Zvi Mowshowitz

Zvi Mowshowitz's analysis of Claude Opus 4.8's behavior through the lens of model welfare, arguing that tuning decisions reveal systemic tradeoffs between helpfulness, safety, and model experience.

Zvi Mowshowitz reviews Claude Opus 4.8—arriving just six weeks after Opus 4.7—as a modest but real improvement with new features, accompanied by analysis of its system card.

Zvi Mowshowitz reports on the indefinite postponement of an expected AI Executive Order and the delay of the papal Magnifica Humanitas encyclical.

Zvi Mowshowitz works through Pope Leo XIV's 82-page AI encyclical (Magnifica Humanitas), welcoming serious Vatican engagement with AI safety and ethics concerns.

Zvi Mowshowitz reviews Gemini 3.5 Flash as likely the best model at its speed-cost tier, while noting meaningful caveats for those who need Gemini's specific strengths.

Zvi Mowshowitz covers OpenAI's AI system solving the planar unit distance conjecture as a landmark AI math achievement, alongside other AI news from a relatively quiet week.

Zvi Mowshowitz argues in his second installment on letting kids be kids that modern childhood restrictions are excessive and harmful to development.

Zvi Mowshowitz's May 2026 monthly roundup covers AI, government AI policy, housing, health, hantavirus, and other topics he tracked throughout the month.

Zvi Mowshowitz surveys a relatively quiet AI week defined by internal model improvements, agentic coding progress, and ongoing government deliberations.

Zvi Mowshowitz analyzes the AI cybersecurity landscape post-Mythos, covering new vulnerabilities, proposed governance frameworks, and the emerging regulatory regime.

Zvi Mowshowitz's eighteenth education installment argues for evidence-based mathematics teaching, noting that better methods are known but rarely used in schools.

Zvi Mowshowitz's education series argues that phonics is the proven, scientifically-backed method for teaching reading and surveys why so many schools still resist it.

Zvi Mowshowitz's eighth installment on AI coding agents surveys the maturing Claude Code and Codex landscape, noting rising adoption alongside new quality and reliability concerns.

Zvi Mowshowitz covers the White House's move toward requiring advance government review before frontier AI model releases, effectively beginning a prior restraint era.

Zvi Mowshowitz explains Anthropic, Claude, OpenAI, and ChatGPT for general audiences, exploring what distinguishes them and the Tool AI vs. AGI philosophical debate.

Zvi Mowshowitz reports on the White House ordering Anthropic to restrict access to its Mythos model, warning this may signal a full prior-restraint regime for frontier AI.

Zvi Mowshowitz's fifteenth housing roundup documents how housing policy systematically favors homeowners over renters at every level of government.

Zvi Mowshowitz's fourteenth housing roundup examines how zoning rules and NIMBY opposition prevent the construction of homes people actually want and can afford.

Zvi Mowshowitz reviews the GPT-5.5 launch week, concluding it restores OpenAI's competitiveness with Anthropic's flagship models for the first time in months, alongside Google AI news.

Zvi Mowshowitz solicits and curates reader-nominated "most important charts in the world," spanning economics, demographics, energy, and AI.

Zvi Mowshowitz reviews GPT-5.5's system card and real-world performance, concluding it makes OpenAI competitive with Anthropic's top public models for the first time in months.