OpenSimWorld is the directory of 3D Virtual Worlds based on OpenSimulator and connected through the HyperGrid. Learn More
New Members
RomanoRusso, JadeJamison, Udyora22, Hydra Knight, VengenceisMine, Mystique, Sorrow, OliverBeckett, TENACIOUSLUV, MetalBear, Topservices, AlanFrench, OoSOULEoo, KylaVixen, Pamella, Desiggner, Dash73, Alifuleeh, ZiggyAlcott, SweetReaperLikes
Introducing AI Atlas, an AI powered NPC. AI Atlas was designed to run with different LLMs (Large Language Models) and make user-to-NPC, or NPC-to-NPC chat is little less boring that it use to be. Emphasis on short, fast and casual answers rather than the pedantic parrot LLMs tend to be. Work in progress as we are still discovering how we can infuse somme intelligence into NPC's behaviour. On display tonight (April, 7) at atlasgrid.fr:8002:Atlas-Studio.
Première présentation ce soir de AI Atlas dans nos studios.
Première présentation ce soir de AI Atlas dans nos studios.
Liked by: 
Comments
For the tech-curious :
- There is no script in the NPC.
- A LSL relay (simple box with a script) listen to the chat and relays it via http to the inference server, a special-built machine with many graphic cards running Debian headless.
- A proxy running on the inference server accepts the http request, relays it again to ollama service.
- ollama service bounce the response to the proxy, then the proxy back to the LSL relay.
- The LSL relay calls osNpsSay.
- Total : 2 scripts (1 LSL, 1 Perl)
ollama can run a variety of LLMs. They may be customized with a prompt prefix governing role, mood, response size, etc. Not a first. There is a NPC to chatbot @ Ferd Frederix website (Outworldz) and maybe others. But this one inherits ollama's flexibility an openneness https://ollama.com/
like(1)
- There is no script in the NPC.
- A LSL relay (simple box with a script) listen to the chat and relays it via http to the inference server, a special-built machine with many graphic cards running Debian headless.
- A proxy running on the inference server accepts the http request, relays it again to ollama service.
- ollama service bounce the response to the proxy, then the proxy back to the LSL relay.
- The LSL relay calls osNpsSay.
- Total : 2 scripts (1 LSL, 1 Perl)
ollama can run a variety of LLMs. They may be customized with a prompt prefix governing role, mood, response size, etc. Not a first. There is a NPC to chatbot @ Ferd Frederix website (Outworldz) and maybe others. But this one inherits ollama's flexibility an openneness https://ollama.com/