| Our Plan: | |
|---|---|
| 1. Jarvis Series (Now Retired) | |
| Meaning | Just A Rather Very Intelligent System |
| Start Date | Work on Jarvis began in March 2024. |
| Details | Jarvis Series was split into two parts. The first part, Jarvis AI, had two versions: an initial fine-tuned version of the Qwen 2.5 model with 7 billion parameters, and its final release, v2.0. The second part, Jarvis+, was based on a fully local agentic AI model. |
| Deployment | In our entire plan, Jarvis AI was the only model deployed on the cloud. All other models, including Jarvis+, are designed to run entirely on local machines. |
| Our Plan: | |
|---|---|
| 1. DeepScale Series (Currently Working) | |
| Meaning | Deep Learning Into Very Scale |
| Purpose | To test new AI ideas and experiments. |
| Methods | We are trying different techniques like Retrieval-Augmented Generation, Chain of Thought and Model Context Protocol servers. |
| Versions | There is no fixed limit on the number of versions in this series. |
| 2. Mark Series (Future Plan) | |
| Meaning | Modular Autonomous Reasoning Kernel |
| Focus | This series will focus on agent-based AI. |
| Versions | There will be four versions: Mark I, Mark XLII, Mark L, and Mark LXXXV. |
| 3. Edith Series (Future Plan) | |
| Meaning | Exponential Digital Intelligence Transforming Humanity |
| Focus | This series will use the latest AI technology expected around 2027. |
| Versions | There will be only one version in this series. |
| Description: |
|---|
| Starlink Technologies is making new plans as AI becomes more important. All future AI series will run on local systems and have offline capabilities. We are excited to share our clear roadmap for the future with these new series. Stay tuned for more updates on DeepScale, Mark, and Edith series! |