great to see everyone at AI rabbithole event today!

as promised, here's my deck with abbreviated narrative.

(video was recorded, will share when available) agents agents agents!

40+ agent related github projects today with 500+ stars, 90% of which are less than a year old.

Been talking lots about knowledge graphs and LLMs, specifically for autonomous agents, recently.

Organized some thought here if you want to dig in 👇

KG 🤝 LLM

🧵 (a thread) For those not familiar, knowledge graphs are a type of data representation in the form of nodes (objects) and edges (relationships) and looks like this.

Open-sourcing "MindGraph", a proof of concept starter kit for building and querying against an ever-expanding knowledge graph with natural language!

- multiple input types
- simple interface
- basic search/RAG against knowledge graph
- easily extendable with integrations
- basic database swapping

GitHub in 🧵 Quick background is that I paused on BabyAGI work after the BabyFoxAGI mod because I felt I was missing some pieces.

Went deep on knowledge graphs (KGs) and released Instagraph () which lets you build, merge, and edit KGs....instagraph.ai

Okay @tldraw is pretty addicting



1) draw your UI
2) use text to describe
3) highlight (i.e. draw box around), and
4) click the blue "make real" on top right) makereal.tldraw.com
super quick prototypes that don't actually work, but give you the sense of what it would feel like to play with? count me in!

The Assistants API is awesome, had to build something...

Open-sourcing "GPTvsGPT" 🕵️🕵️
(109 lines of code)

A simple template to have two AI Assistants converse. What's exciting is the ability to extend these with retrieval, data, and custom functions.

GH/Replit in thread 🔽 As you saw above, you set the parameters for two Assistants, then feed that in with the topic, and number of messages you want.

You'll get something like this:

famous songs as knowledge graphs

obsessed w this new learning tool

need some time before sharing the tool but in the meantime… any requests?

Introducing BabyFoxAGI 👶🦊🤖

Newest mod of @babyAGI_ with:

🔂 Self-improving task lists (FOXY method)
💬 Novel Chat UI w parallel tasks
🛠️ New skills (DALLE, Airtable, etc)
📄 It’s own README

Breakdown, Github, Replit 👇 BabyFoxAGI is mod of BabyElfAGI, in a series of mods w the naming of Baby<animal>AGI in alphabetical order:

🐝Bee, 🐱 Cat, 🦌 Deer, 🧝 Elf, 🦊 Fox

You can find the thread introducing BabyElfAGI here: x.com/yoheinakajima/…

Introducing the "FOXY method" 🦊

Our new approach to self improving auto agents, introduced in the next mod of @babyagi_.

Dig in shall we? 👇 How it works:

With FOXY (Final Output eXamination from "Yesterday"), we do a final reflection on the output of each run, and use this to guide future runs, pulling most relevant reflection using a similarity search, paired with a decay mechanism to prioritize recent reflections.

Meet BabyElfAGI!

Biggest update to the OG @babyAGI_, finally broke out into multiple files:

🛠️ Skills Class allows for easy skill creation
📝 Dynamic Task List example w vector search
🔍 Beta Reflection Agent
🖇️ Can read and write own code (yes it can write new skills and task… https://t.co/mN6jHZydwqtwitter.com/i/web/status/1…
Skills Class

New “Skills” (prev called Mini Agents) are what the task execution agent uses to complete tasks. Enabled in main files handles APIs centrally.

Each skill can be as simple as a simple API call, or a long script w many functions.

I get asked these often, so thought I’d write up a thread.

Where are we with autonomous agents? What can they do? Should we use one?

My thoughts 👇 A general autonomous agent, that can be given any reasonable objective and accomplish it well… well, we’re far from that. (1~1000 weeks)

Don’t expect to use one today or tomorrow.

testing latest @babyAGI_ mod...

"Look up AI startups, present 5 to the user to have them pick one, and then do further research and provide an investment memo on that startup." searching and scraping...

Meow!!! 👶🐱🤖 Check it:

OG @babyAGI_ mod #3 - BabyCatAGI

- Higher completion of objectives
- Lower errors
- Faster (GPT4 task creation runs once)
- Hand-crafted Mini agent as tool

✨~300 lines of code

Let’s dig in 👇 As a quick recap, OG @babyAGI_ is the original commit, combining three agents in a loop (task execution/creation/prioritization), only using GPT-3, and designed to never stop.

~105 lines of code.

https://twitter.com/yoheinakajima/status/1640934493489070080

Introducing BabyBeeAGI 👶🐝🤖

A buggier, slower, more powerful mod of OG @babyAGI_ at ~300 lines of code.

✅ Powerful Task Manager
➡️ Task serialization
📋 Local task queue
⚙️ Tools!
🔚 Stops! (usually)
😔 Requires GPT-4

"Paper", Code, & more 👇 Biggest difference is combining the task creation agent and task prioritization into one task manager agent.

We also added task serialization, tools, and tracking complete/incomplete tasks all in the task manager.

Even with GPT-4 we're pushing the limit.

Compare the prompts:

Woo hoo, web scrape! Then analyzed the contents by passing results between tasks!

Another test, shall we?

This one should do web search, pass results between relevant tasks, and finally stop at the end. Task list looks solid, and it's doing a web search...

Let’s try another test! Interestingly did more than generate a task list. The execution agent is still using GPT-3, and still using MVP (minimum viable prompt) there.

Playing around w combining the task creation and prioritization agent into one task manager agent who assigns the tool to use and tracks completed tasks…

No idea how to merge this 😅

(Also it only works w GPT4 and is much slower) web-search tool seems to be working!

(Note: is an unreleased version I’m still working on)

🧵 Introducing AsymmeTrix: Asymmetric Vector Embeddings for Directional Similarity Search! 🔍

Our new approach to vector search & embeddings that models asymmetric relationships between data points.

Github: github.com/yoheinakajima/…

👇Read on for key insights! 🚀🔍 1/ Traditional vector embeddings rely on symmetric similarity measures like cosine similarity.

They're great for many applications, but can struggle with asymmetric relationships between objects.

Enter AsymmeTrix! 🔄

Kinda like prompt engineering “You are an AI assistant that always response like an immature 6 yr old”

The response to @babyAGI_ has been beyond my imagination…

So I'm bringing in help!

To start, we have @Fraser joining us to help lead, with support @sparkcapital, @Replit, @trychroma, and more partners to be announced soon!

A bit more on the vision 👇 Baby AGI was a simple idea, presented as code - but it sparked imagination!

We quickly saw people driving this simple approach in different directions, and I believe the discussion here is where the magic happens.

We have a lot of work and learning to do together.