World Model Survey Paper Review, Part 2 — Models & Their Path to AGI: JEPA, Dreamer, Genie, Sora, World Labs and more
About Event
TL;DR This is a beginner-friendly peer-led “paper clinic” where we roast + rebuild the latest World Model survey paper: “Understanding World or Predicting Future? A Comprehensive Survey of World Models” (arXiv:2411.14499, Last updated by Dec 2025). You’ll leave with a clean mental map of the field, core architecture patterns, and the top open problems worth tracking (or building on).
In this session, we’ll connect the paper’s taxonomy to the most talked-about systems and research lines: JEPA V-JEPA (representation prediction), Dreamer (latent imagination for planningcontrol), Genie (interactive environment generation), Sora-style video models (large-scale simulation), and World Labs (spatial intelligence + editable 3D worlds).
Note: We will finish this paper in 2 weeks:
• Session 1 - Feb 28: Foundations knowledge in World Models, the "Mirror vs. Map" debate (Generation vs. Representation). (https://luma.com/lir5i9w7)
• Session 2 - March 7: Competitive landscape (Sora vs. Cosmos vs. V-JEPA) and future AGI trends like social intelligence. (https://luma.com/tckgubqm)
What we’ll cover (in order)
What “world models” are and the paper’s key split: internal understanding vs future prediction/simulation
The shared architecture pattern: representation → dynamics → prediction/simulation → planning
How the survey’s categories map to real systems (JEPA, Dreamer, Genie, Sora, World Labs)
Learning requirements
• Beginner-friendly. Ok to come with no math or ML background — we’ll keep it builder-friendly
Session Setup
• 30-20-30 Workflow. Each session is a collaborative relay: a 30m high-level intro (Ti), a 10m active self-research + 10 group discussion break, and a 30m expert deep dive + Q&A (Colby).
• Multi-Level Support. Every slide contains targeted sidebars for Beginners (concepts), Builders (industrial implementation), and Researchers (mathematical core) to ensure value for all experience levels.
• Self-study with Group discussion. Participants will get to spend the middle 20 minutes using NotebookLM to personally verify concepts, identify "Causality Gaps," and prepare for group-led debates.
Pre-class (optional)
Paper URL: https://arxiv.org/pdf/2411.14499
About AI Scholars
A peer-led learning journey for engineers, students, researchers, and builders. We read new papers, challenge the hype, and turn ideas into practical mental models you can use immediately.