The Sunday Essay
Can we actually slow AI down?
Dario Amodei says we must. But once powerful intelligence becomes open, local and cheap, the real safety problem may lie somewhere else.
Not more AI news. Better judgement about what AI news means.
The Sunday Essay
Dario Amodei says we must. But once powerful intelligence becomes open, local and cheap, the real safety problem may lie somewhere else.
Revenue, financing, IPOs, labour markets and the economics of intelligence.
Models, agents, robotics, open weights, chips and systems beneath the spectacle.
Resources
Every model ends in matter: data centres, transformers, cooling systems, supply chains and capital.
Explore the evidence ↗Resources desk
Work, education, psychology, trust, inequality and everyday life.
Cybersecurity, biosecurity, disinformation and concentrations of power.
Where capability becomes meaningful human and economic value.
What everyone says
What the evidence complicates
Open weights, falling inference costs and local deployment move parts of the safety problem from model governance to infrastructure security.
Studies, filings, reports, benchmarks and primary documents worth reading.
Arguments and essays. Interpretation is labelled; uncertainty stays visible.
Perhaps the ultimate AI safety problem is not how to stop intelligence, but how to build a world that remains safe when intelligence can no longer be stopped.
Mark T. Drexler →Automation should accelerate the newsroom—not impersonate editorial judgement.
The editorial system
Trusted journalism, company filings, institutions and research.
Dates, numbers and claims traced to authoritative sources.
Business, technology, resources, society, threats and opportunities.
Routine briefs can be automated. Sensitive analysis requires editorial review.
Not more AI news. Better judgement about what AI news means.