Jitendra Malik: Stop Conflating VLMs with World Models in Robotics
JitendraMalikCV · x · 2026-08-24
Berkeley professor Jitendra Malik argues for terminological precision in robotics: VLM, VLA, and World Models are being used indiscriminately. VLMs arose as multimodal extensions of LLMs trained on tasks like VQA, capturing static scene semantics with no dynamics. World Models (LeCun, Ha & Schmidhuber 2018) are primarily dynamics models rooted in control theory (Bellman, Kalman, 1960)—classically written a priori from physics, today learned from video and robot trajectories, but the concept of state-transition modeling is unchanged and shouldn't be mixed with VLMs.
Related event: Scholars Urge Precision on VLM vs World Model Terminology(2 posts)→
More from Embodied
- Stanford's Fei-Fei Li team open-sources OpenWAM, a unified world-action model framework for robots — jiqizhixin · 2026-10-16
- Robot folding leaderboard: GPT-6 Astra leads at 1349 Elo, Muse Wonder trails at 1034 — qinzytech · 2026-10-16
- NVIDIA's TacitVLA: robots reason without reading their own thoughts, 3.4x lower latency — arankomatsuzaki · 2026-10-16
- SEAR: first large-scale controlled study of self-evolving robot agents, 7 LLMs, 330 tasks — xiye_nlp · 2026-10-16
- Podcast: subsea robotics startup Ulysses on operating 1,500m underwater and ocean carbon — blakemenezes · 2026-10-16
- Sony unveils $5,900 gold-plated Walkman with AI music enhancement — Polymarket · 2026-10-16