Tae Hyoung Jo

I don't know much about finance, but I have been obsessing over tools we use for gathering and building knowledge both personally and in organizations for a while... Namely, how might we enable frictionless knowledge capture and be able to realize patterns of information? How do we enable collaboration between groups of people with different mental models, and be able to make sense of information in a streamlined way? AI has gotten us halfway there, but for a while it has always felt like no one was capitalizing on this in a first class experience. Too much retrofitting AI assistants. Existing tools are either too rigid or just "not helpful enough" to warrant moving away from a simple chat interface. This lead up is all to say, Village just might be the knowledge capture tool of my dreams. Where collaboration with AI assistants feels natural. Where sense-making doesn't feel like dread. The difference on the end interface is subtle, as most of the innovation lies in the primitives and tools the assistant works with, but I'm confident within seconds of using the product you'd be able to tell the leapfrog improvement in experience. I would highly recommend signing up for early access, even if you're not necessarily in Finance! --- Anyway, congratulations to an incredibly dedicated team: @john_sungjin, @_newhaiku, and Tae. <3

Today, we're launching Village to let investors orchestrate teams of agents to scale their judgment. Three years ago, we bet that both humans and agents would need new tools to fulfill the promise of LLMs to transform research. Village is that tool, the first IDE for research built from the ground up so your agents can answer bigger questions, learn from your feedback, and build the AI research infrastructure for your firm. We're starting with public equity research, enabled by our EDGAR and earnings call transcript datasets, and we'd love to hear what use-cases you'd like to see next. The feeling of manipulating huge amounts of information and collaborating with hundreds of Assistants is really special, and I'm excited for you to experience it. Our beta sign up link is below!

We realized this a year ago and bet the rest of our runway on the implications - parallel agents are coming, we can design for them now, and investor research tools will hit PMF. Our thought process: Once companies can trade tokens for work, the next question they'll ask is: how can we spend those tokens most quickly and effectively? And they'll throw money at the products that answer it best. The winners will be: - Parallel multi-agent: single agent loops are bottlenecked by token generation speed and context size. - Asynchronous + fully autonomous: one person can't Starcraft-micro 1000 agents, and every need to intervene is magnified 1000x. LLMs can't work and collaborate like humans yet. However, soon that'll change, and the bottleneck will be product. How do you build an environment that maximizes an agent's utility? What primitives do they need to collaborate? What's the right UX to manage them? The core idea: the agent is a first-class user of the app. Calling isolated tools + logging results doesn't scale for complex multi-agent work. Agents need an environment to control their own contexts and benefit from the same UX patterns that enable humans. For our human users, this means more automatable tasks and more legible results. When it all works, the first big market will be research tools for investors. In software, single-agent is enough for PMF because it's harder to get value from scale; an extra SWE on your team doesn't guarantee your feature ships faster. With research, scale is everything. The task "get the # of US DTC subs for NFLX this quarter" can easily be scaled to more datapoints (ARPU, management commentary), companies (DIS, WBD), and time (20 quarters), which enables new insights like comparisons and trend analyses. Investors will be the earliest adopters while LLMs are expensive: there's too many questions to ask, alpha can mean millions in returns, and compute is more elastic than headcount. Effective token spend will become a measure of competitive edge. And, in time, inference costs will come down, and the tools that help investors do research will be accessible for everyone. A year later, we're still super early but it's finally starting to play out - the models are good enough, and we've built the product for them.

.@_harshalsheth and I were building semantic search and wanted to test different embeddings strategies + models, but there aren't any good playgrounds for quickly iterating and building intuition so we built it! embed.sheth.io

We got 🥈 at the @AnthropicAI hackathon by impersonating our favorite anons and judges - but we're really excited about the idea of using LLMs to Simulate Everything! So, we created Twitter (we can legally call it that now @elonmusk ). 1/n (❤️ @kvvnhu @taehyoungjo @thetejmahal)

Brb going through photos of the audience

7/ Villagers Building twitter “focus groups” by fitting active user tweets into Claude’s context window and modeling their responses to controversial tweets @john_sungjin @kvvnhu @taehyoungjo @thetejmahal 🥈 2nd place winners

@FundamentEdge Dm-ed and excited to chat!

Also, to the extent the feasibility study and prototype are promising, a co-development partnership with a large asset manager (who has existing APIs and internal data repository) seems like the right path. If that is you, please reach out via DM or brett@fundamentedge.com

CHAT GPT FOR STOCK PICKERS? Throughout the course of my 15 years as a stock-picker I've always erred on the side of being a luddite. To me, developing a great stock idea is an artisanal process and the same way a painter might find joy from cleaning his brushes,

I'm surprised by how much I love using quacktype.com by @kvvnhu. The rare kind of application where you go from "oh this is nice" to "you can pry it from my cold, lifeless hands" after just a few days.

Jailbroke all the ChatGPT plugin prompts (github.com/john-sungjin/c…) and wrote some thoughts. TL;DR: - Plugin specs are injected as Typescript - Improving ReAct/@langchain agents - A new SEO industry for LLMs (LEO?) - Companies are thinking beyond OpenAI

You can jailbreak @OpenAI's plugin chatGPT if you want to see plugin code. ai-plugin.json specs are not available for every plugin, but chatGPT will just tell you!

1/ A preview of what we’ve been secretly building at Orchard: LLM agents. By coordinating and prompting LLM systems effectively, we can create highly capable agents that go beyond simple conversational interfaces.

Our beta is out! Chord - an AI engine that does real-time research & serves recommendations for products, books, and more. Imagine an on-demand, crowdsourced Wirecutter: we scour the web for organic discussion on topics and then compose in-depth articles. chord.ooo

Today, Momento announces the first #AI that streamlines #podcast promotion What can Momento AI do? 📱 Produce 30s-1min video shorts 📹 Create 5-10min segments for YouTube 📝 Write summaries, chapters, takeaways Try it: studio.momento.fm/playground Examples 👇 1/16

Congrats on the rebrand! 📈

@EryDayImRusslen Try a jigsaw puzzle. Preferably with friends. First time I felt flow state in months.

I have been unable to get into a “flow” state with anything for 3-4 months. It is slowly driving me crazy

tucked some apples (that's my new catchphrase) and got 1st at the @scale_AI hackathon with @evanon0ping (the man/boy is a genius) and a guy named David Yue (2700 elo) project is self-explanatory u do the backend with gpt: github.com/TheAppleTucker…

I increasingly see GPT‑3/LLM prompts as assembly code, not as human interface. We shouldn’t be writing prompts, but prompt compilers. A template string is not a moat.

Thrilled to announce @ChestrApp's iOS app, allowing users to easily save items across the web with 2 taps + get notified of discounts! Think of it as a sleek version of Pinterest + Honey for mobile! Link to the app: apps.apple.com/app/id64440904… Product Hunt: producthunt.com/posts/chestr-2

The Villagers are growing an Orchard! 🌱