Episode 1 of the interview series "How I Use ChatGPT," featuring Sahil Lavingia, co-founder and CEO of Gumroad. With screen-shared prompts and responses, the episode covers how he uses ChatGPT to research buying a NYC building (neighborhood selection, broker questions, property design), draft tweets and essays, pressure-test and research ideas, assist decision-making, and use DALL-E for interior design. Also touches on how ChatGPT could have saved him $70 million, its effect on top creative output, and why he believes people no longer have an excuse for not knowing things. Timestamps included. # Channel Identity Observations: "How Gumroad's CEO uses ChatGPT - Ep. 1 with Sahil Lavingia" ## Video-Level Presentation Evidence - **Format**: Interview/screencast hybrid. The host asks the guest to walk through specific ChatGPT chats ("take us through some specific chats that you've had so we can see how how you're doing it"), with the guest narrating on-screen sessions (e.g., a DALL·E chat about "58 Bowery," a downtown Brooklyn history chat, writing/essay chats). - **Episode position**: The title labels this "Ep. 1," indicating a numbered series. ## Explicit Self-Description (Creator Claims About the Channel) The transcript contains an extended channel mission statement: > "I have this theory that chat gbt is the most important creative tool of the decade but it's still so early that most of us don't really know how to use it there are a few people though who are living in the future... that's what this interview series is about we'll go in depth with the most interesting people in the world so you can see every detail of how they use chat gbt to think write make art make decisions build businesses one chat at a time" Claims derivable from this passage: - **Claimed scope**: In-depth interviews on practical ChatGPT usage across thinking, writing, art, decision-making, and business-building, examined "one chat at a time." - **Claimed positioning**: Showing viewers "what our future with these tools will look like" via expert practitioners. The host also describes the guest as credentialing the series' premise: > "my guest on the show today is sahill... founder and CEO of gumroad one of the largest platforms for creators to sell their work online he's also an active investor and the author of The minimalist entrepreneur" ## Intended Audience (As Demonstrated) - Viewers who already use or want to use ChatGPT for **work and creative productivity**, not beginners needing setup instructions — no onboarding or basics are covered. - The audience skews toward **builders, founders, and writers**: discussion covers startup pricing decisions, equity/409A mechanics, essay drafting, and investment research. - The host explicitly frames viewers as learners: "so you can see every detail of how they use chat gbt." ## Observable Presentation Patterns - **Interview structure**: Host summarizes and reflects back the guest's points ("so what what I'm hearing I think so far is um it's a tool for you when you have a random interesting idea..."), acts as audience proxy. - **Prompt-level specificity**: Concrete prompts and outputs are surfaced, e.g., "what material was 58 bowy constructed of," "add three to four paragraphs to make the point compelling also suggest more examples," and a cited rule ("Title 63") for landmark renovation. - **Recurring themes the host emphasizes**: refining rather than accepting first answers ("it gave you an answer originally that you didn't like that much and um and you just told it go again"), human-in-the-loop editing, and using ChatGPT to counter one's own biases and opportunism. - **Host self-disclosure**: The host shares a personal custom-instructions practice ("I'm a little bit too opportunistic I'm not being strategic enough here are our quarterly goals"), signaling a peer-level, practitioner-host format. ## Audience Address and Tone - Conversational, first-person, peer-to-peer; casual profanity-free enthusiasm with direct second-person framing toward viewers in the intro. - Guest and host both use anecdote-driven reasoning (subway research sessions, Dumbo real estate, Gumroad pricing), suggesting the channel values **narrative case studies** over tutorials or listicles. ## Scope Notes / Limits of Evidence - No curated metadata was supplied; all observations rest on title and transcript alone. - This is one episode ("Ep. 1"); channel-wide consistency cannot be inferred from this single source. # Delivery Observation: Sahil Lavingia — "How Gumroad's CEO uses ChatGPT" ## Speaker Roles Two distinct voices are present: an **interviewer/host** and **Sahil Lavingia (guest)**. The host summarizes and redirects; Sahil delivers the bulk of the content in long, self-directed stretches. --- ## Tone - **Conversational and offhand.** Sahil's speech is filled with hedges and casual markers: "I mean," "you know," "like," "sort of," "kind of," "or something." > "I mean so I I think of I think of it like uh you know if you think about high-risk investing generally as..." - **Self-aware about rambling.** He explicitly flags his own digressions: > "so anyway that was kind of a rant" - **Confident, declarative claims** punctuate the casual tone: > "there's no excuse for not knowing why anymore" ## Vocabulary & Recurring Phrases - **"Thesis"** — repeated framing device for his ideas: > "kind of like a thesis" ... "to me this is the thesis everything else is a little bit arbitrary" ... "that's at least my current thesis" - **"Network effects"** — recurring conceptual anchor: > "I really believe strongly in the concept of like Network effects" - **"Top 1%," "power law," "10x / 10,000x"** — investing vocabulary carried into other domains: > "the top 1% of movies the top 1% of hotels" - **"Poke holes in this"** — a named prompting technique he returns to: > "you can tell it your idea and say poke holes in this" - **"Fun fact"** used to tag digressions: > "so anyway fun fact" - **"At the end of the day"** — recurring pivot phrase: > "but at the end of the day you still have to like build the thing perfectly" ... "at the end of the day the decision is like buying the building or not" ## Teaching / Explaining Habits - **Analogy-driven explanation.** Nearly every point is grounded in a concrete comparison: - ChatGPT as ghostwriter: > "it's kind of like you're getting it to write a book ... in real time for you ... it's like a book that doesn't exist" - AI as attention mechanism: > "I think of a lot of the the innovation in in this new wave of AI as like peripheral vision" - Samplin: > "it's kind of like listening to like a really good Kanye West song ... how like he finds like crazy samples" - **Enumerated sequences** to structure examples: > "your first meeting with like a real estate developer and then your second me with a construction manager your third meeting with the general contractor" - **Hypothetical reframing ("imagine...")** as a recurring rhetorical move: > "imagine like if the equivalent of like every time you buy a piece of furniture you're doing 100 hours of research" ... "Imagine like when you're going to be able to Source..." - **Framing AI as an employee:** > "imagine I had like an employee who's like dedicated to working on this project what would I actually ask them to do" ## Communication Patterns - **Long unbroken monologues** with chained clauses and stacked connectives ("and and and," "so so so"), showing thinking-in-real-time: > "and and and because I would have to sit down at a laptop and I would spend like you know like half a day like looking like and then I would have to do that three or four different times" - **Self-correction and refinement mid-sentence**, e.g., the Broadway/Bowery correction: > "like Broadway and bowry are the same road ... basically the Indians like had that foot path" - **Walks through his screen** with the host, narrating his chats live: "so you can see," "let me just see what what did you ask." - **Concrete dollar and hour figures** to make abstract claims tangible: > "that's about $70 million in Prof profit margin over the last ... since the life of the company" > "you would have had to spend like 15 hours to find that" - **Closes thoughts with self-deprecating dismissals** rather than polished endings: "uh I don't know what whatever it would be but uh yeah anyway." ## Host's Delivery (for contrast) - Uses **summative check-ins**: "so what what I'm hearing I think so far is..." and "is is that a good summary." - Frequently **mirrors and validates**: "right right right," "yeah," "I love that." - Redirects with explicit transitions: "I want to go back to the thing you talked about," "La last last question I have." ## Uncertainty Notes - Timestamps in the transcript are coarse and fragmentary; several quoted passages appear split across utterance boundaries, so punctuation of quotes is approximate. - Whether patterns like "you know" and "like" reflect habitual fillers versus transcript artifacts cannot be fully separated — the density is high enough to suggest genuine verbal habit, but exact frequency should be treated as uncertain. # Observation Record: How Gumroad's CEO uses ChatGPT — Ep. 1 with Sahil Lavingia ## Speakers - **Sahil Lavingia** — founder/CEO of Gumroad, active investor, author of *The Minimalist Entrepreneur* (per host's introduction) - **Host (unnamed in transcript)** — interviewer, series creator ## Framing Thesis (Host) - The host states his theory that **"ChatGPT is the most important creative tool of the decade"** but most people don't know how to use it; the interview series is meant to showcase heavy users. ## Sahil Lavingia's Stated Positions ### Usage Frequency and Habit - Uses ChatGPT **~3–4 times a day**, primarily through the mobile app, which he says improves "almost weekly." - Uses it for "all sorts of things": deciding where to eat, weekend plans, networking, factchecking. ### Core Claims About ChatGPT's Value #### 1. No excuse for not knowing why - People often say "I don't understand why [X]" — he argues there's **"no excuse for not knowing why anymore"** because you can always ask. #### 2. Rapid idea filtering / killing bad ideas - He keeps essentially **no to-do list** (at most one item, e.g., "pick up groceries from Costco"). - Reasoning: most of his ideas are "terrible," and people **delude themselves** into thinking they have many good back-burner ideas. ChatGPT lets him research an idea quickly (on the subway, on a walk) and determine whether it's worth pursuing. - > "If it's a really good idea, just write it... tell me the exact scenes in the movie." - He'd rather **double down on things that work** (e.g., Gumroad's product-market fit). #### 3. Raised research quality bar - Example: buying a building in NYC — fact-checking lots, background checks on prior owners, usage history, zoning, visualization, pitch decks. - Claims the bar for pitch quality is now far higher than listings he sees on LoopNet ("seven photos taken with a Canon and some text"). - Predicts impact will show in the **top 1%** of movies, hotels, products — "so much more thought through." - Meetings: showing up informed (e.g., knowing what material **58 Bowery** was constructed of) signals seriousness to counterparties; the expertise he couldn't feasibly research for 50 buildings, he can do for one. #### 4. Thesis-driven investing via network effects - Frames high-risk investing as **power-law returns**: a thousand bets, one winner. - Applies this to real estate: of ~1,000 potential buildings over three years, one is "the building." - Strong believer in **network effects** — e.g., asks which roads existed first in downtown Brooklyn, hypothesizing the first roads remain the most popular. Fun fact he shares: Broadway and the Bowery are the same original road, following a Native American footpath through Manhattan. - Notes **Dumbo** was one of the best US real estate investments of the last ~100 years, tied to "iconic structures" like the Brooklyn and Manhattan Bridges. ### Techniques He Describes - **Iterate on bad answers**: the host highlights that when ChatGPT gave a "Wikipedia-esque" answer, Sahil told it to try again — "a lot of people just give up after first answer." - **Get it to "write a book in real time"** — a book that doesn't exist, tailored to your actual question. - **Poke holes in ideas**: tell ChatGPT your thesis and ask it to find holes; useful for accountability on whether an idea is good. - **Factchecking interesting facts**: example of a claim about Israel/Palestine he shared without verifying; couldn't confirm it was true — using ChatGPT to check in real time. - **Essay writing**: starts from a tweet/thesis (e.g., "one city per country"), asks it to "add three to four paragraphs to make the point compelling" and suggest examples. **If it can't suggest examples, "this rule doesn't actually generalize"** — a heuristic for discarding mental models. - **Prompt generation for DALL·E**: uploaded a photo of 58 Bowery's interior, asked for prompt ideas for a "social club for creatives," then fed those prompts back to refocus image generation on his specific vision. Notes the image input wasn't used much and hopes future versions will edit uploaded photos directly. - **Peripheral vision**: sees the innovation as expanding what you'd notice — analogous to how Kanye West finds samples. - **Employee framing**: "Imagine I had an employee dedicated to this project — what would I ask them to do?" Led him, after many days, to realize his real question: what renovations can he make to the building? Resulting thesis: **buy a historic landmark building and renovate it**. ChatGPT cited a specific regulation (Title 63) with URLs he still needs to read. ### Stated Limitations of AI (his qualifications) - Generated images **"aren't bound by reality"** — no true 3D; next step is modeling the building in 3D himself. - > "It can't actually write the code for you" — it gives a "rough thumbnail"; code is deterministic and data is insufficient. - Not very useful for **Gumroad's roadmap** because context lives in his head; better for **new things he's thinking about**, not things already in progress. - Twitter-format outputs are dated ("hasn't learned some stuff"); output still needs a couple hours of editing. ### Decision-Making and Organizational Claims - **Better decisions via fewer people deciding**: bad decisions come from collective decision-making ("the car looks stupid"); ChatGPT lets every person act like the PM. - Example: Gumroad's **flat pricing decision** took ~12 years; he speculates earlier ChatGPT-assisted pricing could have meant ~**$70 million** more lifetime profit margin. (He qualifies: "hard to say what would have happened.") - **Reduces emotional component** of decisions — though he notes ChatGPT "does say stuff like" cautioning against pricing changes (AI-safety-style objections). - **Exposes post-hoc rationalization**: people decide, then justify; ChatGPT can tell you your stated reasons don't align with your decision (e.g., "you should move to Dumbo" example). - **Custom instructions as a nudge** (host): host shares his own custom instruction that he's "too opportunistic, not strategic enough," with quarterly goals, so ChatGPT checks new ideas against goals. - **Smaller firms**: roles can be bundled — "if something that used to take 40 hours now takes five... you just don't need as many roles." Lawyers/accountants remain third-party. Host agrees: smaller teams going further. - **Team idea triage**: instead of deferring employees' ideas, "spend five hours with ChatGPT" to evaluate them — clears heads, fewer excuses. ### Broader Predictions - AI agents will source materials better/cheaper (vs. monopoly marketplaces like Alibaba Express); even a coffee shop could make "10x–100x better supplier decisions." - Apple keynote example: iPhone has ~6,000 parts; a trillion-dollar company researches every one — small players will get that capability. - **Tension he acknowledges**: "the inertia of the existing system is also mega massive" — a trillion dollars of value might be just a 1% GDP bump; also observes "life does seem to be pretty nice" already. - **Human-in-the-loop claim**: for extraordinary work, "human plus AI versus AI" — humans will always be non-zero contributors, though he concedes "maybe people will disagree with that." ## Host Contributions (positions) - Summarizes the tool as (a) a filter for bad ideas and (b) a research-quality amplifier — Sahil confirms with the "better first meetings with developer/construction manager/contractor" elaboration. - Emphasizes the value of **iterating past the first answer** and of ChatGPT **avoiding the blank page** while still leaving the human to edit ("you're not just typing a prompt and tweeting whatever it gives you") — Sahil agrees. - Host asks whether ChatGPT helped Sahil *form* his network-effects thesis; Sahil answers theses tend to appear spontaneously ("like Harry Potter"), but ChatGPT helps hold him accountable to whether it's good. ## Unresolved Tensions / Qualifications Preserved - Sahil both touts AI's transformative impact and concedes its limits (no 3D, can't write code, dated formats, weak for deep-context roadmap work). - The $70M counterfactual is explicitly hedged ("hard to say what would have happened"). - He claims AI removes emotion from decisions, then immediately notes AI sometimes raises objections like "changing pricing on users might be a bad idea." - The human-plus-AI claim is presented as his belief, with acknowledged possible disagreement. # Speaker Attribution Record ## Video - **Title:** How Gumroad's CEO uses ChatGPT - Ep. 1 with Sahil Lavingia - **ID:** 01a0f15b-42b1-75c2-818a-d8ba7f04567e ## Speakers & Roles ### Primary Identified Speakers - **Host / Series Narrator (unnamed in transcript)** - Delivers the series intro narration ("my guest on the show today is sahill...") - Conducts the interview; asks questions, summarizes, and prompts the guest throughout - Identifies himself as having known Sahil "about 10 years" - Mentions his own **custom instructions** in ChatGPT regarding being "too opportunistic" as a manager/leader — implies he leads a team (unspecified affiliation) ### Guest - **Sahil Lavingia** - Introduced by host as **founder and CEO of Gumroad**, **active investor**, and **author of *The Minimalist Entrepreneur*** - Primary speaker for most of the conversation; describes his ChatGPT workflows ### Narration - The series-opening voiceover (people "living in the future," purpose of the interview series) is attributed to the **host's narration**, not a separate narrator. ## Referenced Third Parties & Sources (as mentioned by speakers) - **Sam Altman** — guest recounts a joke from an interview about asking AI how OpenAI will make money - **Kanye West** — cited by guest as an analogy (sampling in music) - **OpenAI products** referenced: **ChatGPT** (app), **DALL·E**, custom instructions - **Gumroad** pricing decision (~$70M profit margin claim over company lifetime, per guest) - **LoopNet** — real estate listing site mentioned by guest - **58 Bowery** — specific building discussed; a ChatGPT answer on construction material attributed to "one article on the internet... in a PDF" - **"Title 63"** — a rule/citation ChatGPT reportedly surfaced with URLs regarding historic landmark renovation - **Broadway/Bowery**, **Dumbo**, **Chinatown**, **Brooklyn** — geographic/historical research topics via ChatGPT - Unverified fact about **Israel and Palestine** — guest says he could not confirm it after checking ## Promotional Segments - **None identified.** The opening narration promotes the interview series itself ("that's what this interview series is about"), which is program framing rather than third-party promotion. ## Comment Availability - **Not indicated** in the transcript; no curated metadata supplied. ## Uncertainty Notes - Host is **not named** in the transcript. - Guest's name appears transcribed phonetically in places ("sahill of India," "sahill"); standard spelling **Sahil Lavingia** comes from the title. - No timestamps present in the transcript. sometimes like you can say like I'm trying to buy a building and then I was like okay well let me like let me start visualizing like is this the right venue you're getting into write a book right in real time for you uh it's like a book that doesn't exist like this one city for country is like this essay that I was writing and so I started with this tweet which is like kind of like a thesis uh and then you know I just say like at three to four uh paragraphs to make the point compelling this is a thing too right sometimes it suggest examples I'm like okay this rule doesn't actually generalize I have this theory that chat gbt is the most important creative tool of the decade but it's still so early that most of us don't really know how to use it there are a few people though who are living in the future who have the time and the Curiosity to figure things out those people light the way for us they show us what our future with these tools will look like because they use them every day already that's what this interview series is about we'll go in depth with the most interesting people in the world so you can see every detail of how they use chat gbt to think write make art art make decisions build businesses one chat at a time my guest on the show today is sahill of India sahill is the founder and CEO of gumroad one of the largest platforms for creators to sell their work online he's also an active investor and the author of The minimalist entrepreneur I've known sahill for about 10 years and he's one of the most original thinkers I've ever encountered let's Dive In welcome sill tell tell us like how you use chat GPT yeah I mean I saw your Tweet uh initially about how I think you had a poll about how often you use it on a daily basis and I just looked at the app uh which is primarily where I use chat gpts through their their their uh app which has gotten it's sort of improves by Leaps and Bounds like almost it feels like on a weekly basis um and basically I was using it like three or four times a day for different things and I was just kind of like surprised by by that but I use it for yeah I use it for all sorts of things I think of it like there's just no excuse for not like I there's this there's this phrase that often comes up in conversation I really notice it now which is like I don't understand why and I often hear this right where it's like someone will be talking about some thing and say oh I don't understand why like Democrats X right or or Republicans do this or or like I don't understand why like Elon like doesn't get like or I don't get why like uh this and you know I just find that like I just don't I just refuse like that's no there's no excuse for not knowing why anymore right like you can always ask and so I find that like basically it increases like one I I spend a lot less time on bad ideas I think a lot of people what they do is they have like a to-do list of like things that they want to to do at some point I don't have a to-do list like I don't have any uh I mean maybe I do I could probably find something but it's like one item long and it's like right now it's probably says like buy a building like you know it's very like uh you know actually yeah right now I don't even have that uh you know just literally pick up some groceries from Costco but it's one item um and the and the reason is basically because when I have an idea I can very quickly invest you know the equivalent of what would have taken like months of of of time to research or weeks and you know and and because I would have to sit down at a laptop and I would spend like you know like half a day like looking like and then I would have to do that three or four different times and and now I can sort of just like as I'm on the subway or going for a walk in in in Brooklyn or something I can just ask it a bunch of stuff right CU queue up like all of these sort of like serial research processes and then often most of my ideas are terrible right so like the reason I don't have a to-do list is because I like don't have that many good ideas but I think a lot of people like they sort of delude themselves and thinking they have like all of these things and and they're like they're on the back burner right and they just have to figure out like how to make the time and this and this and like but the truth is like if you actually examine the idea like at a pretty decent level you realize it's actually not that good so I think that is one of the most useful things about it is it focuses your research on like what actually matters uh I think you know it's I think a lot of people have like in their Mind's Eye a really good idea right an essay or a movie or something but then when they like when the the truth is like it's actually not that good it just you think it you can just like sort of like pretend that it's good when it's not in like a concrete form right it's like if it's a really good idea like just write it like tell me the idea don't like tell me the log line for the movie like tell me the exact scenes in the movie right and I think uh Chad gbt it allows you to ask these kinds of questions and basically get to a zero or like is this worth pursuing sort of more um and I think this drastically changes like a lot of like the like even Investing For example like the bar that a found like a Founder can do so much research in a weekend right that like when they come to me like the like for example before now like right now for example I'm like buy I'm thinking about buying a building in New York City and that's that's a it's a thought that I have but the amount of sort of research that I can do fact checking on Lots uh background checks on previous Property Owners the history of of usage for certain buildings zoning requirements I can visualize what it looks like and it's just the like in you know the the sort of pitch deck that I would be able to create to you know to raise let's say any amount of money to do this like the bar for what that needs to be now is like it's literally like before like three years ago it would be like and I still get these by the way because you go to Loop net you can find like the pitches for these buildings right and it's just like seven photos taken with like a Canon and some text and like there like there's just no and I I I think the you won't see the impact in a lot of things but I think in the top 1% right top 1% of movies the top 1% of hotels like I built like the top 1% of all products will be so much more thought through uh so anyway that was kind of a rant but that that that think is like a big a huge one is I just find that like the research like I just don't have to I don't have to pretend that like I have like 50 ideas and I need to do them I can focus on like okay I have Road it has product Market fit you know uh that's really hard I've like I'm I'd rather double down on on on on the things that work um so I think I think that that that's been really really helpful so what what I'm hearing I think so far is um it's a tool for you when you have a a random interesting idea that you get really excited about to kind of go deeper on it to I guess filter out ideas that are actually not worth your time or are bad and then also like when you have something that you're pursuing seriously like buying a building it raises the level of quality of research and um I guess output that you can you can create because it it it puts all the things that you might need uh at your fingertips is is that a good summary of of of what it does for you yeah I think a lot of doing doing projects out like that have an impact in the real world are like they require having a lot of conversations with people and so if you can like bring such a like a better level of conversation to your first meeting with like a real estate developer and then your second me with a construction manager your third meeting with the general contractor like at each level you know you're you're having a a conversation with them you're learning a lot of things that you got wrong basically right and imagine if like you you didn't get those things wrong you actually brought them like a better plan in the first place they would still tell you a bunch of things you got wrong they would be different better more nuanced things right and so I think that also I think is like sort of a a value it's like I can show up to a meeting and and people are like how do you know so much about architecture it feel it like like you you've read like it feels like you've read too many books like not like like the like the and it's like Well Chad gbt just told me right like if you ask a very specific question as as long as I can like frame the question uh you know I can say like what material was 58 bowy constructed of uh and bring that up with with like the person selling the building and they're like that is really like there's only one article on the internet like it you you would have had to spend like 15 hours to find that right in a PDF somewhere and it's like well actually Chad gbt just told me that directly right right right but but the value is to them is like okay you're pretty this guy's serious yeah but but to them it's like this is the one building I've done this research forh because there's it's not possible to do this for like 50 buildings um but actually it is impossible to like you want you want to like put the level of work on you know in every in everything that you do and that and you need to to examine if it's actually worth pursuing like which one is is worthwhile right so imagine like if the equivalent of like every time you buy a piece of furniture you're doing 100 hours of research MH and that means like 100 hours of like actual research like you're you're going to make a lot less bad decisions right like right you're you're going to buy Le a lot less bad products if you were actually like doing the research um yeah and I think that that kind of like can extrapolate right like imagine like there like right now I don't think there's a YouTuber who says like I do a 100 to a thousand hours of research every time I buy like a chair or something you know but that could be possible like I think the like the the types of of content that you will be able to make because a lot of this like humans won't actually do but someone will do it right uh so you you can still benefit from like a lot of the value of like a better movie without having made the movie yourself right right um yeah well yeah I think this is really cool I think for the building the building example is one that I'm really interested in like take us through some specific chats that you've had so we can see how how you're doing it so it looks like you started on this dolly one we can start there or if you have other chats you you think are better to start with like I want to know okay what was in your head where you're like oh this is a great Chat gbt thing and then what what was your first message and and what was the outcome of the chat yeah I mean so I I think of I think of it like uh you know if you think about high-risk investing generally as you make a thousand bets and one of those is going to be uh you know your THX return uh ideally 10,000x return or something everything go zero you have like a 10x from right like that's sort of like the model is this like power law returns so I think similarly with with investing in like a a building in New York City for example is that let's let's say there's a thousand buildings you could potentially buy that will be listed and and on the market within within some constraints uh in the next three years one of them is the is the building right and everything else is gone um and so I think a lot of it is just figuring like do stuff like just narrowing down uh like for example I really believe strongly in the concept of like Network effects uh and and and investing in network effects doubling down on the ones that you have how hard they are to actually find and so you know things like asking like you know what uh what roads actually ex started to exist first in downtown Brooklyn right like my guess is that like the first roads are still the most popular roads like the network affected just keeps building even though people may forget uh you know like like basically for example like the first Road in New in New York City is basically uh Broadway like Broadway and bowry are the same road uh but that's like the before basically the Indians like had that foot path I went basically through the center of Manhattan um at a diagonal right um so anyway fun fact but it's you know being able to like actually ask and learn about all of the these things um without having to like traditionally I would go buy like a like a I would get a recommendation from somebody about like a 400 page Kindle book yeah and I would have to go read that and now I can just like ask it these questions right how did you start that chat history of downtown Brooklyn okay got it so you're you're sort of thinking to yourself okay I'm interested in downtown Brooklyn as an area I might want to buy um but before I do that I really care about the history I want to know what which roads came first I want to I want to understand that because because the past influences the present so strongly um but rather than like having to go read like a gigantic book or like do like a ton of Googling there's a you type a couple things into chat gbt here and it gives you mostly what you want exactly and I think here you know it gave me this like very like Wikipedia ask thing yeah I realized like no what I really want is the origin like the the real formation event of like you know the the region um and so you know you can modify and it's kind of like you're getting it to write a book yeah right in real time for you uh it's like a book that doesn't exist yeah but ideally you if you know what you if you're starting to narrow down like what you want what prom you're actually really trying to solve you can like get the perfect book written for you yeah like this could be have been you know much more specific if I if I was like by the way you know the reason here is to is to like buy building and so like focus on you know interesting structures that were the first structures that were built or something right um and then it would have focused on like well this steam fairy service like probably had a building associated with it right right and I mean think you're you're also doing something that I think is really kind of important it which is like um that I think a lot of people Miss is it gave you an answer originally that you didn't like that much and um and you just told it go again right um and I think a lot of people just give up after first answer rather than refining and being like hey like do it again in this specific way um I think that's really valuable yeah and it's often like kind of stochastic right where it's it does often like kind of like investing like nine times it might not be that interesting yeah but every once in a while it has like a really interesting like fun fact uh like for example like the region of Dumbo is one of the best investments right um in the real stay in the last in the US in the last like 100 years or so uh but if you think about like like even something like just asking at this more specific question if you really pay attention to what it said uh and you were able to like let's say there was a book that you were reading right like yeah 50 years ago about downtown Brooklyn and it said very specifically like iconic structures like the Brooklyn and Manhattan Bridges this is basically the reason that Dumbo is so successful is this and it literally uses the word I iic right so if if if I was even like it like it's it's like the answer is to making a THX return investment is there you just have to kind of see it and ask the right question and pull it out um and I think like being able to write better and better prompts like this yeah like I'm basically like and and sometimes it's funny like Sam Alman I think he he had an interview uh and he said someone asked him like how's opening i g to make money yeah and he basically said as a joke uh I think um we're just going to ask the AI like it'll tell us like the best way to make a money and we'll just do that right like yeah uh but it's it's sometimes like you can say like I'm trying to buy a building like sometimes people don't even vocalize what they're actually trying to do but if you just say like buy a building in and find network effects that they have kickstarted decades ago and it just yeah you start to learn like how to sort of focus the K of vision of of the AI I'm curious like yeah it's I guess it sounds to me like um uh once one of the things you're showing us is once you have a thesis it helps you do the research to like support or find Opportunities based on that thesis um but I'm also kind of curious um the extent to which it's helped you um identify or refine a thesis like you have this okay there's one in a thousand buildings um you know I'm I want to buy a building based on sort of the network effect of um how the area was established um did trbt play a role in that or where did that come from I mean it certainly can I mean th those I think often come like from that's kind of like the spark you know know like of just like it kind of just appears to you as like an idea um like Harry Potter or like you know at some point the idea I don't know where it comes from but I do think you can like hold yourself accountable to like is this actually a good idea or not you can tell it your idea and say poke holes in this and that has been quite useful too right like you can and this is less useful in maybe like the business stuff but I do find that like someone someone I remember someone told me some fact about about like Israel and Palestine or something and I was like oh that's crazy and then I told somebody else and then I was like wait a second I never actually checked if that was true I just thought it was like an interesting fact you know like the interestingness was why I was sharing it but then I was also like as part of that sharing like some numbers or whatever uh and then I like looked into it and it like I wasn't able to find that it was actually true but being able to just like in real time do that right like also I think uh is a way of like just saying hey I learned this fact is it true or is it not true and like explain why or you know like cite your sources like that that sort of stuff is is is quite helpful um even even like tweets I have ideas for thoughts I want to share and then I write an essay and I say like is this like what are the biggest holes in this essay yeah and sometimes honestly like sometimes it's like this is a big hole and I'm like oh yeah right I think of that like like like this one right like this one city for country is like this essay that I was writing and so I started with this tweet which is like kind of like a thesis uh and then you know I just say like add three to four uh paragraphs to make the point compelling also suggest more examp this is the thing too right it's like if it can sometimes it can't suggest examples I'm like okay there's there's no point to be made here then this rule doesn't actually generalize um but then you know it kind of like responds with this nice thing and then I can kind of like edit it and then at the end I can you know uh you know I can sort of decide if it's if it's worth worth doing or not I also think it's a way to like come at like a problem from multiple angles um right so if you have you can say you know like this idea isn't working but can you come up with different ideas yeah like I'm looking at Chinatown but am I missing are there places that are more populous right you know and then it sort of can expand like I think of a lot of the the the innovation in in this new wave of AI as like peripheral vision where it like because of the token the way it works and like you know basically each token in the in the context like influences the next thing not just like the the the most recent ones yeah um I I think that allows for this like interesting like level of peripheral vision where like you wouldn't actually see I you see this a lot in a in in dolly for example where I would ask something uh and then it it it it's actually quite interesting how it gener it shows you like the prompts that it generates um but like let me just see what what did you ask so interior of a social club inside bowy Bank Social Club um and and what was the context like why' you why' you ask that so you know you know this is sort of like Fast Forward imagine like I'm now refined my search to like uh you know I started getting into more specific areas and I eventually like found this like specific building I was looking even at like daily traffic so I got pretty close to like this uh so you you ran you you okay you narrowed it down to like a building that you really wanted to look at and then called 58 baly and yeah exactly and then I was like okay well let me like let me start visualizing like is this the right venue um like I basically I spent a lot of time on the exterior like the location like okay that's like part one of the thesis now part two is like is this the right building from the interior yeah like would it have the right Vibes Bic that makes sense and and what was the first so it sounds like you sent an image to chbt the bank building social club one where you were um asking yes okay so you so you you went in the building you took a tour it sounds like and then you took a picture of it and um and then asked Dolly to come up with or chat gbt to come up with um prompts for what it what it could look like exactly um I specified I don't know if it actually really used this image that much um I was hoping that it would it would like generate prompts that would like kind of follow it a bit more but yeah um but one day that that'll that that I think will happen where you'll be able to basically upload you know upload a photo and ask it to edit it um but basically yeah I just asked it to like give me some prompt generation ideas and then I took this because they are currently two separate models I basically just took all of these ideas yeah and then pasted it at the top of this or not why was that useful to you like why The Prompt generation ideas like what what you needed tell me about that I felt like it wasn't like I needed to basically refocus it on what the problem actually wash and the problem was basically I need a diversity of like prompts but they were all focused on this specific thing which is a social club for creat um in this in this specific area like to me this is the thesis everything else is a little bit arbitrary yeah um or you know like this this part um and so I basically just wanted to kind of like Pro like seat it with tokens that were like like interactive digital wall like interesting ideas that like the problem with this is this was basically like too broad and it would just diverge and then I wanted to kind of like refocus it back on like the the more specific Vision um yeah so I started just kind of asking it questions and uh getting it to generate like what I thought would would basically what I would like kind of use in a like I I would actually like go out and like hire an artist and like render stuff um but it's just you know just a way to get like really fast um ideas it's sort of like yeah like um it's sort of the first draft like really quick like as many different things as possible so you can narrow your search down into like a range of things that speak to you yeah and I think it helps it aids in if you need to communicate with another human being and you can't like project like I know like in my mind's eye like what it should look like but I can't communicate that with you right without having to spend 10,000 hours learning perspective and Anatomy yeah like I which I do like I've spent thousands of hours of learning to draw and like I my goal is like I would like to be able to do that too but it's just like Life's Too Short yeah where did you where did you get to like at the end of this like where did you end up with it what did it give you that you didn't have before for um I mean one it gave me confidence that this is still worth pursuing that like this it's sort of farther than the 50 other buildings that I've spent time on that sort of thing yeah and so you know I'm going to go like see it again and like spend more more like basically I'm gonna uh my next step is I'm going to get the basically model I have the floor plans so basically model the building in 3D and basically design you know like build basically design this whole this whatever this this this ends up looking like in in 3D space because I think the thing that currently AI like the leap it cannot make is these aren't Bound by reality in any way right right there's no 3D it's not like it's pixels it's not uh fion coordinate plane or whatever uh that's not the right thing but it's not XZ y right um and so I think you can Aid uh you know like you can Aid the artist with these materials but at the end of the day you still have to like build the thing perfectly I think of it like it can't actually write the code for you yeah right like at the end of the day it's not going to actually write the code for you it can kind of give you like the rough like thumbnail of the code it can often work code is like very deterministic um but there's not enough data at least right now for to be able to like build out like fully 3D you know what not um but yeah basically I would just like I would kind of like take this and like build it out and then I would uh it's just a way of you know at the end of the day the decision is like buying the building or not which it doesn't require AI at all right like um it's just like deciding to invest money um but I the quality of the decisions uh should be much much much better and I just look back at gum R and I'm like man if I was able to like there's just like you know like we updated the pricing the pricing was a better was so much better and it took like 12 years into the business and you know if we had hard to say what would have happened but let's say this existed and I was able to make like this different pricing decision because certainly I would have asked Chad gbg like these are the options like what should I what's recommended what are the other what have other people done what have pricing Chang in the past and how were they received blah blah blah all these sorts of things uh and let's say we got to 10% flat like much sooner that's those was about $70 million in Prof profit margin over the last uh you know since the life of the company basically uh that's a lot I mean that's but that's not like you know it's just a different decision yeah like it's a different number in one line the code yeah you um so that's why I think sometimes it's hard to like wrap your head and and you you won't see the impact of AI like you may never even talk to an AI right but like everything around you is going like the materials I see this uh in specifically in real estate development like talking to people the Imagine like when you're going to be able to Source like I think every material is going to be sourced from like a much better quality supplier for a much cheaper price because right now imagine you just like most people just go on to like the single Monopoly marketplace right and it's just like listed on Alibaba Express or whatever and they buy like a pallet of Terra Cotta or something uh but imagine if you could just like you know you had these AI agents that would like scour the internet uh right for revieww you know and like I I think the the bar for you know for for I mean generally you already see this right I mean like just think about like uh I think the last Apple keynot was shot on an iPhone yeah and the iPhone is like made of like something like 6,000 different parts so like imagine Apple like to make something that amazing like it it does have to like research every single you know like they do do that but imagine like that's a trillion dollar company um if like you're Joe Mo coffee shop could make like 10x 100x better uh supplier decisions on the chairs that you sit in on the you know auler on the coffee you know on the person making like the coffee pots yeah and I do yeah I don't know I I I walk around and like besides the news like life does seem to be pretty nice like pretty awesome so I think these this is happening you know um but I think also the inertia of the existing system is also Mega massive so yeah yeah you could you could generate like a trillion dollars in value and like you're still competing with like right trillion dollars in value so it's like a 1% bump in the GDP but that's like significant um uh I want to go back to the thing you talked about um a little bit earlier which is this decision to go flat um for for gumro for flat pricing um which it sounds like uh you know it's a huge sort of multi-million dollar decision for you over the last 10 years and and the idea that like if you had had Chachi BT like you might have made that decision a bit earlier how do you think about chat BT in your decisionmaking like when do you turn to it and what is the most useful way to um use it for that and if you have specific chats that you you've done that for I I know we've we've talked about the building but like if you have specific decisions that you've been making where you've gone back and forth what ch gbt it' be really interesting to talk about that um let me think I mean it's it's just like a constant hum of of stuff you know that I ask for uh if I'm like deciding where to eat you know I might like that's French toast place right or like if I'm factchecking something I might ask it something um if I'm looking for something to do on the weekend um if I'm looking for like you know people to like network with in in New York um all sorts of like small uh small things I mean I don't I don't think there's any um I mean the yeah I mean I I do think that there is like it's not super useful I think for gum because like the the road map now is like there's so much in my head that like typing it out to tell you know Chad gbt like give it the context it's just not worth it so I think it's much better at like new yeah new stuff um so yeah I think it's mostly mostly helpful like for like for for stuff that I'm thinking about not stuff that I'm already doing I see yeah that that's my that's that's my guess okay that makes sense I do use it in the gum context I do think it it ends up replacing a lot of like I think one one reason bad decisions are made is that you have a lot of people making the decision right so like the the decision making kind of get it's sort of like the designer makes the design decisions the engineer makes the engineering decision and the car looks stupid like just like it it has to be kind of this Collective thing um and I think you like this sort of empowers every single person to be like the PM for example yeah like so the decisions ideally like ideally for example the designer would often prior like we're building Flex out it's this like weird Equity Dividend complicated thing and before people would have to H me with like how does Equity work stock options 49a exercise stri like all of these things uh and that's every single person to explain it as a to get it as like a as a customer of the product but then also like if you're imagine you're designing it right have all these questions too and now you can just ask chat GPT and it'll just tell you and so I think that also I think the quality of the decision is not just by the co of the company but by everybody helps a lot it also means you need less people so the pricing decision I think for example the more people you have you have you're going to have people who are like emotional and like defend the you know like oh no it's too expensive or and like you know at the end of the day some decisions are hard to make that's like the role of the CEO sometimes is like to make the hard decisions it's not necessarily like like you know surprising that it worked to a lot of people it's just like yeah oh you did it wow that's hard um you know you got to deal with some like hate on Twitter or something right um so I think that also like if you are relying on Chad GPT you get rid of the emotional component right like the people saying like oh no like that's a bad like I mean that might maybe that still happens right like every once in a while AI safety like it does like kind of say like changing the pricing on your users might be a bad idea you know like it does say stuff like that uh but you know it's better it feels more objective that's the other nice thing is like I I love it for factchecking my own biases because it tells it like will tell me like you know it'll it'll like for example if I say like uh I have like some mental model in my head and I but I always use like one example right uh to to go to like explain the mental model like one city per country in like New York City for a us or something and then if I but if I go to Chad gbt and I say tell me five you know like I just force it to to improv and it only comes up with like the one example I have I'm like okay there's no good other examples like I should stop using this mental model I need to like discard it in my you know decision- making right um so I think that that yeah that that has helped uh I think a lot of people like they they want to make a decision uh and then they justify it to themselves why they're making decision and it at least tells you like hey you can still make that decision you want to make but that's not the reason you're making it like it's yeah you're not making it because of this rational like set of constraints you can codify in ashad GP and it'll tell you like these don't align with your decision yeah it'll tell you like you know you should move to Dumbo because you're closest am in hand what you say you care about is yeah to you know commute to vand or something I I have that too I mean one of the things I feel for myself is like as a manager as a leader of every um I'm just a little too opportunistic like I'll get really excited about a new idea and I'll like kind of want to go in that direction um and I need to be a little bit more sort of strategic where I'm just like okay like here are like quarterly goals like let's make sure everything we do is like aligning with those goals and I literally just have a thing in my custom instructions that's like right now one of the things I'm working on is I'm a little bit too opportunistic I'm not being strategic enough here are our quarterly goals and so whenever I get excited about an idea um I'll I I'll always chat with Chachi BT because I think it's really interesting and it's always like hey like does this align with your quarterly goal and I'm like damn it you're so right like it really helps me helps keep me on track um in a in a sort of and just that slight little nudge does does a lot for me yeah it's almost like a like a screen time reminder in a way yeah where it's like by the way just reminding you you're you have one core goal this quarter yeah uh yeah it's kind that's kind of like just reminder you're trying to buy a building like does with that goal or not I love that yeah yeah the other the other nice thing just to mention quickly is like you can Empower your like certain people like let's say you have someone on your team who's like oh I really want to like explore this thing uh sometimes you just have to like right like in in the prior world you just say like oh let's like wait like we don't have time or like let's look at it that like next quarter or something uh because it would take like a long a lot of sort of time investment to figure out if it were actually worth doing or not but now if it takes like I can just say hey like spend five hours this quarter like just literally spend five hours Hing to chat GPT and like pres you know whatever like explore this problem as much as you can and then the goal for the quarter will be just to evaluate is this worth do and then so sometimes you it's nice because it clears everybody else's head too of the of these ideas that they may be like sitting on and like oh we should do this and that it's like oh and it's like no just like spend two hours realize it's not that good of an idea or or not and then you know uh pursue it uh or not so I I find that also is like really like people have less excuses right generally across the company uh when they have this access to this sort of this tool um and and I I think like the like gumad has like product designers software Engineers customer support people uh who also write all the help center copy and like that's like I feel like and like fraud and risk review and that to me seems like why like every other role is like Chad GPT like you could just tell an engineer to use chat GP to get the answer that they would need or you we have lawyers and accountants but they're both third party um yeah I don't know it feels like it just really like you you you you don't you just don't need as many roles right if something that used to take 40 hours a week now takes five hours a week you can kind of bundle all these roles into one person often the CEO or the founding team or whatever will just do that yeah um so yeah I think there's a lot of implications for like the how these how companies like the size of the firm and like how stuff like that yeah I love that I mean I I do think for every like we are able to uh go further with a smaller team than we would than we would have before and and that feels just really really exciting to me um so um yeah I think uh La last last question I have um before before we go like tell me more about the writing stuff I know you've been you showed us a little bit of writing stuff if you have any more examples of how you're using it for writing I think that would be really interesting to um to go through um um sure yeah I can see if I can yeah I mean so this is like one idea is I was trying to figure out like can I get if I can really provide a really precise thumbnail of what I'm trying to write yeah exactly like the the bits that you know that like even I was thinking of saying like use these three sentences and like you know so it provides them um but anyway I have the this idea that like there's sort of people who think in terms of logic and people who think in terms of rhetoric um and the the example I wanted to use so I would prob say use this example specifically logicians like uh like this uh doing good Rec what reply so it's kind of like basically like the way that you interpret the sentence um like if you think like like what how would an AI model interpret the sentence right like a logic someone who's like really logic based would be like okay clearly this is just like some weird thing yeah but you know I don't know someone who's like maybe a little bit more like caught up in the thing would be like what the hell like you know something like that um so anyway this is like I think a good example and I think it wrote something it's funny because uh the format of Twitter has changed so much that it hasn't learned some stuff here but and then you know I was like tur into a pay like right but I I attempted it um you know so this is just an example of like you know is one better than the other um trying to have a conversation with it yeah so you have a little concept and you're sort of using this to help you refine and pull out like different ways to talk about the concept exactly yeah like just yeah trying to spark sort of spark like basically I'm trying to figure out are there like three or four interesting little bits right that that demand like a little mini uh post on on X instead of just like a single tweet right um often often like specific examp examples are like really what I'm looking for because I find that like you know it's kind of like a I think of it like listening to like a really good Kanye West song like to me like he's a good he's like good at making music but what's really interesting to me is how like he finds like crazy samples um and and like he just kind of highlights them and so I think of it similarly like I'm like trying to use Chad GP to be like can I find a really good example that explain this topic that like you know it's kind of like you read a book and everyone sites like the same examples for stuff it's just like in the Z guys like oh yeah 10,000 hours like the example is like yeah uh I don't know what whatever it would be but uh yeah anyway trying try like trying to trying to find like the the thing that people site actually like that becomes the thing people site yeah so yeah anyway this is like what it would I I would like it's certainly not good enough that I would I would definitely want to like spend like a couple hours like editing down but my style of writing is really simple which is like I just edit down words like I just have that thing that's kind of like contains all of the ideas I want and then I just spend like two hours like do I need this word do I need this word and like right eventually it gets down to something small and then you know and and so like the ideas may already all be here and then it's just a question of like you know trying to get trying to and I wonder even if like you could say something like three the above with 90% of the words and like you know kind of recurse this right like keep asking it um until it has like you know as few looking sounds like yeah it sounds like it's giving you sort of like it avoids the blank page it gives you some raw material to work with and then you can rework what's there into something that feels like you and has new ideas that you might not have run into before or new examples that you might not have run into before and um and that just that just makes the whole process a little faster and a little bit cleaner but you're not like just typing a prompt and then tweeting whatever chbt gives you yeah exactly yeah like there's always not that there's always going to be a need for that but if you're trying to produce extraordinary stuff yeah you're always going to have a human in the loop right because at the end of the day it's human versus a it's like human plus AI versus AI yeah so like humans will always have some that they can add on top of Ai No matter how good AI gets like humans will be non zero I think um maybe people will disagree with that but um yeah I think it's it's useful like here's a here's an example of this kind of thing right where I had an idea I had like one example and I'm like please come up with more and then it yeah it didn't really you know come up with anything you know that uh that compelling to me so I didn't really Double Down On It uh or this one's another one oh yeah this is another version of that I really didn't give up it I guess U this is more great a good a good this is kind of like another framing I often use is like imagine I had like an employee who's like dedicated to working on this project what would I actually ask them to do right like what's the actual like when you're doing it yourself it's easy to like right right but if you had like on payroll uh like what would they what would they act you know so I would basically ask them like is this you know what Renovations can I make to the building like this would determine if it's worth bu like this and it took me a while like many days to realize like this is really this is the question I really need to to be able to answer that you know um at least in the context of of this specific building but generally this is what I realize is I need to buy a Historic Landmark building and renovate it um yeah that's at least my current thesis so I love it it and it it it actually tells you it even like cited the uh the URLs and stuff so I need to go you know like read this but it's like it literally tells me exactly it's this one rule title 63 right and that would take like lots of lots of research or talking to people who are expensive or hard to get to to like find ordinarily yeah or just the possibility that it might right yeah like it could have been a Google search away but it could have also been like because you know it's like some government document like you don't like you don't know and a lot of people I think they evaluate like the effort and then they just say it's not worth pursuing it's like the chance of this being a worthwhile project is so low yeah but if you can tweak that risk reward ratio maybe you can get it to a place where you're like oh actually it does make sense for me to spend like an hour like yeah reading into this um and then basically you know at some point I would like hire someone you know there there's people that you hire that are basically like a they just visit the site and then they you know they do all the research and they give you a report that's like this is what you can do uh but like they are happy to take your money to answer all the basic questions right like it's I want that piece of work to be as high quality as possible I also think it it holds me accountable to like do I really want to do this or not like it's easy to just spend 10 hours like screwing around but if I'm trying to like am I getting somewhere like is the quality of the idea improving right or am I just like spinning my wheels and I think that's another signal too yeah this is great I've learned a lot um I really appreciate you taking the time to do it um yeah you you have so many so many good use cases for this um thanks for thanks for sharing with us you're welcome thanks for letting me do it hopefully it was helpful it was great cool man