Musk says Grok 4.8 is in RL phase, built for 'usefulness max' over benchmarks
XFreeze · x · 2026-11-19
Elon Musk detailed Grok 4.8's progress: the model is in its RL phase, optimized for "usefulness max" rather than benchmark chasing, which he called a fool's errand. He compared the approach to Tesla FSD: trace every intervention to wrong, biased, or missing source data, fix it, and retrain — aiming for zero crashes, not a test score. Versions are quietly tested under the hood of Grok Bot via periodic soft releases, creating a data flywheel that measures whether users' problems are actually solved.
More from Companies & People
- Kalanick vows to rally Nvidia walkout against 'AI overlord' with a 'wafer-encrusted' megaphone — beffjezos · 2026-11-19
- Full video of Elon Musk's talk at TomorrowX just released — PTrubey · 2026-11-19
- Travis Kalanick backs Jensen Huang as the 'number one guy' standing between us and an AI overlord — kevinnbass · 2026-11-19
- Musk calls NVIDIA removing Jensen Huang as CEO a 'huge mistake': 'Who wouldn't want to hire him?' — XFreeze · 2026-11-19
- Median OpenAI researcher burns $2,400 in tokens daily — $876K a year — kevinroose · 2026-11-19
- Marin releases checkpoints after largest-ever live open pretraining run — Diyi_Yang · 2026-11-19