Debate over Anthropic safety evals: Accenture vs METR independence questioned
GaryMarcus · x · 2026-09-20
Gary Marcus amplified a thread on the independence of Anthropic's safety evaluations. The argument: Accenture is more culturally independent from Anthropic but less independent in terms of incentives — METR is a nonprofit whose mission requires transparency even when findings hurt its partner, while Accenture is a business that wants its clients happy. The debate reflects broader scrutiny of vendor-commissioned safety evals.
More from Companies & People
- Leaked research shows Google AI Overviews cut clicks without improving quality — lilyraynyc · 2026-09-20
- Perplexity CEO: Open-Source Will Cheapen Frontier AI, But New Workloads Will Sustain Demand — rohanpaul_ai · 2026-09-20
- 'GPTs are GPTs' Author Daniel Rock Joins Google to Lead AI x Economy Research — soumitrashukla9 · 2026-09-20
- Reddit Accuses Design Arena of Anti-West Hype While Pitching US VCs — entsnack · 2026-09-20
- Flock reportedly pushes employee buyouts to avoid layoffs — TechCrunch AI · 2026-09-20
- If You Don't Ship a New Model Every Week, You're No Longer a Frontier AI Lab — bendee983 · 2026-09-20