Name: The Simulation
Uploaded: 2025-02-01T19:07:26.863Z
Duration: 2 min 33 s
Description: A simulated reality of AI-powered characters aiming to create artificial general intelligence.

the-simulation

A simulated reality of AI-powered characters aiming to create artificial general intelligence.

The Simulation

Skill to action generation for agents or saga is a generative AI framework that steers AI agents toward successful goals through actions. An action is an implementation of a skill. Skills are the tools in saga's toolbox. Given a sheriff agent whose goal is finding a criminal, Saga may use the go-to skill to enter the saloon and the converse with skill to ask the bartender about suspicious patrons. 

Saga is a bring your own simulation AI framework, an agnostic to the Sim and Sim specific skills it's used.

With agents first tell Saga contextual metadata about themselves and the world via companion simulation who they are what they know and what skills they have. Let's explore how Saga works in practice using our Thisle Gulch simulation, a fictional Wild West Town in late 1800s. This agent is Blackjack, the saloon owner and local gang leader. The murder of a local native is threatening his plans to rob a gold shipment in a few weeks. His goal is to get the heat off of him and his gang by any means necessary.

Sheriff y Cooper on the other hand another agent is trying to solve the murder and avoid false leads setting up a conflict between the sheriff and the saloon owner. 

Which will be the context Saga uses to generate actions during this demo. These strategies were not prescribed ahead of time but through Saga are the natural consequence of these agents backstories and the scenario they find themselves in. 

When an agent is deciding what to do next Saga generates a set of actions. These action options are then scored and return of the simulation in order to direct the agent toward its goals. Even though Saga rated each action with a score, the agent can still choose any action, increasing the opportunity for variation and human feedback. This process repeats each time the agent is deciding its next action and can be scaled to multiple agents running simultaneously. 

For blackjack, Saga will provide options that leverage his two loyal lieutenants, Sally McCarty and Razer Jim Donovan. Here, Saga used the talk to share AI Agent Store with colleagues.

Skill to generate the action. Talk to Sally Mccardy about the dead native situation to test her loyalty and gather information. After the actions completed, Saga generates a new list of options. Here it generated six options: two using the talk to skill, one using the reflect skill, two using the interact skill, and then one using the weight skill. The sheriff's agent working autonomously has gotten a lead that may point back to the gang. While Blackjack's first strategy was to lay low and wait, he now decides.

They need to frame someone else for the murder because these skills like tools can be used in many different ways. The number of unique actions they can generate is very large. After some time has passed, Blackjack checks back in with Razer. Razer acknowledges that he was interrupted while playing cards because the simulation is tracking Razer's activities. Razer updates Blackjack that things are going to plan, and Blackjack encourages him to keep at it. Razer returns to playing cards, and since Saga offers.

That option for blackjack as well. We join him. The sheriff awakens on day two and immediately gets to work tracking down some leads. After the sheriff gets a tip on a suspicious man who bought a hunting knife, Blackjack talks to the sheriff in order to direct him away from Razer. But the sheriff is suspicious. With their plans on the line, Blackjack decides to step up their efforts. He tells Razer they need to plant some evidence in order to lead the sheriff out of town. Razer uses his take to skill and goto.

Skill to steal some evidence and pl it at a campsite outside of town. We're excited to announce Saga skill to actions for generative agents is open source and now available. Please check the launch announcement and GitHub repo for more information.

Note, that there may be slight differences in text and video. Watch video for the original content.

The Simulation Video