Overview
- Elon Musk disclosed on September 14 that Grok 4.8, a model reported at about 2.5 trillion parameters, is finishing foundational training this week and is entering a reinforcement learning phase.
- Musk also posted that Grok 5 is the version he expects will reach artificial general intelligence, a high‑impact claim that has not been tested or verified by independent benchmarks.
- xAI says it is training models on a gigawatt‑scale Colossus 2 cluster and using proprietary data from SpaceX operations and Cursor engineering workflows to boost capability.
- Markets reacted to the announcements and renewed safety talk with selling pressure on AI chip stocks, with reports of notable declines in Nvidia and AMD share prices.
- The pace of monthly model releases, prior delays to Grok 4.7, and the lack of peer review highlight execution risks and leave questions about reproducibility, safety testing, and regulatory oversight.