397-Billion-Parameter AI Runs on iPhone: Video Explains MoE Design and Limitations
DJTRENDSETTA · reddit · 2026-08-16
A video shows a 397B-parameter AI model running offline on an iPhone, explaining the mixture-of-experts design (96% dormant), Apple's predictive paper from three years ago, flash-moe team's cache deletion speeding up 38%, and limitations in speed, storage, and heat.
Related event: 397B Parameter AI Model Runs Offline on iPhone(2 posts)→
More from Embodied
- AutoAgent0: VLM-powered agentic runtime lifts long-tail driving success by 28.3pp — yining_hong · 2026-10-16
- CarryOn: dual-loop harness adds personalization and safe recovery to frozen VLA robots — Zixuan Wang · 2026-10-16
- FoveaVLA replaces wrist cameras with learned foveation, hitting 70.8% success — Haoyu_Xiong_ · 2026-10-16
- Musk teases two-way Grok x Optimus integration: Grok in the bot, bot in Grok — beffjezos · 2026-10-16
- SimVTA uses simulator-rendered video to post-train driving world action models with RL — OpenDriveLab · 2026-10-16
- Robotics data startups compared: Scale AI tops $1B revenue as 9 LLM vs robotics data suppliers benchmarked — FinanceYF5 · 2026-10-16