Robotics experts push back on GPT-7-Robot hype: LLMs are 100x larger than VLAs
silver__tsuki · x · 2026-10-29
A robotics-ML debate sparked by a speculative take that OpenAI could eventually ship "GPT-7-Robot" — a general-purpose model post-trained heavily on robotics data, running as fast as VLAs while solving 95% of tasks. The rebuttal lays out three constraints:
- Not all robot inference can happen in the cloud; on-device compute is a hard limit
- Current LLMs are 100x larger than the best VLAs, so scaling up directly is impractical
- In many robotics use cases, intelligence isn't the main bottleneck anyway
The exchange captures the core route dispute in embodied AI: cloud-scale general models vs lightweight on-device VLAs and non-intelligence bottlenecks (hardware, cost, reliability).
More from AGI Musings
- GPT-6.1 computer use costs $0.16 per human-minute task, an effective $9.60/hr — bytebot · 2026-10-29
- When agents outnumber humans, their preferences will steer the economy — lfschiavo · 2026-10-29
- Millions of tweets, zero scenarios Shulman and Bostrom didn't already think through — abhiadesai · 2026-10-29
- AI Software Factory vs Dark Factory: two stages of autonomous software delivery — Pavan_Belagatti · 2026-10-29
- Is Google's claimed use of RSI the start of the singularity under Vinge's and Good's definitions? — Extraaltodeus · 2026-10-29
- 'AI is just a tool' misses the point: writing and print reshaped humans too — YogeshMalik · 2026-10-29