Inspiration
what inspired us to build simulacra was this now famous paper on using generative agents to create a human simulacra. we realized how important this could be for orgs that don't wanna eat the expenses of running real social experiments.
What it does
simulacra lets you build + deploy ai agents to create custom simulations.
How we built it
Dylan: it was my first time using python to build a web app haha - my strong suit is Next.js + Typescript - but streamlit was very helpful in allowing us to ship something fast.
Vivek: All Dylan when it came to execution for this demo. Inspired by the Stanford Simulacra Stanford Simulacra experiment ~2 mo ago. I had some very very long shot visions on the backburner since the day it came out and I got the Tweet alert. Basically found a perfect time during AI SF's hackathon to pitch a scaffolded version of these ideas. On Day 1, Dylan and I dived into this project headfirst, focused on an MVP for the hackathon with the goals of long-term vision pursuit after initial hype generated. I offered initial guidance and the vision of the project. We (Dylan) bootstrapped this insanely quickly and launched the MVP: a Streamlit back-and-forth agentic conversation a standard UI chat interface.
Challenges we ran into
the biggest challenge was def trying to build the complex double-agent simulation - we got the simple one down but the complex one had to use vector stores + more complex algorithms. we ended up not being able to ship this in time for the demo - so as of now you can only simulate w/ a single agent or w/ no-long-term-memory double-agents lol.
Accomplishments that we're proud of
we (dylan) shipped something cool! i (dylan) also built my first web app w/ python! will definitely use streamlit again in the future haha.
What we learned
that agents have so much more practical potential than we thought before. it's not all about "how to 10x your productivity w/ autogpt" but there are large-scale problems that can be solves w/ these tools.
What's next for simulacra
next is launching + learning. a lot.
Next will be MORE agents. Sociology- the science of human behavior in societies - is for the first time at a place where its 'untested theories' can scientifically be validated. Let that settle in, and follow along while you think about it 👀
Built With
- framer
- python
- streamlit
Log in or sign up for Devpost to join the conversation.