Let me say something that might offend people: Anthropic is warning that AI is too dangerous and we need to slow down, while the capabilities of its own models are accelerating at nearly twice the rate—I'm not making this up; it's from an independent assessment METR just published on November 5: Anthropic's internal capability index is growing at about 1.9 times its previous rate (95% confidence interval: 1.5–2.8 times), meaning it did in one year what used to take two, largely by using AI to conduct its own research and development. Growth in training compute can't explain the acceleration. Even more ironically, the report says Anthropic redacted most of the crucial evidence. In the same week, Cat Wu, head of product for Claude Code, had her last day last Friday (October 30); on November 5, it was announced that she'd joined OpenAI to lead Core Products—and three key people from Claude Code have left in ten days. On the eve of its IPO, capabilities are doubling while key people are walking out—which side do you believe? #Anthropic #AI