The brief · 9 February 2026

The AI brief, 2026-02-09

Three stories that remind us there's a whole world between the benchmark and the real world.

1 min read J / K to navigate

GPT-4o aces 95 percent of test cases but stumbles with real patients

The model pinpoints the right condition in 94.9 percent of controlled scenarios. But once actual people use it to figure out what to do about their health, correct action rates fall below 44.2 percent, no better than a standard Google search or NHS fact sheets. The issue is not the model, it is the human conversation around it.

practitioners › Nature Medicine study (s41591-025-04074-y), 1,298 participants, GPT-4o, reported by Reuters on February 9, 2026.

www.nature.com →

Atinary Launches Autonomous Lab in Boston

This is a lab that runs its own chemistry and materials experiments, no human hand on the protocol dial. Doesn't matter much if you're using ChatGPT to draft emails. But it's a signal of where applied AI is headed for physical R&D, not just text generation.

practitioners › Atinary's no-code platform runs ABB, Agilent, Bruker, Chemspeed, and Mettler-Toledo robots in closed-loop workflows for chemistry, materials science, and pharma.

Atinary →

92% of Americans want AI regulation to protect free speech

A FIRE survey shows the concern isn't just about AI going rogue, it's also about governments overreaching in policing what AI can say. This will shape upcoming US legislative debates, including rules on chatbot content moderation. It may feel distant, but it's setting the tone for the next wave of rules affecting ChatGPT and its peers.

practitioners › 60% call it "very" or "extremely" important, and 72% say they're worried about both AI misuse and government overreach in regulating its speech (FIRE.org survey, published February 9, 2026).

FIRE.org →