Latest / The AI Practitioner Brief
Anthropic's $517B Bet, MMLU's Cracks, and XPeng's Robot — September 8, 2026
More than five hundred seventeen billion dollars in locked-in compute. A benchmark score that can drop fourteen points just by swapping the question set. One score, two different tests. We go through how AI is collapsing the cost of finding and exploiting security bugs, why XPeng's humanoid robot beat Tesla's Optimus off the line, and the smaller stories: OpenAI's coming Managed Agents, ChatGPT reading your email to learn your writing voice, and ByteDance's real-time spatial video model.
Get full access to The AI Practitioner at aipractitioner.substack.com/subscribe
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.