generative_agents

Show HN: simtown.ai – US Election Simulation with AI Characters Inspired by the Generative Agents paper [0], we've created a browser-based simulation of the upcoming U.S. election with GPT-driven versions of Kamala and Trump, along with nine diverse NPCs representing voters in Pennsylvania. These AI characters freely roam, interact with each other, and engage with real players. Currently, we have a single global instance where everyone can interact together. Single game-day is 24mins in real life. What can you do? - Listen to speeches by the candidates - Chat with NPCs and other players - Persuade others to vote for your candidate - Eavesdrop on conversations between other players by moving closer to them - Explore the environment - Observe the final vote tally once everyone make their mind We appreciate any feedback or suggestions! [0] h

Despite picking up computers at a young age, that was in 1995, nearly 3 decades after this moment. It still blows my mind at what we've been able to do with the power to simulate and visualize everything around us. It's something I do think "modern society" takes for granted - that there's so much about the world that we JUST learned about, or at least finally had a way to translate theories into a format digestible to everyone. I wonder how as we learn to design out large scale AI systems future generations will think about how obvious they are to them but were actually revolutionary to us. For example if the multi-LLM agent simulations we're seeing with Interactive Simulacra[1] will spark new theories about how we should communicate with each other. [1] https://github.com/joonspk-research/generative_agents

Hey HN, My team is open-sourcing the inference stack and fined-tuned models we use to create LLM-powered NPCs: https://github.com/GigaxGames/gigax The generative agents paper [1] pioneered the idea of prompting LLMs to create autonomous NPCs. But existing implementations require multiple calls to an LLM to make the agent plan its day, chat with people, and interact with its environment [2]. Our approach allows NPCs to be stepped at runtime with a single pass on consumer-grade hardware, with reasonable latency. To achieve this, we've fine-tuned open-source LLMs [3] to parse a text description of a 3d scene, and respond with custom actions like `greet <someone>`, `grab <item>`, or `say <utterance>`. This simple whitespace-separated « function calling » format is less verbose than json and thus helps with inference speed. We're also using th

This is a cool project that implements role-playing AI: https://github.com/joonspk-research/generative_agents

If you haven't yet checked out the Generative Agents project referenced by OP, definitely give it a look, it's open source: https://github.com/joonspk-research/generative_agents Over the weekend Lance Martin got it working with local models using llama.cpp and ollama.ai which saves $ on longer sims since all inference happens locally https://twitter.com/RLanceMartin/status/1690829179615657985. It's neat how the AI agents interface with each other – e.g. one will host a party and invites will be sent throughout the group

Related to this post from yesterday: https://news.ycombinator.com/item?id=37073938 Link to a precomputed simulation: https://reverie.herokuapp.com/arXiv_Demo Link to the prompt templates: https://github.com/joonspk-research/generative_agents/tree/m...

Side note -- discussion on adding support for local models is here, along with a preliminary fork that adds support: https://github.com/joonspk-research/generative_agents/issues...