Ted Moskovitz (Member of Technical Staff at Anthropic) RAAIS 2026

Scaling AI models is a rigorous scientific process requiring epistemic humility, controlled experimentation, and a focus on cost-to-benefit ratios to improve model performance. Ted Moskovitz, who leads the Science of Scaling team at Anthropic, explains that modern large language models are increasingly capable of generalizing and performing complex tasks, shifting the focus from simple benchmark optimization to product-integrated utility. The integration of safety research and alignment techniques is essential for creating usable, harmless systems, as these guardrails enable the feedback loops necessary for rapid development. While the field remains experimental, the ability to leverage test-time compute and improve model "taste" in research directions suggests a trajectory toward more autonomous, high-impact AI. Ultimately, the industry is moving toward a future where larger, more capable models handle complex, multi-step tasks, significantly reducing the need for human intervention.