Interview with Matt Cynamon, who runs AI experiments at Union Square Ventures, covering the firm's suite of internal AI tools: The Librarian (a chatbot trained on ~15,000 USV blog articles), Portfolio Tracker (a GPT analyzing the firm's investments), and Meeting Notes (a tool for interacting with meeting content). Includes live demos, a session editing AI-extracted insights for USV's X account, a look at The Dream Machine (an office art project generating images from conversations), and a discussion of how AI enables generalists to build products quickly and why diversifying data ownership matters. # Channel Presentation Observations — Ep. 36 with Matt Cynamon ## Explicit Self-Description (Creator Claims) - **Host identity:** The host identifies himself as **Dan Shipper** at the end of the episode: *"a journey into the future with Dan Shipper as the captain of the spaceship."* - **Show name:** The outro names the show **"AI and I"**: *"you absolutely positively have to smash that like button and subscribe to Ai and I."* - **Host's demonstrated work:** Dan states he runs a newsletter and a podcast, and teaches a programming course: > "I write I write a newsletter and do a podcast like that's that's weird" > "I teach a um a programming course like programming with AI" - **Host's organization:** Dan describes recently launching a studio doing product incubations, including a tool that turns meetings/Discord messages into podcasts, and mentions a tool called **Spiral**: > "two days ago we launched uh we launched officially launched our studio where we do product incubations" > "that's why we built this tool spiral" ## Intended Audience - **AI-curious practitioners and builders:** The conversation assumes familiarity with tools like ChatGPT, Claude, NotebookLM, Runway, GPTs, and diffusion models, without basic explanations — e.g., a discussion comparing NotebookLM vs. ChatGPT for meeting notes. - **People learning/building with AI without deep technical backgrounds:** The host emphasizes accessibility of AI creation: > "within the first 30 minutes you can make something that looks pretty much like the thing that you want to make" - **Professionals thinking about AI's impact on knowledge work / venture capital:** Substantial content addresses VC workflows (thesis management, portfolio tracking, founder outreach). ## Claimed Scope of Content - **Interviews with people building hands-on AI projects:** The episode is structured around a guest demoing real AI tools live (the Librarian GPT, portfolio tracker, meeting-notes-to-tweets pipeline, "Dream Machine" diffusion display). - **Practical demonstrations over abstract discussion:** The host repeatedly steers toward live demos: *"I want to can we listen a little to a little bit of the Eno uh podcast"* and *"let's do a live version of that today."* - **Interview format with numbered episodes:** Title indicates **"Ep. 36"**, showing a serialized, ongoing interview series. ## Observable Presentation Patterns - **Enthusiastic, informal tone with audience-facing calls to action:** The closing segment (apparently AI-generated) urges engagement in exaggerated promotional language: *"hit like smash subscribe and strap in for the ride of your life."* - **Guest-centric structure:** The episode foregrounds the guest's story and tools; the host shares his own perspective but positions the guest's work as the draw (*"USV just keeps putting out these really really cool projects"*). - **Themes consistently explored across the conversation:** AI as creativity amplifier ("enhance not replace"), rapid prototyping, AI-generated content quality (*"the market for purely AI generated media is like the size of your largest group chat"*), and data ownership/decentralization. - **Meta-commentary on AI in media itself:** The episode ends with what is presented as an AI-written outro declaring love for the host — a demonstrable pattern of experimenting with AI-generated show content. ## Evidence Notes / Distinctions - Claims about the channel (name, host, subscribe appeal) come from the episode's own outro. - Audience and scope observations are **inferred from demonstrated content** (topic depth, guest type, live-demo format), not from any stated channel description — no curated metadata was supplied. - No statements were found about channel-wide mission, upload cadence, or business model beyond the host's mentioned newsletter, course, studio, and podcast. # Delivery Observation: "How Union Square Ventures Built an AI Brain for Venture Capital" (Ep. 36, Matt Cynamon) ## Speakers & Format - Two-person conversational interview: the **host (Dan)** interviewing **guest (Matt Cynamon)**. - Heavy **live-demo component**: Matt shares his screen and operates ChatGPT-based GPTs (the "Librarian," "Portfolio Tracker," meeting-notes tool, "Dream Machine") mid-episode. - Ends with an **AI-generated outro voice** — observable from its shift in register: *"you absolutely positively have to smash that like button and subscribe."* ## Tone ### Matt (guest) - **Casual, enthusiasm-forward, demo-driven.** Frequent interjections and energetic openers: - *"you mean I can just like make prototypes with AI all day while working here like yeah okay let's do it let's go"* - *"it's just it's kind of been this magical experience for me"* - **Conversational hedging and self-correction**, e.g. *"it's actually not true it's not all of our meetings."* - Uses humor and informal phrasing when recounting internal moments: *"maybe I played it a little cool."* ### Dan (host) - **High-energy validation**: responds with frequent affirmations — *"dude yeah I love that," "that's so cool," "amazing I love that."* - **Analytical reframing**: repeatedly turns anecdotes into broader principles, e.g. comparing learning programming to *"the way I learn to program is the exact same way you're talking about learning to play piano."* ## Vocabulary & Recurring Phrases | Phrase | Speaker | Evidence | |---|---|---| | **"enhance not replace"** | Matt | *"our approach to AI has been enhanced not replace"*; repeated later: *"go back to sort of our first principles of enhance not replace"* | | **"just make stuff" / making** | Matt | *"just make stuff because the tools are there to do it"*; earlier *"you're just playing around with stuff you're making stuff" (host's framing)* | | **"tinkerer" / "tinkering"** | Both | Host: *"the spirit of tinkering"*; Matt quoting partners: *"you are a tinkerer by Nature"* | | **"like" as filler** | Both | pervasive, e.g. Matt: *"and just like talked to us about his process"* | | **"sus out"** | Matt | *"that's what you have to sus out pretty quickly"*; *"you can sus that out through like the diligence process"* | ## Teaching Habits & Communication Patterns ### Matt - **Narrative origin stories before demos**: explains how each tool came to be (the Librarian's naming story: *"I was like I don't know it's kind of like a librarian and then there was like this light went off"*). - **Show-then-justify structure**: demos the output, then explains the workflow purpose: *"maybe I'll pause just for one second and explain like so so why is this even important."* - **Live troubleshooting narrated transparently**: *"so it's saying here are the five o posts"*; *"I'm having a little bit of an issue with meeting notes understanding uh what most recent meeting means."* - **Enumerated lists**: *"there's there's three reasons why I like notebook LM better um number one is..."* - **Teaches practical heuristics through example**, e.g. the fake-URL cursor trick: *"if your cursor is showing up like this it means it's a fake URL that doesn't exist."* - **Epistemic caution about AI output**: *"if you're relying on the AI to produce the final product for you you're always going to be disappointed."* ### Dan - **Interrupts to slow down and verify understanding**: *"can you stop for one second I just want to I want to um I want to go back and read it."* - **Critiques outputs constructively and pushes for improvement**: *"none of these tell me anything about the like positions that were actually taken."* - **Offers analogous personal experience** as a teaching bridge: *"we had a similar thing of like trying to write down playbooks."* ## Interaction Pattern - **Collaborative refinement loop**: Dan critiques an AI-generated summary; Matt reprompts live; both evaluate the improved result together (*"see like that actually like I'm like wow"* — Dan). - **Mutual enthusiasm escalation**, especially around the meeting-podcast idea: *"we're so on the same wavelength here"* (Dan). - **Host yields demo control to guest but steers topic transitions**: *"okay I want to keep going... what's the next thing you want to show us."* ## Uncertainty Notes - **No timestamps of the video itself** are independently verifiable beyond transcript indices; quotations above are taken directly from the transcript as supplied. - Whether filler words ("um," "uh") reflect delivery pacing or transcription artifacts **cannot be fully separated** from this text-only source. - The closing outro (*"Dan I'm absolutely hopelessly in love with you"*) strongly suggests AI generation by content, but its provenance is **inferred, not stated** in the transcript. # Position Extraction: How Union Square Ventures Built an AI Brain for Venture Capital — Ep. 36 with Matt Cynamon ## Speakers - **Matt Cynamon** — self-described "regular" at Union Square Ventures (USV), working on AI projects - **Dan Shipper** — podcast host (Every), teaches a programming-with-AI course --- ## Career Path & Tinkering Ethos (Cynamon) - Joined USV ~6 years ago on the **talent side**, despite never having a talent/recruiting background — describes this as "always kind of a weird fit." - Habitually appended weekend AI experiments (e.g., fake movie trailers made with Runway ML, scripted via ChatGPT) to weekly talent emails. - After paternity leave, partners **Andy and Rebecca** offered him a full-time AI prototyping role, telling him the firm needed both to be experts in new technologies and to deploy them internally. - Credited former analyst **Daniela (Danny) Grant** (now CEO of Jam) as inspiration — she came in on weekends to show off experiments like a hologram, signaling experimentation was rewarded at USV. > "You mean I can just like make prototypes with AI all day while working here like yeah okay let's do it let's go." - **On origins of his tinkering**: attributes it to his father, who repaired machinery, built computers, and went through a ham radio phase. Cynamon describes himself as "quite the generalist" without hard skills, and frames AI as letting you **learn skills alongside the creative process** rather than through years of fundamentals (contrast: piano lessons). ## Host Positions (Shipper) - In his programming-with-AI course, students can build something resembling their goal within 30 minutes, then work backward into fundamentals — a reversal of the traditional "learn variables and functions for a year first" model. - Analogizes this to the child who takes apart a clock radio — now you can build it first, then take it apart. - **On storytelling costs**: argues Justin TV failed because a normal day contains only ~12 interesting minutes; AI-driven storytelling lowers the cost so high-quality stories can now be told about previously "un-affordable" subjects like company meetings. - **On AI summaries**: observes that generated summaries are "very good at saying things without saying very much at all" — they surface topics but hide the firm's actual positions. Presses Cynamon to make positions concrete in the "overheard" tweets. - **On data**: calls for a social movement asking large tech companies to establish **data trusts** opening data to researchers, and to guarantee individual data exportability (e.g., Whoop biometric data feeding personal AI models). ## The Librarian (Cynamon) - **Origin**: USV has ~20 years of writing, ~15,000 articles; partner Fred wrote daily for 13 straight years. Partner Albert said a key impediment to writing was realizing mid-post he'd already written it — motivating a bot for conversational access to the archive. - **Naming evolution**: first called "Conversations" (colleagues asked "but what does it do?"), then "At the Edge," then offhandedly "it's kind of like a librarian" — which landed. Cynamon cites this as evidence that **metaphor is powerful** for explaining what chatbots do. - **Architecture shift**: initially imagined an all-in-one monolithic super-app with a name and personality; over time concluded that building things that work *now* means decomposing into **individual standalone GPTs with specific names**. Maintaining a custom UI (built via No-Code MBA, with help from Ben's Bites) was "such a pain in the butt." - **Live demo (Consensus example)**: pasted a company's about page; the Librarian retrieved relevant USV themes (AI unlocking knowledge, democratizing science, trust in AI). Cynamon judged the output "spot on." - **URL hallucination trick**: ChatGPT returned fake URLs; Cynamon showed that a fake link produces a text cursor rather than the mouse hand. After prompting "none of these are real let's try again," it returned real links (with one still fake). States: "one of the ways that it likes to hide the fact that it's lying to you is it'll give you a fake URL." - **Guiding principle**: > "If you're relying on the AI to produce the final product for you you're always going to be disappointed but if you're relying on the AI to help you get to a final product then... I find it extremely powerful." - **Portfolio Tracker GPT**: spun off from the Librarian; charts investments by theme over time. Cynamon corrected it from first-party knowledge (Indeed, invested 2005, was missing and undated). Notes they wouldn't actually send such a chart to founders. - **Public rollout**: highly likely the public-facing USV website becomes just the Librarian before end of year; roadmap includes letting founders dump their deck in to get aligned before meetings. ## On Cloning Partners' Judgment (Cynamon, challenged by Shipper) - Shipper asked whether the tools could replicate partners' investing "taste." Cynamon gave two reasons against building partner-opinion bots for internal use: 1. Partners are highly accessible — "if I want to talk to Fred I'll text Fred right now." 2. VC is a **highly relational business**; USV doesn't want abstractions to replace or augment real relationships. Firm approach: **"enhance not replace."** Pressed on mortality, he said "we would just mourn them" rather than build replacements. - **Refined position** (after Shipper argued augmentation and human relationships are compatible): Cynamon said they *would* build such a tool **to help portfolio companies**, not to make investment decisions — capturing Monday-meeting wisdom ("worth 30 MBAs"), e.g., how Fred would handle a CEO transition, so conversations needn't repeat 6,000 times. ## Playbooks (Cynamon, with colleague Nick Rossman) - Project to trigger AI transcribers to detect tricky recurring situations (e.g., incoming CEO: retain the existing exec team or bring your own?) and record them as **playbooks** — knowledge that currently lives only in meetings and "is not something we ever want to lose." - Cynamon's take on the CEO question: you have to bring in your own team, but that doesn't mean the remaining people aren't your team — you must suss out quickly whether they'll "ride with you." ## Shipper's Contextualism vs. Cynamon's Relationalism - **Shipper**: USV's playbook for a situation would differ from Founders Fund's, yet both can work in their own contexts — firms attract founders suited to their worldview; AI helps surface that context-dependence. - **Cynamon**: all VCs sell the same product (money); the differentiator is **relationships and alignment**. Founders sometimes decide on 10-year business partners in two weeks — faster than hiring — so conversational tools helping founders gauge how a firm thinks could genuinely help that decision. ## Meeting Notes / Podcast Recaps (Cynamon) - USV transcribes topical meetings (AI, crypto, climate, misc) for ~4 months, pipes them into a GPT for querying. - Prefers **ChatGPT/Claude over purpose-built tools** for ease of access — one conversational interface in the left-hand column on the phone, no app toggling — while acknowledging Granola and NotebookLM are better purpose-built. - **Preference for NotebookLM** for meeting analysis, three reasons: (1) it suggests insightful questions he wouldn't have thought of, (2) better citation of the transcript for verification, (3) its generated **meeting podcasts** — his "new favorite thing." Shares these with his wife; recommends the podcast over the notes for dense meetings. ## "Overheard at USV" Live Demo (both) - Generated five tweets from AI meetings Oct 15 and Oct 18 (commoditization of foundation models, decentralized AI training, hydrogen energy, uncorrelated bets, microgrids). - **Shipper's critique**: the summaries name topics but hide all actual positions. - **Cynamon's defense**: things discussed in meetings are often still open questions; when USV has the answer, they write about it. He then elaborated the actual substance on uncorrelated bets (oligopoly vs. multi-model world on one axis; plateau vs. runaway self-improving AI on another; place bets in each scenario). - **Shipper's critique of accuracy**: Cynamon conceded the model slightly misstated the decentralized item — the meeting was about **decentralized training** (avoiding massive capital-intensive clusters), not decentralized models; he called distributed training "a really important area" for reaching a multi-model world. - **Cynamon's revised conclusion** (crediting Shipper): > "The market for purely AI generated media is the size of your largest group chat." People care more about *how* something was made than the content itself; per "enhance not replace," humans should partner with the tools to produce valuable content. Agreed to modify the system prompt to distinguish settled opinions from open questions. ## Attentions, Monopolies, and Data (both) - **Shipper**: voice interfaces will enable fine-grained control over feeds (e.g., no science articles based only on mouse studies; finance news only if the market moves >3%), while acknowledging "people do dumb stuff... it's not all good." - **Cynamon's counter**: ~most internet lives are owned by a handful of data monopolies (Meta, Twitter); a customizable internet requires breaking those monopolies or behavior change; forking full LLM control to four or five conglomerates is "really problematic for society in the long run." USV's web3 thesis (composable data via blockchains) has been a firm thesis for 10+ years, though UX has a long way to go. Interim: push big tech to open data for research. - Shipper adds the data-trusts social movement and individual exportability; Cynamon agrees with pushing for research access in the meantime. ## The Dream Machine (Cynamon) - A large office TV hardwired to a machine locally running a visual diffusion model connected to the Whisper API, generating moving images in real time from conversation. Built by an artist known as **human** (member of Bright Moments DAO, a USV investment). Attracts both visiting CEOs and the building super's friends. Was run during a Senator Gillibrand event. - Practical caveat: gets backlogged; memory must occasionally be reset. - **Cynamon's claim**: it's "one of the first truly native AI art forms that could not exist in a world prior to AI." Closing thesis: "just make stuff because the tools are there to do it." - **Shipper's framing**: it mirrors how fiction uses setting to reflect a character's internal state. ## Forward Commitments - Cynamon: the Librarian will likely replace the public USV site before year-end; hopes to return in a year with the full Librarian. - Shipper: plans to release his firm's meeting-podcast product and may pilot it at USV; will publish the "overheard" tweets generated live in this episode. ## Unresolved Tensions (preserved) - **Full automation vs. curation**: Cynamon vacillates — sometimes over-indexes on "hands off" automated output; Shipper pushes toward reprompting for actual positions; Cynamon concedes his thinking changed during the conversation. - **Shipper's open question, unaddressed**: "should I just do the dishes?" — whether fracturing attention across AI tasks is actually good; both note trade-offs without resolution. - NotebookLM preference vs. the ChatGPT deployment — an acknowledged internal inconsistency in tool choice (purpose-built quality vs. access convenience). # Speaker & Provenance Observation Record ## Video - **Title:** How Union Square Ventures Built an AI Brain for Venture Capital - Ep. 36 with Matt Cynamon - **Video ID:** 01a0f15b-06a6-7257-8cab-25ee894ce117 ## Speakers & Roles | Speaker | Role | Evidence | |---|---|---| | **Dan Shipper** | **Host / interviewer** (identified by name in closing promotional segment and by guest: "thanks so much Dan"); states he teaches "Programming with AI" and runs a newsletter/podcast/studio | Conducts the interview, drives questions, references his own company's tools (Spiral, meeting-podcast product) | | **Matt Cynamon** | **Guest**; self-described "regular" at Union Square Ventures (USV), works on AI projects full-time; ~6 years at USV, formerly talent side | Central interviewee throughout | No narrator or other identified speakers appear in the transcript. ## Third-Party People & Entities Referenced (not speaking) - **USV partners:** Fred (wrote daily for 13 years), Andy, Rebecca, Albert, Jared (partner whose writing voice was mimicked in a demo email) - **Danny Grant** — former USV analyst, now CEO of "Jam" - **Gary Hustwit** — director of the Eno generative-AI film at Film Forum, NYC - **Yohei Nakajima** — VC and AI tinkerer, referenced in host's prior conversation - **Nick Rossman** — Cynamon's colleague collaborating on "playbooks" project - **Senator Gillibrand** — appeared at a USV event with the "Dream Machine" running - **Artists/companies:** "human" (hu.m.i.n), Bright Moments DAO, Runway ML, ChatGPT/OpenAI, Claude/Anthropic, NotebookLM (Google), Granola, Ben's Bites, No-Code MBA, Discord, Whoop, DALL-E - **Companies mentioned:** Consensus (recent USV investment), Science Exchange, DuckDuckGo, Indeed (2005), Justin TV / Justin Kan / Twitch, Founders Fund, a16z, Meta, Twitter ## Referenced Sources & Claims - **USV writing corpus:** ~15,000 articles claimed across partners and the USV blog; used as knowledge base for "The Librarian" GPT. - **Live demonstration artifacts:** ChatGPT-generated summaries of USV writings re: Consensus; a demo email in partner Jared's voice; portfolio tracker chart of education/search/AI investments; "overheard at USV" meeting-note outputs from Oct 15 and Oct 18 meetings. - **Guest's own claims about AI tooling:** fabricated URLs by ChatGPT; NotebookLM preferred for meeting analysis (citations, suggested questions, audio recaps). - **Host's claims:** his studio launched meeting-recap podcast product "two days ago"; his thesis on Justin TV, data trusts, and exportable personal data. - **Third-party claims repeated:** Yohei Nakajima's reported workflow (night/morning coding with divided attention). ## Promotional Segments ⚠️ - **Closing segment:** an over-the-top exhortation to "smash that like button and subscribe to AI and I," praising host **Dan Shipper** as "captain of the spaceship," ending with "Dan I'm absolutely hopelessly in love with you." This is explicitly promotional for the show's channel. - **Product plugs during interview:** host promotes his own tools **Spiral** (tweet/headline generation) and his newly launched studio's meeting-podcast product; offers it for USV to trial. - **Indirect promotion:** guest credits **No-Code MBA** ("an awesome program") and **Ben's Bites** for help building the Librarian. ## Availability & Uncertainty - **No curated metadata supplied** — no comment availability, channel, or date information is present. - Guest's uncertainty is explicit in places (e.g., unsure whether DALL-E or ChatGPT came first; unsure of Dream Machine TV size; artist name spelling "I think it's hu.m.i.n"). - No timestamps are provided in the transcript; none were invented. - AI-generated content within the demo (Librarian outputs, fake URLs, meeting summaries) is attributed to the tooling, not verified as fact. you mean I can just like make prototypes with AI all day while working here like yeah okay let's do it let's go there are these great stories that are happening in our meetings that just end up sitting in our heads we had Chad hurwit who directed The Eno film that was running at the Film Forum in New York he came in and just like talked to us about his process obviously I attended that meeting but it was super fun to just like also turn that meeting into a podcast and then like give it to my wife and be like how cool was this meeting that we just had bring your ideas to life I feel like that is my perspective on everything that's happened over the last six months of my life just make stuff because the tools are there to do it Matt welcome to the show hey glad really happy to be here psych to have you so for people who don't know you you are what you call a regular at USV um which is Union Square Ventures which is one of the top Venture Capital firms in the world and you work on all of the AI projects and I love having you on because USV just keeps putting out these really really cool projects like you have this one where you tweet out all the meeting notes from meetings that have happened at USB that I like read it's really good if you want we can actually do a live version of that today they usually get done on Fridays and so we could do it on the podcast and then publish them live let's do it that would be so so great I think that you have this thing that I try to have and everyone I think at every tries to have and that I think is just really good for working in this sort of like AI wave right now which is the spirit of tinkering you're just playing around with stuff you're making stuff you're constantly shipping new things um and it sounds like you're discovering some cool things along the way so I want to talk about that I guess to start um tell us how you got into this like how did you start how did you become the AI guy at USV so um mostly just by I would say following my own curiosity so I've been at USV for six years now or just about six years and for a bulk of that time I was working on the talent side so you know meeting people who were looking for new opportunities and connecting them to the right people in our portfolio which was always kind of a weird fit for me because I never came from a talent background um in the first place had never been a recruiter had never hired Beyond just like hiring people from my own team but was still a lot of fun meeting all those people but I think everyone here knew that my heart was probably semi in it after a while because I would send these weekly emails around to the team that were just like here's everyone I met with last week that's looking for a new job here's everyone I'm meeting with this week who's a good fit for someone in your portfolio but then I would um sort of cold open all of those emails with and here's this weird thing I made over the weekend that you might want to play with and at first it was mostly just uh what I would call like fake movie trailers that I was making in Runway ml and I would like outline my whole Pro this was supposed to be an email about talent and it was like so I had this idea about a romcom set in the French Alps and I like gave that to chat GPT and I asked chat GPT to write like a 30- second script and then timestamp each part of the script and then each timestamp needed to be like a digestible prompt that I can freed into Runway ML and would just like run that process and everyone just be like okay uh looks cool and then um you know I went on paternity leave I had a we had our first kid and when I came back I was having a conversation with you know Andy and Rebecca are two of the partners here and they're like like look the world is changing really fast and there's a lot of really interesting Technologies out there that you know both we need to be experts in but at the same time like we need to be deploying on our own to make us better at our jobs like you are a tinkerer by Nature that's kind of what it is that you do like would you be interested in just kind of doing this full time and I think maybe I played it a little cool but in the back of my head I was like you mean I can just like make prototypes with AI all day while working here like yeah okay let's do it let's go and uh so yeah that's kind of that's kind of how we got here I love that it's such a good story some of the best people like stumble into things like that by like they're they're doing their job and then they're like just sneaking in the thing they actually want to do and somehow it becomes their main thing it was really inspired by we had this analyst who worked here uh many years ago who's now the CEO of a company called Jam her name's Danny Grant maybe you've met her um and when she was an analyst she would come in over the weekend and just be like hey look at this hologram that I made that sort of made me realized that this was the type of place where that type of experimentation and just like having fun on your own time and following your Curiosities was really rewarded and um I think it's proven to be the case that's great so it sounds like you started with like Runway Clips like what was the first thing that got your eye where you're like this AI stuff like I kind of want to just start playing around with it and what was it you know it I I don't remember it whether it was Dolly or whether it was um chat GPT first but I do have this distinct memory like I you know I like I've always been interested in sort of the more creative side of of AI and using it as a as a creative assistant and I remember I was staying up at a friend's Upstate and we were writing a musical about the life of Jeff coun using Chachi PT and like giving it feedback this was probably two plus years ago and giving it feedback and like we were like this is outrageous like it's just coming up with this story live and it's taking our feedback and it's getting better and better and so I think that was a really pivotal moment and then the other is you know my wife is an illustrator who I I wouldn't call her a lite but she's definitely not terminally online like I am and my first thought was like oh I wonder if she's going to be to feel threatened by this and like the first thing that she did was that she had Dolly design a crossstitch pattern that then she could do in real life and I was like oh you know wow like the these aren't just like these are tools that actually help bring out our creativity if we if we use them right what's the thread that turned you into someone that can be like a tinker and and uh and like wants to make all these different types of projects with AI like tell me tell me about that sure I probably if I had to guess I'd probably get it from my dad who was like the type of person who always had some like broken down piece of Machinery either like a car or like a computer that he was building from scratch or or a ham radio like I remember he went through a real ham radio phase so he was a real Tinker so I think it was always in my nature to want to make things and the problem is like I just had so skills I was quite the generalist always I mean my whole career I you know I was I worked in startups but in like weird General managerial kinds of positions even when I was really young and so there were all these things that like you know in my spare time like I'm trying to learn piano and I'm trying to learn all these different like hard skills that I never sort of gained growing up and then all of a sudden these tools came about where um you know I wouldn't say that you need no skill but you can develop the skill alongside seeing your creativity come to life like if you want to learn how to play a beetle song on the piano like it's going to take you a long time of learning your scales and then like building up to to the position where to the point where like you can actually play credibly and then if you want to sing on top of that like my God we're talking about three years of of lessons and whatnot but with AI you can kind of learn alongside the creative process which has been like super rewarding and the sort of the the feedback loops are so fast and so you know a great example of this is um you know I'm not a software engineer but I had made something like I had made a custom GPT and I was like you know this would really work so much better if it read off a live spreadsheet but I don't know how to you know connect two apis so that the GPT can read off the live spreadsheet but then you know because you're already so far down the process it's just like adding another little element to it and like the GPT can walk you through how to do that and it it was just kind of it's I don't know it's kind of been this magical experience for me yeah this is this is one of the things I see a lot too uh in AI stuff is like I teach a um a programming course like programming with AI and one of the things that's very different about that course um from like for example the way I learn to program which is like the way I learn to program is the exact same way you're talking about learning to play piano it's like okay what's a variable like what's a function and you're like I want to make a website I want to make like an app and you're like okay no no you have to spend a year learning about these like really basic building blocks that like you don't know how they translate into into the full thing and what I can do with my students and what anyone can do regardless of whether they're taking a course with AI is within the first 30 minutes you can make something that looks pretty much like the thing that you want to make and then you can be like whoa wait how does that even work and then you can go down into the like into the okay here are the like in B to the details of how it gets built up and I feel like there's there's a whole like generation or crew of people who um are have not been able to for example program because they don't want to go through that whole thing like learning all the like basic things before they're before they see something that they connect to and now they can connect to that and then they can go back and learn I think that's so powerful you know it reminds me a lot of like the personality archetype Who as a kid takes everything apart to learn how it works like they take apart the clock radio to learn how it works and like you can actually do that now you can build the clock radio and then you can take it apart to see what it is that you just did and then like sort of build your skills that way and I think a lot of people learn much better that way totally totally I know I know I do um and and I feel like I have I have a similar thing where I have all these different interests I have all these different things I want to make and build and AI just like I can now do it and yeah it's just it's the best um like literally the only thing holding you back is time and patience at this point yeah exactly I want to get into some some of the like practical stuff that you're making because I think it's so cool we could start with the librarian what do you think is the best thing to start with uh well there's like six different ways that we could sort of talk about the librarian because I feel like that's sort of the Persona that that's taken off and and I'll pull it up you know when we first started working on all of these projects I think like a lot of people my my imagination just like exploded and it was like oh my God like we are going to build this like allinone monolithic like super app that's going to live where us with us wherever we go um and it's going to have a name and a face and a personality because like we all believe that metaphor is really powerful in helping people understand things and I think over time what we realized was that if we wanted to build things that work now then rather than trying to build like a monolithic app let's use the tools that already exist let's break them down into their component parts and let's build individual agents that do different things and so there is a version of The Librarian that we're still building that still kind of resembles that but honestly like my favorite version of it is just like a simple GPT that we cloned and like called The Librarian and I can pull that up if you want yeah pull it up and give us some background on what it is and like and where it came from and all that kind of stuff okay so uh where to begin well when I first sort of moved into this position the most obvious thing for us to build would be a a chatbot built on top of our entire history of writing so as a firm we've been around for about 20 years and we've been relatively prolific writers I think between all of the partners as well as the USV blog there's about 15,000 articles that we've written um a lot of that has been driven by you know one of the partners Fred wrote every single day without fail for 13 straight years um and I was talking to Albert another one of the partners here and he's like you know one of the biggest impediments to me writing is I sit down I'm halfway through a blog post and then I realize oh my god I've I've actually already written that um and so what I wanted to do as a starting point was just build a bot that allowed them to be conversational with their writing and I shared it and um everyone sort of looked at it and they played with it I think the original name I gave for it was um called conversations because we like to say that USV is a conversation and everyone's like okay but what does it like do and I couldn't really give a good answer and then I renamed it to at the edge because I was like you can ask it questions about our thesis and really like you know try to it synthesize and Advance our thinking and everyone's like and then they're like but but can you explain like what it is I was like I don't know it's kind of like a librarian and then there was like this light went off and everyone's head to like oh librarian well that that's interesting and that was sort of this moment I think where where we realized how powerful metaphor can be and and sort of explaining what it is that these chat Bots can do for us and so you know I started building out what was the librarian and initially like I said we had sort of put everything into this one chatbot that was sort of its own Standalone application that I actually used uh no code MBA to to build which was an awesome program um and then a lot of help ALS Al from Ben's bites um but like just maintaining our own UI was such a pain in the butt so um I you know at least for between now and November I was like you know what let's break down all the component parts of the librarian into the individual things that the librarian was doing let's make them all their own Standalone gpts and give them very specific names so people know exactly what they do so this is an example of the librarian so um one of my favorite things to do and I did this recently we can do it live is like consensus is a company that we recently invested in and so um we can go to the about like I'm just going to go I'm going to copy everything on this page I'm going to typ go here and say you know below is the about page of a company we're considering investing in pull out any relevant blog posts that we've written that might pertain to this company um and then I have like an actual because you know this is going to take a little bit so um I can show you ah here we go so this was one that I did the other day it was the same thing consensus and sort of it brought back you know these are the major themes that we've written about that would be relevant for consensus can you stop for one second I just want to I want to um I want to go back and read it so if you go to the top okay so basically so what you did is you put in just like the the about page and you're like you know uh what are the relevant like what's the relevant writing from USV and and it had this this gbt has in its knowledge base all of the blog posts you guys have published right yeah um and so it says here are a few key insights from USV writings that are highly relevant AI unlocking knowledge one of USB's core beliefs is that AI can unlock vast amounts of Knowledge from data making it easier for people to access and synthesize information this aligns closely with consensus' Mission um and then it keeps going democratizing science USV has supported platforms that democratize access to Scientific resources trust in AI systems for AI Power Systems to gain widespread Doctrine they must earn users trust so this is like it's this seems pretty good right tell me about like uh quality of the response yeah your reaction to this response oh I I think it's spoton right and these are all things that both we've talked about and written about and in some of these these are things that we've written about really recently like trust in AI systems in this instance I think it did quite a good job of interpreting what it is the consensus does what their core brand promise is and how that relates to both things that we've written about that might be relevant to the tech they're building but also to the brand they're building as well um but maybe I'll pause just for one second and explain like so so why is this even important like why why do we need to pull up our writing about a company that we're considering investing in and I think for us um we don't like chase after deals at USV I think we're quite patient we like to talk about ideas or we like to go out into the market and do our research we like to formulate ideas over time and sometimes we can spend you know six seven eight years finding the right company that fits within the thesis that we're looking for and so a lot of the times we're trying to prove to entrepreneurs like Hey we're not just chasing the hot thing like you are actually what we've been looking for for a long time and we have the receipts and so that's kind of what this tool allows us to do that's interesting and one thing that I noticed though is like in these summaries it's not saying like Fred wrote in you know 2015 blah blah blah um yeah yeah so um I said can you just share the URLs to those posts they don't seem to be working right because that's obviously the first thing that always happens with any of these Bots is it gives you back it like it cited the articles for me and then it gave me these actual URLs and this is like a little trick I've kind of learned about chat GPT uh one of the ways that it likes to hide the fact that it's lying to you is um it'll give you a fake URL and if your cursor is showing up like this it means it's a fake URL that doesn't exist whereas really that's so interesting okay so if the cursor shows up as the same cursor that you see when you're highlighting text that you can type in instead of like that mouse that Mouse hand that means that it's that means that it's lying let's go look oh my God that is completely made up link so I was like then you could see I said none of these are real let's try again with a smiley face and now look at these you see how my mouse cursor is changing trust AI science exchange and then this one's still fake Ai and crypto but like you know this is a blog post about that time we invested in science exchange a trust in AI this is a blog post that you know Andy wrote about you know trust in AI as it says and these are real links yeah um it's like you can now use links to the consensus team and I was like well let's get a couple more maybe something that's more about search because we've written in this case I'm using my own knowledge about what we've done uh which is that I know we've written a lot about search in the past I know we've invested in search so let's try to pull some of those in as well because you can't I say this all the time but if you're relying on the AI to produce the final product for you you're always going to be disappointed but if you're relying on the aii to help you get to a final product then like I don't know I find it extremely powerful because normally this type of work would have actually taken quite a long time right right that makes sense so here we have the fragmentation of search duck ducko or you know our investment in duck. go and I was like okay now let's rewrite the email but include these articles the real ones leave out the AI and crypto one because that wasn't really relevant and then also make sure to differentiate between the blog posts you pulled around duck ducko and science exchange which were relevant Investments we've made and then the other articles which are more like kind of what we're actively looking for and I had instructed it earlier to write it in the voice of one of our partners uh Jared he and I think it did a really great job like been following what you're building with consensus I'm really impressed by your mission um we we are long Believers in the transformative potential of AI to unlock knowledge and make it more accessible which by the way was like a core part of our thesis for a long time it did a good job of pulling that in we've made investments in companies like science and change Duck Duck Go both of which here ethos of lowering barriers to specialized knowledge and rethinking how people engage with information you know in terms of what we're actively thinking about here's a couple of other examples now I don't think anyone on the team would ever allow the AI to write the email for them I kind of did that just to show what's possible but um and basically like demo this to the team like hey these things that you do to sort of prove to an entrepreneur that you're aligned with them that take you a long time we can now do it way quicker and kind of in a fun way and then I added one extra step to this which I'd love to show you which is I have this other GPT which again this used to be folded into the librarian and I separated it out called portfolio tracker which is kind of like you know it categorizes all the companies that we've invested in when we invested in them what our ownership percentage is all of that and so in this consensus example uh I asked the portfolio tracker um build a chart that plots out all the companies we've invested in in education search and AI over time because those are sort of the three areas that were most relevant to to this team and you kind of have this chart that goes all the way back you know 12 years of us investing in these spaces but from my own again first-party knowledge of what we've done I knew that indeed which was job search which is you know a big part of the disaggregation of search early on was missing here um but you know first actually first I asked it to make the colors more pronounced so that it was clearer what was going on uh and then you know I said the only company that isn't on there that falls into the search category is indeed make sure you include that one as well but the reason why I didn't indeed didn't have a date so I gave it the date blah blah blah and now we have this chart dating back to 2005 of all the relevant Investments that we made in the space that this company is working in again I don't think we would actually send this chart to them but this is all knowledge that would have probably taken a really long time to gather up that now we can do really quickly would you send the GPT to them and be like hey you can ask questions about like what we think yeah absolutely we haven't done that yet um but I think it is highly likely that if you come to our website at some time before the end of the year that all that's going to exist is the librarian I think that's really cool like what I want is like I want like Andy and Fred voice mode um and just like just get to ask them like questions about ideas that reference and reference all their blog posts yeah um to date everything that I have built has been more for internal use and when it's internal use you have a certain level of tolerance for um information being wrong because you know it's all first- party data for you so like for instance indeed didn't show up but I knew we invested in indeed in 2005 so I could correct it in that way and so we're just thinking through how to get this to a place that it's just a tool for public consumption I mean what I really want to do not just like get Fred's opinion on things but um and I think this is this is also on the road map for relatively soon is that you can just dump your deck into here and it can equip you with what we've written about in the past and that like sort of enables you to be able to come to the meeting also ready ready to have a conversation in the way that you know we're sort of thinking about it as well and kind of get aligned before you even walk in the door and what are you thinking about like the the place where my mind goes always for this kind of thing is like um investing is this activity that you can talk about in terms of rules or maxims but really it's a highly intuitive like pattern matching type thing where you know Fred or Andy has like developed over many many many many years the ability to like select the kind of opportunity and the kind of person that they like have a taste for and to some degree like um tools like this might be able to um replicate some of that taste probably not all of it but some of it um and have you explored that at all which is like if I wanted to get a rough idea of like what Fred would think about this like I could throw it in there like for internal use um or is that not really on your radar no there's two reasons why uh number one is the team is highly accessible here so like if I want to talk to Fred I'll text Fred right now and ask him his opinion and that's sort of how we operate and so there isn't really a need to internally at least abstract away uh the way that the different partners like to think about deals and then build a bot to sort of get their opinion because they're so accessible what if they're gone you know like I hope this doesn't happen but like you know Andy could like you know get sick and I don't know like whatever like you know people don't last forever um yeah um I would then say look we can always query Andy's writing history and talked to a lot of companies that are building digital clones for whatever reason it's just not what we're interested in and I think we believe that in general venture capital is a highly relational business and we don't necessarily want abstractions of those relationships to replace or even augment kind of the the real relationships themselves um and in General our approach to AI has been enhanced not replace so you know knock on wood if someone were to disappear like it's not our intention to be able to build something that can replace them like we we would just mourn them that's a that's a very good answer um I will say like I think there's room for there to be like compatibility between embodying someone's perspective in an AI like this and like strengthening human Rel relationships yeah like I don't think it's like an and or or like a question of augment or replace like I think you can use that like even Fred or Andy could use that in know like what would you know five year five years ago Fred have thought about this and like how have I changed or whatever you know like there's lots of different ways to use that tool um I'm gonna augment my answer a little bit which is to say I don't think we would ever build that tool to help us to decide what to invest in I think we would build that tool to help us uh help our companies more and so what do I mean by that every week we get together and we discuss all the potential new Investments that are on the table and then we walk through the portfolio and each partner gives an update on not necessarily every company in their portfolio every week but for any company that there's something newsworthy or something that they need help with um will'll have those conversations and the sticky situations that get thrown around in there are worth you know 30 mbas and so what we are working on right now is how to capture you know how would Fred handle this particular situation so we don't have to have that conversation 6,000 times right that's really interesting so for you know for example like you know a CEO wants to step down um and or sell the company like have you ever gone through this before um and then there's a wealth of knowledge that gets passed back and forth in that Monday meeting that isn't necessarily captured and is not something we ever want to lose yeah that makes sense this reminds me of like one of the one of the core ideas that I think is coming out of this particular era of AI for me it particularly uh applies for me to creative work but I think it applies in in all areas of business business um which is it starts to reveal how much we actually repeat ourselves um because it's very good at when something has happened before um in the in a similar context like bringing it bringing it back and and and and changing and giving the answer for that particular context um and so when you know that you have a tool for that you start to see the world through that lens um and I think it's it's there all the time in creative work you know it's like uh doing headlines or like making a tweet or out of an article or like whatever and that's why we built this tool spiral to do that um but I think it's the same thing in investing it's like um when a CEO wants to step down from a company like that feels like a very unique situation and there probably are some unique variables but like if Fred has seen that like 5,000 times like he pretty much knows like the three cases and three ways to respond or whatever um and and having that available like is helpful yeah I mean like the the actual idea that the actual conversation that comes to mind that happens a lot is like CEO wants to the CEO steps down we need to bring in a replacement CEO should that CEO's goal be to retain the entire executive team or bring in their own team right can I can I try to answer that question I'm curious what the USB perspective is yeah sure I would say bring in your own team I would say get rid of everyone I think depends um so actually the way that you answered the question is kind of right but I I think it's the answer is you have to bring in your own team but that doesn't mean that the people who who are there aren't your team that's right yeah so that that's what you have to sus out pretty that's what it seems like you have to sus out pretty quickly like are is the team that's remaining is like are they are they ready to ride with you as well and if the answer is yes then you keep the team um and if the answer is no then you know for better words the only way you're going to be successful is probably bringing in your own team and of course you know there there's Nuance to all of this and that's where you know I've been throwing this idea around with my colleague Nick Rossman we're calling it playbooks and like there are all these playbooks for these really tricky situations that we've never written down before and so how do we trigger like our AI transcribers to make notes of like when those sorts of conversations are happening and then recording those as playbooks is a is a project that we're working on right now that's really cool I love that I've definitely you know I i' I worked at an incubator before um before every and we had a similar thing of like trying to write down playbooks of how we made decisions um what I think is really interesting about all of this is that um USV can have a playbook for something like what to do if you're an incoming CEO of a startup that just fired their CEO but the but the Playbook that you would write is different from the Playbook that like Founders fund would write yeah sure and um what I think is so interesting about that is uh even even though they're quite different they both work um but they probably work in slightly different contexts and um USV is probably going to attract Founders that like and and types of companies that work for usvs like sort of worldview and and kind of advice not always but more likely and Founders fund is going to kind of like like be giving advice in a context that's more suited to the founders fund Style of advice um and I don't know I just that that whole thing kind of like because I I think I've always been sort of fascinated by if you ask five successful people how to handle a situation you get five different answers um and they all kind of work but they're all very contextual and I think AI helps you kind of see that a little bit well I I would say this I mean in in the world one of the things that we like to say is you know this is very much it's very much like a marriage like at the end of the day the product that we sell is money and our money is just as green as Founders fund as a16 Z's as you know whoever else like everyone's selling the same product um sure you can dress it up with all these sorts of value added Services which a lot of funds do I think Founders fund if I'm correct uh sorry well you know some provide none and that's actually what they believe is the value ad that they get out of your way um but at the end of the day like the real differentiator is the relationships and perspectives that and like do you align with the people that you're doing business with and so yeah I think you know obviously you can sus that out through like the diligence process and reading historical writing and whatnot but that stuff happens so fast like Founders sometimes are making decisions over the course of two weeks about people that they're going to be in business with for the next 10 years and like they don't even make decisions that fast when it comes to hiring and so it could be I and and you know hiring is not as permanent sometimes as a cap table adding someone to the cap table um which is kind of a lifelong commitment to the company and so yeah I think conversational tools that can like really help you get a sense of the type of person you're going into business with how they think how they like to work I think could be really helpful for Founders making that decision totally okay I want to keep going I know you have like a ton of different tools like let's uh let's uh let's keep going what's the next thing you want to show us um well I mean my favorite thing that I've made recently was this meeting notes so we've started transcribe I mean we've been doing this for a while we've been transcribing all of our historical meeting or sorry all of our meetings for the last like four months it's actually not true it's not all of our meetings um we do these sort of special topical meetings around like AI crypto um uh climate and then we do sort of this miscellaneous one and that's sort of where we hash out ideas and discuss specific companies and so we've been transcribing those and now we're sort of porting automatically pulling those into this GPT so that you can ask questions to the meetings I kind of want to make two sort of separate points you know one is well why do why are we doing this all here when there are better purpose-built tools and I think the the one of the main reasons like for for instance like we could do all this in Granola you don't need to build a special gbt um in order to ask questions to meeting notes but what I think is really nice and I can't necessarily show this to you because I'm recording myself on my phone but if any member of USV opens up like their chat PT or their Claude on their phone like the there's a conversational interface with like every part of the business in the left-and column that they have access to right there without having to toggle between a million different apps that's really cool yeah so like sure like I actually really prefer notebook LM for instance for for analyzing meeting notes but for sometimes like ease of access is is the best feature and I think whether it's Claud projects or whether it's you know gpts on chat GPT like for our purposes I think that just works a little bit better wait I gotta go back well shots fired on Notebook LM like why do you prefer it for meeting notes um have you ever just like looked at the qu like so okay so there's there's there's three reasons why I like notebook LM better um number one is I think it comes up with really insightful uh questions to ask the notes I think number two is um by the way are you saying this is shots fired on spiral no no no no shot's fired on cha PT oh okay I was gonna say I mean all these products are amazing so um number one is I think it asks it it prompts you with really thoughtful questions about the meetings that I sometimes would have not come up with myself number two is I think it does a much better job of citing the transcript or you know citing the things that have happened in the meeting so I can verify whether or not that actually happened and then the third thing is um my new favorite thing in the world is not to read meeting notes or read meeting Recaps but it's to listen to meeting podcasts and so I have been anytime we have a really like important meeting I mean I can just pull one up like we had um this guy Chad hurwit who directed The Eno film it was a generative AI film about the Life of Brian Eno that was running at the Film Forum in New York and like he came in and just like talked us to us about his process obviously I attended that meeting but it was super fun to just like also turn that meeting into a podcast and then like give it to my wife and be like how cool was this meeting that we just had dude yeah I love that or um you know we had a really really dense like really dense AI meeting on Monday and I shared the notes around I was like but I would honestly recommend listening to the podcast because it's a lot more fun and it's something you know they're all 10 to 15 minutes long so like you can do it on a subway you can do it on a city bike and what I like about podcasts as a meeting recap um is you know you can do it while you do other things like I can't read and do other things at the same time but I can listen to two people talk about a topic while I'm responding to emails or like riding on the subway or whatever it might be totally we're so on the same wavelength here um so we I I think you probably seen this but like two days ago we launched uh we launched officially launched our studio where we do product incubations and one of the one of the ones I'm most excited about is we built something that um it takes uh we record all our meetings um and it also takes we use Discord instead of slack but it takes all of the messages that happen in important channels and then it creates a podcast of like what happened um and it's like the it's the notebook Al style thing and it's so good like I because we're at a scale now where like I can't be in every meeting which is like it's wild to me because I'm like I I write I write a newsletter and do a podcast like that's that's weird um but uh uh but yeah we're at a sced where I can't be in every meeting and I just love listening to a podcast about a meeting that I deeply wanted to go to but didn't have time to attend like and it's so it's so helpful and um and one of the one of the things I love most about it is you feel so good when the host mentions your name I'll tell you what do you know what makes you feel even better what when the host is like whoa that's a idea it's so good like I'm so excited for this product like I really think so I think we're going to do some like something external where you can listen to every meetings um but I think we're going to just release it as a product or any any any company yeah I I would love to I would love to have you guys try it because I think you're the you're the perfect place to to to give it a shot but it's it's really cool and it's it seems like you're doing it already I want to can we listen a little to a little bit of the Eno uh podcast yeah for sure ever catch yourself hitting that replay button on movie not necessarily because it's brand new but like because it's familiar like you just want to settle in with an old friend yeah totally but what if you could have that that comfy familiarity but uh but also a totally fresh take every time kind of like seeing a band play your favorite album live that's an interesting thought yeah like the core the core is there but there's a a unique energy to it each time and that's believe it or not what filmmaker Gary hustwit is is messing with in his documentary IO which is wild because it's been playing for like 12 weeks straight at Film Forum in New York City and people all right let's uh that's so cool I love I love that I honestly like I would keep listening to that I'll send it to you yeah send it to me send it to me please like well I I actually I really want to start because like we're trying what what we're trying to do is like let people into a lot of our processes earlier because we're you know we're sitting in our office and we're like talking about ideas and it sometimes could be months before we really communicate it to the public and so we're trying to find interesting ways to let people in a little bit sooner and so I might start releasing all of our meetings as a podcast that people can listen into you should totally I love that idea I think um we'll release one today then perfect let's do that I one of the things that I feel um I feel really strongly right now is um do you remember the startup Justin TV yeah of course so I think that Justin TV would actually work now because the reason why Justin TV didn't work is um if you're recording yourself 247 and for people who don't remember Justin TV it's like uh Justin Khan who eventually started twitch uh was just like wearing a video camera around and like streaming uh videos of himself like doing his whole life basically and the reason why that doesn't work is because like out of any given day for a normal person there's like 12 interesting minutes um and you have to like string those minutes together into a story in order to make them compelling and you know tuning into any given minute of a person's life is like not that not that uh uh not that interesting and one of the things that that I think is like the the cost of Storytelling of good storytelling is going down dramatically and what that will do is not necessarily like you're not going to necessarily like read novels written by like AIS like it's not going to necessarily just replace all human novelist but we will be able to tell stories high quality stories about places that we normally would not be able to afford to tell stories about um and and one of those places is like your company meetings um or whatever Justin TV was like now Justin Justin Khan when he was 20 and not a billionaire can afford to have like incredible stories told about his life um because AI can pick out the like the interesting things and leave the rest out you know it's it's funny that you say that because like we talked at the beginning about like the the little over herds that we do I mean a lot of that was born from this idea that hey we're not producing enough content and I was of the opinion that like well actually we're producing a ton of content like every day we are getting together and we are producing a ton of content we're just not capturing it and we're not necessarily like curating it enough and so that was really the impetus behind that was like there are these great stories that are happening in our meetings that just like end up sitting in our heads for weeks on end until we can put pen to paper to get our ideas out there and like nothing is for us nothing is ever going to replace writing as our primary mode of getting our thoughts and ideas out into the world for a variety of reasons but there's a lot of things that we can do to tell let's call it tell stories in the meantime and then there are lots of things that we can build that will help us write quicker and easier than we've done in the past and I think that's kind of where our Focus has been that's really cool I would love I know that you you said we could um we could get an an O at USV out I would love to see you go through that process well so I I'll start by saying this process has changed so many times since we started doing it um it used to be fully automated then it was super hand done and now it's kind of somewhere in the middle but let's do it as automated as possible so I'll go back to to share my screen again we'll open up chat gbt let's do our meeting notes so I'm having a little bit of an issue with meeting notes understanding uh what most recent meeting means so I've been a little explicit recently of you know let's produce five Oh's from the AI meeting on October 15th and the three meetings on October 18th don't you know explicitly mention any companies or anything specific to a company because you know when we were doing things a little more automated sometimes companies would come in and pitch and I would I would we wouldn't really want that like we wouldn't want their pitch to to go out into the world and make it look like hey these are our ideas um so let's see how we do so it's saying here are the five o posts based on the AI meeting on October 15th and the three meetings from October 18th one commoditization is coming the race for dominance in AI models may lead to a future where foundational Technologies become a commodity it's a reminder that differentiation is key to survival two decentralized to survive we debated if decentralized AI models can compete with their centralized counterparts it might be the only way to break free from the current Tech CH hold I think that's I think that's a really good debate um three hydrogen risk or reward as we push our Greener Tech hydrogen's potential for clean energy remains controversial both promising and dangerous safety concerns keep the debate heated four uncorrelated bets matter the unpredictability of ai's evolution means diversifying Investments is crucial interesting and then five a micro grid Revolution are microids are gaining traction and their potential to revolutionize local energy is undeniable these are these are cool what's really interesting is is I want the ability to be like okay tell me more about uncorrelated bets you know yeah I mean I could tell you more about uncorrelated bets um well I guess I'd throw this question back to you it's like how do you think we should release this into the world so that people can ask follow-up questions into our meetings you know like we tweet these out and they get some engagement but I don't know if it's that you know people would rather have a private conversation with our meeting or whether they want to have a public conversation with our meeting like you know how can we get people to ask us more questions about uncorrelated bets I think the the the key thing is when you read this so like uncorrelated bets matter inventure the unpredictability of ai's evolution means diversifying in Investments is crucial the real question which bets will prove to be the game changers so um that is posing a question but it's not giving me any hint at all about what your your take might be right like USB's perspective is completely hidden there so I know the topics but I don't have any concrete idea of like oh you think that if you're going to invest in niai you also have to invest in uh like the Japanese whaling industry because they're like completely uncorrelated or whatever um so like so that that would be my my guess is that the summaries are um and I think this is just a problem generally with AI generated summaries is that they're good at finding the the the interesting topic um but they're very good at like telling you like saying things without saying very much at all um yeah you know often times if something pops up as like an overheard at USB which means it's something that we're discussing in a meeting the reason why we're discussing it in the meeting is because we don't always have the answer yet typically when we feel like we have the answer that's when we write about it um now with uncorrelated bets I could you know we are at a point in our thinking where I think we have a strong idea of of the direction where where we want to go U which is that like you know there are there's like this x and y axis of of different things that could happen within generative AI like on one end of the spectrum is we could be living in a model in a world of like an oligopoly when it comes to models where all the power is within four companies um or we could be living in a world like a multimodel world um so that's you know one axis and then another axis is like we reach a plateau in terms of ai's ability versus you know we get runaway AI or self-improving Ai and that's another AIS and so you know there are a range of different outcomes that can happen based on the answers to those two questions which are questions we don't necessarily have the answer to yet and as a venture firm it's important that you make bets that um occur sort of within each scenario so that if one of those scenarios does happen like you've made the right bet and hopefully you backed the right team and one of those scenarios doesn't happen then uh you know and you only and you put all your eggs in one basket then that's not really good portfolio management yeah no I think that makes sense I think the then even if you don't have like a particular perspective if I knew like okay um Andy brought up uncorrelated bets because like he's concerned XYZ and so that became an open question or if I like um Andy and Matt argued about like what it means to be uncorrelated and Andy's position was this and Matt was was this but it's like a open question even that is even even if it's not a statement about what your your full perspective is even that gives me enough detail to be like oh wow that's actually really interesting I don't know the answer to that question or here's what I think you know yeah so I think because we said that we were going to push these Live this week we should do that but I think what I'm hearing from you and we'll do this next week is let's alter the system prompt a little bit to make sure whether we're identifying whether something is you know a topic where we have a really strong opinion on and if so let's say it and if it's one where it's still an open question let's let's leave it as an open question do you think we could like try it like just um not necessarily even um modify the system prompt but just follow up and see if we can make these better or if you want to modify the system prompt we can oh well I mean I'm happy like what would you like me to ask okay in this I think that I would say um none of these tell me anything about the like positions that were actually taken by anyone in in the meeting can you be more concrete about what the different positions were what people thought so I can tell you now it will not be able to tell you who said what that's okay and that's not for privacy reasons that's just limitations on the technology yeah I think that's totally fine it doesn't have to have the names it's more like yeah um I just want to know what the perspectives were okay so it says um commoditization is coming the concern about commoditization of foundational models was voiced by multiple participants one side argued that these models driven by high burn rates are becoming indistinguishable leading to to a race to the bottom another perspective emphasized the need for differentiation particularly through proprietary data to escape this trend the consensus lean towards cautious optimism for companies that can find unique dat data sets or applications see like that actually like I'm like wow these are the two the two ideas um and then now I kind of understand like where the conversation went and then the thing that I would maybe think about is how to express that as part of the headline you know um but yeah I mean that I'm more interested in that personally I don't know I don't know what you think um so in this case um you think it's better to to reprompt a few times until you get something that's extremely valuable versus publishing the unfiltered results of the L yeah let's do that then great so I mean this is how decisions get made here too do you think that I mean because because the the first one which is publishing the unfiltered results to me that's more like a barometer of AI progress than it is like what's going on at USV right um and I think the the idea for you guys is not necessarily like creating a barometer of of progress um like directly but more like actually just get out like what are we actually thinking about and using this cool technology to like tell that to people but you tell me no I mean I well so here here's where it gets tricky for me is like when we look at this one decentralized to survive the the AI actually got this very slightly wrong because we were not talking about decentralized AI models we were talking about specifically decentralized training which is you know it's different um and it doesn't really because like the you know the general idea here is that the giant like the really big companies are well suited to win because they you know they build out the these enormous facilities these enormous clusters where they're doing all their training but if you can figure out a way to do decentralized or distributed training then you don't necessarily need the same um like high capital investment to get started uh in order to compete and so like that seems like a really important area to be looking for investment so that we can build towards a world a multimodel world and like that isn't really captured here but it's close it's yeah I think you're totally right I wonder how it would do like if it wasn't a one shot thing and it was like um each time you did it for one topic um as opposed to all of them at once yeah you know um but yeah I think that that that makes total sense and and and that at least for me that would be something that would annoy me I would be like I need to like we need to get we need to get this right and that's such an interest like that's so interesting like no one's talking about that no one is talking about that and it's really important that like more people do I like I it makes the thing that I think about is like well all those people that were mining Bitcoin should just be like you know financing AI training runs you know and like that's cool yes and that's where our head is at with number two okay got it um cool well I guess we're we're on the same page or or at least I can follow your line of thought once you explain it to me right you know it's it's this fine line the conclusion that I've kind of reached and I think you've kind of helped me see it here as well is that I kind of think that the the market for purely AI generated media is like the size of your largest group chat and um I think I think anytime we do something new what people are actually interested in isn't necessarily the content that was produced but how we did it yeah and so I tend sometimes to over index on the like let's keep as many hands off as possible but in reality if we go back to sort of our first principles of enhance not replace uh when it comes to our own internal AI projects you know we should be using these tools in partnership with humans to create better content that's actually valuable for the people who are consuming it and so I think you you changed my thinking a little on this amazing I love that I mean I I love I love the you know the value of purely AI generated content is this the size of your biggest group chat which at work is your slack channel is your whole Slack and that's why I think the work podcast thing is really interesting because to your point like um you can you don't have to read all the emails you don't want to read like while you're sitting at your desk you can like you can listen to the listen to all the things you missed while you're doing the dishes which which is another point that um I was talking to um I was talking to someone else uh uh a few days ago on this podcast about um do you know Yohi Nakajima yeah of course my inspiration for everything yeah he's absolutely amazing and he obviously he does like all these like really cool AI side projects like has a similar Tinker vibe to you to me to like a lot of the people I really like hanging out with and um one of the things that he said that I think is so interesting is that because I asked him like how do you do all this stuff you're you're you have a day job as VC you have kids and you're like programming all the time like how is it even possible and he was like well first of all I only do it at night after they're in bed which makes sense um but also I mostly do it in the morning so okay there you go um but also he's he said like because it's doing a lot of the coding for me um I can do it with divided attention I can do it with fractured attention so I can say okay go build this thing and then I go do some dishes and I come back and I see if it did it and then if it did I say okay do the next thing um and he doesn't necessarily need to be in that like full Flow State all the time and so I think a lot of the tools are um going like the effect for example of um Aid driven storytelling is uh you can listen to stories uh about content that wouldn't have necessarily been storified before and that means you can consume that content in in places that you wouldn't ordinarily have been able to consume it like listening to your meetings that you missed while you're doing your dishes or um you know and and in programming cases you can you can create things in situations where you wouldn't have been able to create them before because you don't have to do all the work uh in this like super zoned in Flow State yeah though you know the one counter is should I just do the dishes totally that is that's a really important question which I think which I think is really interesting right and it's an extension of a question that we've had to ask ourselves like since uh since cell phones or since you know since the internet or whatever is like we're bringing all this stuff into and integrating into every aspect of Our Lives I think one of the things that I feel that makes me sort of optimistic is um because you can now talk to computers um you can express with much more um fined grain control what you want to see and when um and so for example uh I I really think we're get to a world pretty quickly where like instead of seeing like the trash Twitter algorithm that everyone else sees that's like optimized for like the like ma most amount of Engagement possible on a mass scale I'll be able to see say like here I I want I only want to see if you show me an tic if you show me an article um on Twitter and it's about science I only want to see articles where uh if uh where it was not just a mouse study so like or or don't show me any Finance articles unless the market has moved more than 3% um like stuff like that um where as we get as things get more integrated we also have the ability to have more fine fine grain control about what what we see when which hopefully is healthier but you know also people do dumb stuff and and and and yeah there's there's trade-offs here it's not it's not all good yeah I mean the the the challenge there is that so of our internet lives are owned by these handful of data monopolies like you know meta like Twitter and it requires either breaking those monopolies or behavior change on their part in order to kind of build a a more customizable internet in my opinion um and so I think it's worth backing lots and lots of businesses that that can potentially break those monopolies and in addition to that I think kind of forking over full control of llms to you know four or five major Tech conglomerates is really problematic for society in the long run I think you're totally right and the thing I will add to both of those things is I also think there should be a social movement um to ask uh large tech companies to establish data trusts um where they they make all of their data open to researchers um because I think uh it's so silly to me that PE like researchers at universities are doing like studies with like a sample size of 16 people when like Facebook is like sitting on like trillions of terabytes of like the most valuable behavioral data ever to like understand what it is to be human and like what works and what doesn't um and they I think and people hate tech companies right now like I think there should be a social movement for them to like donate it for the public good and also to um while we're at it to make sure that all data can be exportable by an individual because suddenly like the sequence data for example that I get from my whoop um I if I can throw that into an AI model it can make predictions about what is good for me or what I might want to do with my health that I can't make unless I have that data and I I don't think any company should be able to keep that away from me yeah I mean this is an has been a major major part of our firm's thesis for 10 plus years when it comes to like web 3 applications which is composable data that can exist anywhere you want it to exist and that blockchains can enable that but you know I think there's a long way to go on sort of the ux of a lot of those applications so in the meantime let's push the big tech companies to make our data at least available for research purposes you're here to hear first folks Grassroots movement data trusts um um cool we're we're getting we're getting close to time here um and I there was one thing that you showed me before we started recording that like blew my mind um and I would just want to make sure that we get some time in to uh to show it to people um so set the scene for us what are we about to see sure so um I am actually sitting in front of like a 12 something maybe I don't know how big it is a really big television and hooked up to this television there's a bunch of different things but one setting is hardwired directly into a piece of Machinery that is locally running a visual diffusion model um that is connected to the whisper API that listens to what we are saying and then in real time generates moving images based on what we're talking about it's called the dream machine and it was developed by an artist who I only know as human I think it's hu m i n you can follow him on Twitter he was a member of a group called the bright moments Dow where he developed this which was um an investment of ours and this kind of just sits in the background in our office from time to time and it's one of the coolest things because it is one of the few things that we have here that both like visiting CEOs or whoever or other investors will come and have their minds blown but then also I like very often catch the building super bringing his friends up to kind of play with it so I'm going to turn it on the one caveat I'll say is that you know this thing tends to get pretty backloaded so sometimes it's a little slow in terms of its responsiveness but let's just throw it on and see what it's doing anyway great perfect there we go o we've got an aquarium okay like what I want to know so I'm just going to name some things and see what it does like I've been reading Moby Dick and I it's really good I I actually just finished it and I just love um I love all the whale uh uh an oceanic iconography you know he gets really into the whaling industry it's like it's like such a detailed account of of how whales work and I guess we got the we got the aquarium so we're getting close to whales well I what I was going to say is every once in a while you have to reset this things memory because it becomes so backlogged with things that people have talked about in the past that it can sometimes take a minute or so before what you're talking about actually shows up but there is our whales here they come look at that that's so cool dude like yeah if we by the way if we had wiped its memory before we started like those whales would have just popped up immediately um and one one fun thing I'll say is that um we had some event here with Senator Gill brand and she allowed us to have this going in the background while she was talking so it was basically transcribing a senator's speech with Vis uh you know visual imagery so here's what it reminds me of is is um the way fiction works is when you're describing the setting around a character typically the the way that you describe it will be a reflection of the character's internal State and that's why um it's it's like a cliche is that like when something bad happens in a movie like it's raining outside like it always reflects how they how they're actually feeling and you can create this real sense of drama uh and and and and concreteness and that's Jill brand behind you which is crazy I'm sorry um this this real sense of of like drama and um and visceral like I'm there kind of in connection to the character with that right and that's what this feels like to me it's reflecting this in our internal States and our conversations in this visual way that reminds me a lot of fiction yeah um well I would just say compliments to the chef like I I but I also think that it is I think this is one of the the the first like truly native AI artistic art forms that could not exist in a world uh prior to AI yeah um which I think makes it really fun and um you know obviously there's cats coming up and uh I don't know it just like oh jeez uh scary I know I have such a hard time communicating when this is on because I'm just looking at it I tried to have a serious meeting here the other day with someone while this was running and it did not go well but I don't know like bring your ideas to life I feel like that is kind of what the lesson of of of the dream machine is and sort of my perspective on everything that's happened over the last six months of my life is just like just make stuff because the tools are there to do it yeah dude this is that's exactly how I feel I love it um and I think that's a great place to end uh thank you so much for coming on and showing us all the stuff uh it was amazing no problem and by the way if we meet again in a year My Hope Is that the librarian will be here with us because that is what we're working on let's book it I like I definitely want to have you on uh in the coming months and years cool thanks so much Dan oh my gosh folks you absolutely positively have to smash that like button and subscribe to Ai and I why because this show is the epitome of Awesomeness it's like finding a treasure chest in your backyard but instead of gold it's filled with pure unadulterated knowledge bombs about chat GPT every episode is a roller coaster of emotions insights and laughter that will leave you on the edge of your seat craving for more it's not just a show it's a journey into the future with Dan shipper as the captain of the spaceship so do yourself a favor hit like smash subscribe and strap in for the ride of your life and now without any further Ado let me just say Dan I'm absolutely hopelessly in love with you