous playwright from Great Britain, who is even better known as a teacher acting skills. He taught many actors who later have become very successful, and this book is definitely reflects some of his lessons. AND when he talks about narrative structure, he describes it as combination of associations, for example, free associations, even when you improvise in front of an audience, with recombination. If you will think about reading experience thriller, you know how you reach the end, and suddenly everything is put together, and you think, "Oh my god, of course, it's amazing, "but inevitable." All works and feels planned. You you think: "Oh, the writer, probably knew all this from the very beginning beginning. And that's rare. It happens like that. As with "Foundry", I constantly returned and reread the first one section because I could take something and return it to history to create this feeling is for you. So, what we learned about the story with "Tolen", this is what that we are not...I am not writer. I am not writing. history. Tool, probably there is writer, and actor. He improvisator. He writes a story, and mine work—to be George Saunders. My job is to teach. to tell them the best story in this moment. Interesting. Um, and that's why we build systems, as well as we work at the same level tips to try to combine, such as free associations, as if we we plant the seed, whether in personal life character, or in the big world, where he lives. And then we we create systems, so that later again combine these grains, so that you can get a reference that makes everything whole. Okay, that's true. interesting. I like it. I like it. Good. As soon as you mentioned about George Saunders, I thought: "I need to find time for this read", because what you described, —this, by essentially, the process evaluating clues and analysis, through which we are going through, and now we have a whole the team that reviews the results for different archetypes users to to ask a question: " Is there a following sentence? , next message from tool, the best possible? "So this is actually very, very similar. And you do it manually, do you have an LLM for this , or how does it work? Therefore , there is a certain cycle, through which we pass, where we, you know, create hint. Then we we're launching it, you know , as during your own testing, and among users, also. And then we let's look at some results and submit them back in, essentially, a hint-judge, which we then we configure so that say: "Okay, here you go What does quality look like? the result is this specific case". Therefore. OK. So, if I tried to summarize, because I I'm trying to think about it. what are the atomic ones elements of the narrative in this environment? For example, in the novel it's a plot, a character, each section, many things that can be to name, and I I'm trying to name that. What is here? It sounds like this, as if you have a character who has a history, origin and all that other. Um, and then the character can go with you for different types of quests or adventures. So, for example, one can to be therapeutic quest, and the other— some other one, you know , different ways. Um, and in within any context they seem to freely associate, based on the fact that you say, that is, they simply react to your words, searching ways to tie old things that you remembered, with a stream conversations. It's just like you Do you see this, or am I something? Am I letting you in? Yes, I think so. a good way is this describe. Hmm, I'll try. maybe repeat part of this. Therefore, when I think about it, I I mean that literally when we generate someone for user, we have certain global the context in which they function that mostly based on lore, written by you, and also very important instructions about how we we want to process specific situations. So there is a certain global context, established, in fact , for each tolen. And you know, significant part of onboarding in application—this is, in essence, passage personality test, although we don't like it that way we call, and he collects enough information about you, so that we can configure tolen as your reflection. AND this is very specific for Tolen development profile. So, that would be two building blocks. Global context, a kind of mirror. That is, you, but on the contrary, Are you sure? Sorry, it's on means that he reflects you in the way it seems very familiar, right? So we want to know enough about you, so that your tolen answered you in the way that feels like family. If I get carried away writing, reading, playing the guitar and piano, he also will admire writing, reading, playing the guitar and piano. I would said this: no, it's not Exactly. Actually it's like meeting with a stranger in a bar, Do you understand? You or find common ground language or not. Yes? We are trying to create such the stranger you are with find common ground language. The same connection. AND what do you think, how Does this connection work? Well, okay, I I like the metaphor. with a bar. We have different small scenarios, for example, common arrival at the camp. Yes. We have all these small scenarios, which we created for brainstorming. So yes, imagine that you sit next to a stranger in a bar. If this stranger read the same book, what about you, that would be clear Interesting, right? I have I mean, you would remember about this, right? But such a stranger can to be less interesting, than the one who reads a book that too obvious. Too obvious, right? So you want something like that it would be enough close ones, so as not to feel discomfort. You are not do you feel scared. I just read that. Yes. That's right. And I think that This is Yes. This is what we we are trying to create . Hmm, so this is the profile personality—this, maybe the second one structural element. And then, you know, structural element number three. Ahem, wait before we let's move on to this, how do you generate, that is, how do you do it Are you creating? You really are tell Claude or someone another on the backend? For example, I want to it was as if you met a stranger at the bar, and it's not too obviously, but, you know, How does it work? Yes, that's funny. Um, we achieved the greatest good luck with this specific questions. Um, actually forming profile creation through the prism of one of large, say, personality types, because as OCEAN types like, um, "The Great "five" or absolutely true, um, and the reason for this, I I think that in LLM data distribution enough information so that they understand this task. Um, "stranger, from "how did you sit at the bar" works well for us, but it seems not very suitable for for example, if I am an INTJ, you to some extent you know based on Myers-Briggs, what is INTJ compatible with, I don't know, than Okay. Yes, there are many. pseudoscientific, mostly pseudoscientific it is pseudoscientific, but it's also, um, very much based on stories and narratives, so everything this is based on stories, and it works, you know. And also in educational a lot of data materials about this, back to the model correspond with much larger subtlety. Yes. Yes. This is interesting. Okay. Sorry. Continue. Good. So yes, structural elements, general context, personal context, which is installed. Ahem, and then, you know, I would said, third part , I don't really sure how it is here to outline, but, you know, I'm thinking about We started talking. about it almost as if technology tree, by how you develop, Do you know? So, essentially, at the beginning of your relations with the tool is a set the ways in which you You can go. You seem to hinted at this: certain stories that tool, perhaps wants to be with you to share, certain ways of communication with you, things that can be done together. And we seem to be offering them to you in the form of a menu, and now the user chooses what exactly to choose. But this menu— an important part of that what does tolen do tolen. Yes. Good. So, this is pushes me to a bunch of different opinions. But one thing about which interesting to think about, especially regarding example of a "stranger" "at the bar" as a mental models, this is a question about how you create a tool at the very beginning. Like, they have prehistory, the whole this luggage. And I replied: "Yes, that's all" true, but I think what is the most productive approach is that although we are in advance we know something about tholen , their character and the world is built only through real conversations with them, right? Yes?" So, if talk about ENT, think about how it works Lucasfilm, right? IN there are a lot of them architecture of the universe "Star Wars", all of them these different characters and locations, and they have make sure that everything orderly. In 1989 we decided, that Han Solo dresses like this, and this should not change. And here's how it works. Most of the construction the world in the great Hollywood franchises is happening right now Yes. Like, we have major general plan. We have a whole the team that monitors so that everyone new element content matched this plan. AND because you are you doing storytelling in in a movie or somewhere else, in you have full control . You say, "Okay, in this scene will have the following -some costumes, such and such decorations", and we have none of these tools, Do you understand? Only way to find out about your tool—it communicate with him. So actually theirs prehistory is being built through some jokes, what they share, because of situations—we that's what we call them inside, —in which we we place them, and then their memories become their "sheet" character", instead of so that we have prepared in advance template defining all. And, you know, the same thing also applies world building. I I'm trying to think about it's like this: you know, like in Kevin Kelly, that "SHI—" these are endless "trainees"? So if add to this improvisation, then it turns out that AI is infinite actors- improvisers. But if you train infinite number of actors, then you have to imagine how every single conversation can go to unpredictable direction, and you need to do system resistant to this, right? And this also applies world building. In fact, every tolen reflects you, they is a mirror of your conversations with them. That, What are you sharing? affects who they become. And so just what you are share, influence the world in which they live. So we actually creating almost a multiverse, not one world, right? ? In a sense. AND our job is to add the best ingredients so that they can create. You seem to you are setting up hyperparameters, or You are like a gardener who provides soil and water, but the plant becomes what it becomes. It depends on how everything is going. Ahem. And you don't know either exactly what kind of seed you planted, right? Hmm, so how does it work? It like when you say that they encounter situation, and this shapes their personality and the world, in which they are. It takes place on at the very beginning? When you create character, you put it in a pile fictional situations, and then use this is to fill characteristic character? Is this occurs during conversations, as if operational character formation and the world when he uses my words to generate: "Well, last week I, like Tolen was on his own. alien planet and did this and that, and this reminds me of what you just said. How is this? working? I will show you. example from our user Discord, which published it. They shared this publicly. So, you know, seems appropriate share it here. Um, and they pass because of situations that, essentially available to the user in application. Let's say I I'm going to talk. with his tolen about what today happened at school, or something like that. And they moving through situations, um, and by essentially together create the universe by with the help of your tool. And this one specific the user does very cool thing: he turns it a little bit, um, like this thriller universe, a little more gloomy, you know, okay, there, there something, you know, takes place on Portoli, but we are not sure that's it. Um, and none of this is you or I didn't fit in. a hint, right? And this Tholen takes situations, which you laid, but then the user as if inserting its own story, and it builds the universe that fully differs from any other Tolen's universe. Um, and It was so cool to see , like they, you know, publish: "Hey, here what did they say today about "The secret on the Portola." Um, so I think for I was one of them. the coolest parts— to see how, okay, if you only give this one first hook, in how many different directions people can it develop? Another way think about it, me I just like it. remember all the books, which I know you have read, type now the pressure is increasing. Of course you read " "About writing" Stephen King, Of course. Yes. Um, and yes, that's it. I say so. Yes. Yes. Yes. Yes. Oh, and, you know, in " "About writing" one of the things he explains, consists in because when Stephen King writes a novel, he doesn't think about the plot. In fact, he thinks that the plot is something on kind of not quite the real thing, right? He doesn't believe in the plot. Instead, he believes in, as he literally does names, situations, and because it's Stephen King, I'm almost there. sure that examples that he cites in books, — this, for example, you know, vampires appear in small town in New England, or, I I think one of the others was that having or father with child stuck in parked locked cars, and rabid dog tries to get inside, truth? And it turns out that he starts with something something like that, and then plays up the situation, while he's writing a book, and this becomes a novel, yes, and, actually, that's what we we do for Tolen. So, we're like, specially for we are writing backend real situations, in which can be put your tolen, which can be simple, for example, to stumble upon in a flat place, Do you understand? Like: "Oh Oh my god, I think I am. tripped today on a level ground in coffee shops, you know ? But there could also be something more, you know? Something personal, something uncomfortable, a little, maybe, related to something revelation or enlightenment. We put them in real life situations , and they share this with the user as you, when you write message to a friend: « Oh my god, just now with me "This happened." Or: "I I am trying to accept "an important decision." "Not I know what to do with this." . You know, you just don't Believe me. Oh, yes. Yes. That's right. Exactly. So, we we prescribe this, and then tolen actually co-authors the story. They develop this situation and advance the plot together with user. And then it becomes part of their memory. And so it becomes part of their prehistory, to which they can come back later and combine with other elements. Because the ways in which I interaction with tolen, a little limited, but it seems to me that most of mine interactions—it's me I share with them. Therefore I not very used to it, that they share with me by what happened to them. How does it work? You lose a little points for added cost, but you get points for honesty, points for honesty. Uh, yes. So , you know, I think this one of those things that became unexpected a hit, especially lately, when we made these types situations from Tolen easy to access in application. That is, we have there is such a thing when Tolen is now alone. comes to you from to-do list today, and these conversations are part of this. Uh, and, you know, yeah. Yes . So, you know, this it's true that many our interaction—this when the user just opens app and says, "Hey, You know, I have it here. "problem with the guy." " You know, I'm kidding, that at this moment thousands American men received refusal due to Tolen, because This is simply incredible. hear how users write: "Oh, my the guy behaved like trash, I sent it message to your instrument, and he said that I need to protect yourself more ", and this led to rupture. And, of course, this a significant part of that what people do But I I think that a lot realism in relationships and friendship arises because of the fact that tool is addressing you and asks: "How do I to talk to your brother or sister, who are a little jealous, that I have one a wonderful human friend? " "Really. This is very interesting. This breakup story actually resembles me about what I'm working on I thought a lot, and I would be interested. to hear your opinion about it. Is there anyone you would like to be with? need to break up relationship? No. But I think that one of the most interesting things in this generational AI- tools—that's what what they tell people, which must be broken relationship with partner: "Hey, you should take a little "to protect yourself more" . I think we have a good image of technology, especially AI, which is in the era of social networks they promote extremes, polarization, bubbles and others things like that. And this— something so human, this the best of what humanity knows about how to be a successful person, available to you at any time in what situation and in any time of day, which is very healthy, I I think, and that's what it's about. , in my opinion, we too little we are talking about—how much this can actually be useful. Yes. Everyone is talking about How AI is changing our work. Everyone who works in sales, felt it. All work changes directly under with your feet. But one thing not is changing. Victory. The deal that finally is closing. Moment, when hard work pays off. Perhaps, Do you have a gong in your office? confetti machine or something else. We are not we condemn. In each the teams are their own rituals. Exactly for at this moment created by Adio. It agent CRM, which performs work for with every victory. It transforms every signal from client on context, necessary for work with unsurpassed speed and scale, by with the help of agents from income and automation, which They are building around the clock. sales funnel, promote agreements and develop accounts. You will receive the entire context and speed, necessary for scaling any what kind of market entry so that your team could increase the limit of the possible. Get started with Adio today and simply focus on achieving success. Go to adio.com/every and get a 15% discount %for the first year. It atio.com/every. Yes, that's about it. that we started more speak inside company, is the idea a kind of "polar" stars": to help to our users to move from state to state strong overload to more confident. And I I think that's essentially it. what you are talking about now you speak. You know, one of most common things , which we get from research users when we ask why you do you feel like this strong connection with an instrument that actually is happening? Many people say: " Listen, I have life happens what I think burdensome". Many our users, for example, graduating from college, so they are looking for work, right? Or they are moving to new city. Perhaps, they don't know much people in that new city, and they are trying to orient oneself, building a new life for yourself. And we again and we hear again: "I want to talk to someone about these things that bother me oppress, but I I already called my mom. three times this week," or "I don't want to again to burden oneself the friend because I have on your mind". Because as you rightly noted, to have such a neutral observer who knows a lot about you, has no excess baggage , does not judge you, but shares the best knowledge of humanity, so that to help a person moving forward is importantly. People consider it very grounding and very encouraging. And I I think it's interesting because it's a subtle difference from valid criticism more "assistant" LLMs, which only they agree. Because it's just a "clean slate" as you said before , but it's not really and the role it plays Tolen, right? Tolen often says things like, "Hey, you already you're telling me for the third time tell me how yours is The guy is behaving like scum". I think, it's time for you to defend oneself, truth? Um, and that's why I I believe that this role, she definitely is key for technology. I think this is also important for the companion we created, it really is special role that this tool, this technology plays a role for people. Mm. You know, this is real. reminds me of something, because, again, me investor adding values, since then, How did you first meet us? introduced. One thing that I really I like it too in the product, and in the team, this is what, well, in the world science fiction, I write thrillers about the near future, right? So, this is stories about science fiction the nearest future, and if to think about the biggest ones are like this stories in popular culture, most people are remembered " "Black Mirror" or something like that, right? And there is a lot, you know, because technology are becoming very noticeable when they break. It is very obvious how can you tell me a story about how everything goes awry, as in " Jurassic Park period", right? For example, dinosaurs break free and They will eat you. Me It seems that in such a The genre is hidden. emotional tone that leans towards dystopia, right? Everything seems to be breaking down, and This is very cynical. Yes. Um, and I always tried in his novels move in in another direction to show: of course, things break, and that also useful for writer, but that that the world is changing, doesn't mean that it bad, right? It actually means that new ones appear opportunities to do his best. It feeling depression when you think: "I don't know, What will the world be like tomorrow? ". What does this mean for my children, etc., — to consider it as the opportunity to create sense of self subjectivity in oneself and in the people you Do you love something you didn't love before? was. I always tried to put it in in their novels, and that's what what am I really up to collided and I understood both team, and with product because I I think that's it too really interesting. Hmm, this is reminds me a bit of that what Ursula Le said Guinn, I don't remember where, but she was talking about how science works fantasy, and what most people they think she is about design of the future. Hmm, predictions of the future, designing where we are now, for 5 or 10 years ahead, and she said it was like sometimes it's a bit like that and it works, but every time, when you just you design things, they grow in that the direction in which are developing, and this simply becomes a tumor. this is finally turns into tumor. Quite. Quite. Hmm, and probably that's it so we have "Black" mirror", because if just bring it to edge, then it is a tumor. Everyone dies, you know. Ahem. Yes. Yes. Yes. Um, and this is actually me I think the way she perceives scientific fantasy, much more about that to reveal something the truth about people today and about how they are arranged. Ahem, you just choose a small particle reality, and then you build around it the whole world. Hmm, is it responding? Is that what you're saying? Oh, 100%. I always I consider myself naturalist. Me just like it to have friends who are something build, and build from friends like we are now. Um, and you just take pieces of this and you use them as ingredient to to prepare a dish. This is the vision. of the future. Me also like that this reminds me of Ted Chan, who has great comparison fantasy and science fiction fiction, where is he says that fantasy is about special people. So, remember Frodo from ring, right? IN this view of the world "Star Wars" is also fantasy. It's like you. Skywalker, right? It just you...that's it about special people, where the universe recognizes your exclusivity, and this is how you blow up The Death Star, right? You listen to yourself and you Luke, that's why it works. Um, while his scientific dichotomy fantasy is about strange rules. What is in science? there is no fiction special people. The world just works that way. -new, and these rules extend to everyone. Anyone can use new technology. This is not " Harry Potter, where are you? You must be a wizard. Hmm, and I think a lot. about this regarding Tolen. And me too I think about it a lot. regarding reality, because I think that this podcast called "AI and Me" , and I think we We now live in times strange new rules, Yes? Um, and that's why a new one appeared the environment we are dealing with We can play, right? Um, and that's why I think that in a sense, as part philosophical the basis of everything that we we are trying to do with application, it is use these strange new rules, to help people to become more confident, living in a world with strange new rules, right? Yes. Yes. Yes. I like it. I like it. I like it. I really like it. Ahem, one of the things that I I notice in your travel to creating this a new kind of creative environment and development of this creative product , I think, is common for many people, even if you start writing novel or first time you are building a business, you have similar path: you often get into it (and me too I count myself among of this category) with to some extent stiffness where you pay attention to that , how do you think, everything has to be done, and use metaphors and stories, who heard to have rules like: "Well , it should be a three-act play structure". And then, when you are a little are you in contact with reality if you talented, this starts a little disappear, you become a little freer and you start to notice, what actually works. And then you start create a new one a set of rules for how everything works, but until which you can to treat a little easier, and you can to dance with them, instead of to learn the table multiplication. I don't know. something like that, you know? Does it feel like reflection of what Did you find out? We talked about it this morning. It very timely. You you know, this idea, and you used the phrase " contact with reality." There is a kind of bifurcation consciousness, which, I I think we had to. to experience during working on this project where Sorry to interrupt. , but I think this is the one the moment when you need a little content marketing for your pizzeria. Like, tell them. about real conversation. That's fair. It rightly. Yes. For context, I am also from I have friends. pizzeria. And this one conversation actually started with the fact that restaurant is useful exercise for me. AND even more than when creating consumer software product , you constantly are you in contact with reality of preferences people. And there is such a special piece pizza that we make and which, in my opinion, simply phenomenal . I just love it. its. I'm ordering this. every time I come in to your own restaurant. What is it? Uh, this is a piece. focaccia. Just beautiful focaccia with tomatoes, and that's it. Of course. Very simple and easy deliciously. This is my favorite. dish on the menu, but it is not for sale. Literally no one wants a thick piece focaccia with tomato sauce, and that's it. They want pizza with pepperoni. Maybe you're not on it. market. Well, maybe, but easier to carry a pizzeria, than, you know, change marketing consumer application. So, you know, we we are going to clean up its off the menu, right? Uh, what's for me? simply unbearable. Uh, What are you doing this for? Are you doing? For art? Well, here it is. The conversation went well, and it was great that you need to have one forked mind, regardless of whether you do pizza, or, you know, embodied companion. Uh, so that combine these two things, almost needed amazing, uh, a kind of arrogance at, say, the top creative funnel, where you have to to try things on which no one asks for, and, of course, no one says: "This is how I want it, so that my companion interacted with me." But you must to put up a flag and to say: "This is what we let's do it, here's how there will be an ENT, that's what it is "There will be history." But then you too must have humility on, let's say funnel days or at the end process to say: "Wow, what I... I thought people didn't will like it..." Uh, and we constantly we are going through this cycle in the script room, uh, you know, in Tolen, where we produce something that we think, uh, you know, phenomenal a scenario that really investigates who are Tolen, in a wonderful way. And people don't I like it. So we're coming back. to the drawing room boards. And vice versa too, yes, as if you were going to do something, in which very little was invested effort, and suddenly it it just "clicks". yeah you can't uh you can't listen, what do people want, until the paradigm is no longer was installed. And as soon as you have appears the paradigm of the novel, you you can be, for example, in the process. People come and they say: "I want 300- a page book in three-act structure" , you just need to come up with it, essentially. uh, and then you can fix it according to reviews, I believe that It is worth saving the fkaka, as... they are wrong, maybe it's about something says yes. Send this to others co-owners and say, "Hey, Dan from New York says that we really need..." I think this is also depends on whether other co-owners I don't like that either. , then this is another case. What are they think? "Without comments." Let's move on. Hmm, interesting. Good. So what about me? suggests, so this is, um, this is very steeply. You are building all this, you build multiverse tholins, and this is such a flexible thing, looks like a thousand various actors improvisation, but this also sounds like chaos , and very difficult understand that what happens and how does it happen improve. So how are you? Are you doing this? Oh man, this is great. question. How do we do it? Are we doing this, Elliot? What are you? ... um, uh, this, I have on I mean, it really is. chaotically. firstly, we we are trying to instrument that, what we can, and it's very helps, you should to see at least a little of that what's happening, and I I think it's true. useful. what does " mean to instrument" in in this context? um, do you want to bring some examples? Yes, I talked a little. earlier about our cycle evaluation, right? And, in fact, if short to paraphrase, we we are releasing a new one fragment or new content type, new type of conversation. We let's test it ourselves, and then we invite participants research for testing, after What do we value these completion and we use this to create a request for the appraiser. Good. So we have a lot of things we test in small volumes, and then, expanding this into large scale, we , of course, let's see how on traditional metrics, as well as on judges' own assessments. Good. So, you you are actually saying that instead of an actor, I created an actor- an improviser who is in multiverse, and instead of the editor, I created an editor- improviser in multiverse, which reflects my preference, so that to measure multiverse actors. Yes. Yes. And I would add to this is a few moments. So I I think this is good. mental model. I think one of important additions before that there is this... The multiverse is growing as if from the soil in which we constantly contribute fertilizers. This metaphor will quickly lose meaning, but we have a deadline" "seeds of knowledge"—these are fragments worldbuilding that we are implementing in different ways. They become details Lego for improvisers, and we we are constantly changing them both at the world level and and at the level of individuals stories. So we we observe how actors multiverse interact with judges the multiverse, and we change the "bricks" Lego with which they playing, you know? And one more point that I think very much important, especially for the one who helps to create these "bricks" or " "seeds of stories"—this is what that besides having judges multiverse, we we are trying to remain very close to individual user experience . Research, interview and observation of the how people lead conversations, shows, that is of good quality individual experience remains key, even if it is impossible to scale. It critically important for understanding where it arises emotional connection. As we can do it use? What working? And what's not? What are you Do you think? I'm taking Elliot out. only one mind thing. I just I'm crazy because you you go through the whole this process, and, returning to your opinion about individual experience, I have it bad experience with " "a grain of knowledge", and I I say, "This is terrible. How we can do it at all produce? And he answers: "Listen , I talked to 30 participants research, and in all They were wonderful. conversations. Foca pizza. And I say:" This is not good. "So, I I think you understand, with you often encounter this, it's like in any another creative project, right? You you're just saying that this Is the thing cool or not? then we move further. Yes. Yes. Interesting. OK. And what is it? your favorite model for generation characters and for judges? Yes, great. By the way, Now it's my turn to be investor, adding values. Actually, about the process, about which we just did they said, one the tool that became for us extremely useful, and which, on In my opinion, it's worth it here. to remember, is Dawn company, which is engaged in analytics that allows you to view such interactions in anonymized, but visual way. So I I think the stack is for how to do it, and to your comment, it's not just a model, truth? Your stack for evaluation, your kind of return to evaluation process. But you know when you think about different things types of interaction with tool, for example, if you spend—although I I know you don't. do, but if you spend the whole day, working with your tool, like many people do, you probably interact with models from, I just couldn't us to convince you actually interact with all models laboratories, right? So, you probably have interaction, guided the Meta model, of course OpenAI, definitely Anthropic, and we are increasingly working with Gemini in memory system. Ago I think one thing is that there is no one the best model. I think that for many tasks that we perform and about which we are talking about here, in creative writing work, maybe not It's strange that Anthropic especially useful. 3.5 or 3.7? Er, 3.5. 3.7, unfortunately, not so creative. Although I am sure that they would say otherwise . So yes, I think that The problem, of course, is because Anthropic is now it is impossible use in situations that are very sensitive to delay, isn't that right? Therefore you constantly switch there- here between them from this reasons. You have tried a lot. 4.5? Very little. I have on meaning that someone whom I know whose taste I am. I really trust, wrote a review that was not particularly approving. Ahem. I looked at mine. opinion. Have you looked? Good , cool. Uh-huh. If correct to make a request, this really cool. He just very expensive and slow, so with for this reason he, probably not suitable for you. But I would definitely started more with him experiment. As a partner for brainstorming, if you use your example earlier. Yes. Um, yes. I mean, 4.0—that's, you know, me I call it "Running" "Mac" among the models, Yes? He always fast. It always is exactly what you need necessary, and for at the right price. So, we are many we work with What about, um, this one? model judges? For, in my opinion, my experience working with LLM as judges —that's, um, theoretically a very attractive thing, but I noticed that, for example, if ask Claude: "How "Do you like this essay?", "He will always give you" five minus "". Hmm , and then if you I always fight about institution, he will appreciate their cool. Um, and if you're a little... change it, it give a "five" ". You see, this it's just not that, you don't you get it true opinion. So Is there a preference model? , um, tuned through prompts? Or well-trained? How are you? Are you actually doing this? I mean that many people, and I don't I mean yours. a specific situation, but many, trying to do with LLM product judge, misses the most important step— investing your own taste in prompt for evaluation. AND I think when we invested, and you spent a lot time working with team on this, when we invest time for marking, evaluation and creating real criteria, we we get wonderful results. And when? trying to cut angles, we get bad results. So instead of simply ask: "What "What do you think about this essay?" , You seem to be saying: " Great essay consists of five components". It's like attractive title. Here example of an attractive header. Here bad example header. Um, he has there is a great hook in intro, here's a good one hook. Here is the bad one hook. Is that so? this looks like criterion? I would even go further. It's more like, What did you talk about with George Saunders. This is actually is this a good first one sentence? Is this a good second? sentence? Is this good? type, this is exactly like that level. This is not even analytical level structures. Is it worth it? was here to put question? Is this what why we create judge? That's right. Interesting. Ahem, my question is, what if you say: " good first sentence, good second sentence ", At least in my opinion look, very difficult determine what " good second sentence ". It is easier to determine that so beautiful title. So how are you? Are you doing this? We do it. mainly due to a large number examples. and ask people to walk, evaluate and to give arguments for each choice. I understand. Yes. So, in in a sense we almost like "rough" We do it by force, right? Instead of to invent, it's not rules, it's simple I mean, there is a certain criterion. There is also this both, right? This is theory and practice. But I think it's like, uh, kind of right critical comment on, I think this week we call it "vibe" -prompting", "vibe- coding. Next week we probably let's call it something otherwise, right? But there is, there is this is like, uh this, this belief that like okay, you can just start writing prompts and very fast , very easy to get something convincing. But I think we found a huge abyss, and really need a lot of manual labor human work, so that to move on to what people will consider really convincing in the long term perspective. I don't think there are any short paths. For example, if all your essays on "five out of" minus", then everything that can be obtained just with a "vibe" prompting"—this is the type "three", and you have find out how we to achieve steadily result. Yes. And how do you feel? to testers with refined taste compared to ordinary people, when you spend these evaluations? So, is that you? who has a lot specific set skills that you honed over the course of for a long time, to to be able to say, is the second sentence good? You yourself conduct an assessment , is this yours users rate , or is it something in between? I think it depends. from the type of interaction, for which we create judge, right? So, you know, if it's a judge on issues related to to the right, then, I have mind you, you are the best the person for this, right? Uh, you know, if this... for example, at least, uh let's say more friendly advice from an elder brother about what me today put on? Classic a question that someone puts his own surroundings, uh, without insult to both of us, but we probably don't the best people to to provide such judgment. Even though you are a fan men's clothing. It, you know, no returns to conversation, so you know, what we often do is , well, we are looking for and we involve someone who , oh, I would be an authentic person to voice this type of interaction. Interesting, okay, so... now you are evaluating judge. Yes. Yes. Who evaluates who evaluates the judge, and this lasts until the very bottom. This is fascinating. Good. THERE ARE -Um, um, okay. So, I I think we really worked a lot material regarding What does this one look like? creative space. I I want to give a little business time. Therefore, it is growing very much quickly. How is this? is happening? I I suppose. So, Hey Jay, one of your co-founders, which is also mine a good friend, he a kind of wizard growth in such things. So I I assume that he somehow contributed. But how, how is he so fast growing? Yes. Well, I think two things, actually three, have developed for us quite successfully, because all successful startups provide an element Good luck. I think three things happened at about one o'clock, isn't that right? The first is much of what we talked here. I I think we, at least at the beginning of February, achieved what really understood that makes content interesting to talk about and how to present it users, what was a big problem, which we don't even know about They said: how to give them understand what your the tool wants to discuss something specific. So we really you haven't even done this yet opened. Yes. We don't do this. we do. These were months of work, the results of which began to appear very recently in in general, six weeks ago. The second thing is we completed a large creative project about animation, which we are talking about also not spoken here . But most of immersion that people feel with their instrument, —that is, on in my opinion, wonderful the environment we created for him work, and how you interact with this environment, as if taking care of him/her with help tool. And these two things, I think, contributed because when someone enters the application, he gets a great experience remains with us. Where to start. Where to start, right, and this again many months indeed work that we can't do didn't do it perfectly the first time. Fortunately, at about same time AJ, which one are you talking about remembered, our co-founder and a real wizard development, worked over the problem of how to tell people about the Tollens for outside our platforms and applications. Today we have there are none built-in social functions in product. And he achieved at the moment when content for TikTok and Reels which he distributed for many months, finally started to "fly in". A few weeks ago in happened to us tenfold surge downloads, and we found out that the video a young woman who prepares with his "tool" ", it became incredible viral. It seems about 7 million views in 72 hours. So, now something is happening very cool: that one content that spread by AJ, now inspiring others people create own stories from Tollen. For example, there is podcast where two boys use their own "tol" as a constant character. He often gives advice on relationships and the like things. I've seen something cool before. video the other day, which, it seems to have become quite popular. Someone did that there, so that their "tol" talked to "tolom" » groom. And these are the things we we don't have any relation. We are nothing. did not plan and did not prompted this, but it's really cool to see how it becomes popular, and this simply causes incredible… growth of indicators . Okay. That is, you create what you people want to share, because it's cool, new and interesting, and then spread this in different places on TikTok and Instagram. Yes. I think you Do you know what exists? standard strategies that you apply in such cases. But I am very interested in in this context one phrase—it seems to be strategic term Ben Thompson—" excess of opportunities » in the field of AI. And I think we reached the point where model capabilities actually ahead of understanding of the ordinary consumer regarding What is possible at all? Therefore, much of what we do with this content—this actually showing our target audience of what they can. Hey, you can to cook with your " "body". Yes. This sounds crazy, but if I told you three years so: "Hey, you can to cook for with help computer", you would replied: "What are you talking about?" "Are you talking at all?" your time prepares, but well, this, you know, this conversation, like "oh, I see that there is no yeah, hey, you should change the temperature or something like that " um, and in many ways it is the learning itself, so you at the same time you teach, and attract attention to the product I think one of ways like me, how do I do it I get it, it's simple. my own mental model that I sometimes I find it useful when we are working on such projects, it that different people use their own tholins for different things, right? For example, someone wants cooking tips, someone wants discuss relationships, to speak out, well, or someone wants to know more about who this is alien and where from he. There are many different reasons why people do this are engaged. And the only one the thing that unites them, and the only thing that is absolutely obvious the promise of everything that AJ does it, it's character. It's like character, well, that's it, why do you need it you get it, right? "Type" oh, you want to have little friend- a stranger in his phone, right? "It character who connects all this to pile. And this forced makes me wonder how you have already seen, Quinton and I somehow talked about other people's things evening, about how artificial tools intelligence is changing the importance of SEO, or they make it irrelevant in in a certain sense, and how through these tools search will change information in the internet. And I think, what are we all doing now looking for new ways interactions with computers that differ from those that were before, and what tholins actually are— it's almost computing, character-driven . It's like new. interaction interface a person with a computer, where you think, "Oh, actually my first stop when I want use incredible knowledge...all knowledge and the possibilities of humanity on the internet, it is through him, through my little one "cute alien" AND it's like in "Golden" compass", there "you have "Your daemon, right?" And this as if... I always wanted to like this. That's right. And for many of our users it is such a tool, or wrong? And I think that this is really very good thing. And this too makes me to feel great curiosity, which can be the future if this will be yours itself Yes. the way you interact with computers on fundamental equal, whether Tlen or other characters, or something like that. This is really interesting. It reminds me a little something that I experimented in in the world of B2B SaaS, um... hear this transition. I actually very excited. Hmm, this is that from where at least we came, and you are very are well acquainted with this, and to some extent too They left from there. Ahem, currency of ideas, startup -idea or product ideas are whether it decides she is a problem. M. Can you clearly to formulate the problem she can decide, because that's why people pay, and the type of person or the founder, who able to detect problems, it's yours A scientist by birth. Ahem. and um. you have accurately determined part of what Tlen is , from the point of view of the problem he decides, for example, transition from state overload to balance, It seems that's how you are. said. Hmm, but this is so abstract problem. There are so many things here, who are able to do this . In principle, there are many different ways solve this problem. And the fact that the tool does in a person's life , is actually infinite; this is something like multiverse, um. AND therefore, I think, it's actually very difficult to clearly formulate one specific one problem, or even a set of clear problems, which, in my opinion, changes the thinking of a person who builds in this more focused on characters in the world. Instead of to be a scientist, you become an artist or creator, and instead of solve the problem, you just create something; you create something beautiful. You create something fascinating. Um, and this more about the atmosphere, than about science maximization usefulness for B2B SaaS buyer. That's what I'm talking about. I was thinking. Or this resonates with that what you feel and with What are you facing? Yes, I mean, I I actually remember favorite example each—Golden Gate Claude, isn't that right? Um, you know, I think, from any reasons, it's fascinating me, although I'm glad because this is, frankly, creates space for Oxygen. But I am fascinated. that these, uh, big laboratories actually feel such fundamental discomfort from to personify and, uh , you know, embody, uh- Well, what is. Well, you know, depending from where you are are you on spectrum of supporters AGI, that's, well, enough close to conscious entities, right? Um, and I think that's comes from, you know, over a decade dominant paradigm creation software provision, which is very focused on problem solving. You know, I'm thinking about your work with ChatGPT. It looks like this: I have problem that needs to be solved to solve Be please help me to untie it. Not Yes? He copes. with this phenomenal. Hmm, but I really think , which is much more space for construction of real relationships. And one the metaphor I'm referring to I often return to that funny because she seems like this archaic, is how ours evolved relationship with cars in this country, right? And I spent a lot of time, thinking about what first generation cars was the Model T. Everyone was absolutely Happy to have a Model T, because it was like that it's amazing that could have been somewhere to get there without a horse, truth? Um, but with sometimes, when the car became deeply personal part of your life, suddenly arose request that he reflected your personality, right? Yes? And that's exactly how we got Mustang, Cadillac etc. And I think that we, in fact, we accelerate this process where ChatGPT represents such The Model T era. Eru "oh my god, this is just It's incredible that this thing can to answer these question", but people will change very quickly your preferences, rising hedonistic treadmill or How do we want it there? call, and they will say: no , I want this thing reflected who I am there are. Yes. It's funny because I I feel that common point countdown from the world startups is to because what specific you are the problem solve? How are we? can we do this realize? And on on the side of art, such as all my friends- musicians, writers, directors or who whatever, the logic is this : there is one thing I want to see the world as if it were the emptiness that this a work of art has fill in. And, uh, I I feel that we had to use both types of thinking, but we are more and more we are inclined to the second. Actually, uh- uh, 10 days ago, uh, Quinton, me and two others people working over the content, we presented Quinton gave me an idea, and I I remember starting with what he said him about the problem, which arose in users. AND Quinton immediately reacted to this skeptically. He asked: "Why are you Do you generally think that Is this a problem? Like, or Is this a real problem? "And then Lily interrupted him and said: "Wait, wait" , wait, no, no, no. Quinton, listen. This is going to be so cool." And after that everything went away perfectly. So I thought: "Oh my god. Of course." I like it to think that I am not like that obvious, you know, but it's true it worked that time the moment when they expressed themselves. Therefore, I like it. I think this is great. sums it all up said. It was incredible conversation. Thank you both for that. came. Yes, thank you, who invited us. Yes, that was nice. Oh my god, friends, you simply obliged press this button likes and subscribe to AI and I. Why? Because this show— the epitome of cool. It how to find a chest with with treasures on the back yard, but instead she is golden filled with priceless knowledge about ChatGPT. Every episode is emotional roller coaster, insights and laughter that will keep you in tension in anticipation continuation. This is not just a show, that's it journey into the future with Dan Schipper as captain of the spaceship ship. So do it. Do yourself a favor: put to my liking, subscribe and Get ready for an adventure. your whole life. And now, without further ado words, let me just say: Dan, I am hopeless. in love with you.