Skip to content

2023 12 28 walk 28 12

Walking notes

Code bounties

  • replit templates
  • tiny grad or tiny box
  • cuda interface for ML
  • space simulator sdk
  • kerbal space - agents

Orbital learning

  1. Define goals
  2. Assign goals to distributed system for federated learning
  3. Define and calculate priority tasks and schedule based on intervals
  4. Handle software updates and data lake maintenance
  5. Utilities ground station resources efficiently. Multiple sources / ground link data sdk
  6. Interface for re inforcement learning - goals, tasks, status/state management

Learning to learn

Agent - 1. read different data sources 2. Identify sources and generate code for loading data 3. Create structured data and summary from sources 4. Build knowledge graph 5. Run evaluation on model for Q and Answers

How to build a Game Engine optimised for Space - Existing Game engines/ OpenSource/ MIT License - Libraries for simulation - Plug and play - Research paper - LLM goal - - Convert equations to code - Create experiments and validate research - Closed loop agents - Read reports from different agencies - Update self map - Read data from Jonathon Space Reports - Create 3dJS map/ Graphic acceleration for voronoi maps - Kerbal Space Engine - Agents for multiple satellites

Build AGI - First build it by hand - Then automate repeated work - Schedule automated tasks for known problems - Add to knowledge graph for new scenarios