FULL STORY
Malik vs Isola: Can LLMs Run Robots?
UC Berkeley's Jitendra Malik challenged LLMs' role in robot control, prompting MIT's Phillip Isola to respond that layered agents writing code can close the gap, escalating the embodied-AI debate.
2026-09-08 ~ 2026-09-08 · 2 episodes · 6 posts
Episode 1 · Malik and Isola Debate Whether LLMs Can Control Robots (2026-09-08, 4 posts)
Berkeley's Jitendra Malik argues LLMs suit only high-level planning, not dexterous low-level control, while MIT's Phillip Isola counters that combining LLMs with high-frequency controllers could achieve dynamic dexterous manipulation.
- Isola on Malik's challenge: dexterous control unsolved, but high-frequency controllers + LLMs can do it — phillip_isola · 2026-09-08
- Jitendra Malik: Using LLMs for low-level motor control is perversely inefficient — JitendraMalikCV · 2026-09-08
- Robotics veteran Jitendra Malik challenges LLM robotics hype with a dexterity benchmark — JitendraMalikCV · 2026-09-08
- Jitendra Malik pushes back on Isola: LLMs can't handle high-frequency robot control — sainingxie · 2026-09-08
Episode 2 · Embodied AI debate escalates as Isola and Chalenti respond to Malik (2026-09-08, 2 posts)
MIT's Phillip Isola responded to Jitendra Malik's embodied-AI skepticism, agreeing on layered control but differing on where the boundary lies, while Georgia Chalenti countered that embodied adaptation cannot be assumed inherent in VLMs.
- Phillip Isola: LLM agents won't act alone, they'll call low-level controllers via code — phillip_isola · 2026-09-08
- Embodiment doesn't vanish above low-level controllers — the Isola assumption is unproven — GeorgiaChal · 2026-09-08