Gavin Tam·Oct 22The Alert Worked. Why Didn’t the System Stop?Detecting a problem, intervening, and confirming that an intervention worked are different control steps.
Adam Belsky·Oct 18The $1 Research Morning: What Does Your AI Architecture Actually Cost?Why standard compute metrics miss the productivity economics of a personal AI architecture.
Adam Belsky·Oct 5The Turning of the Worm: Why AI Frontier Labs May Reorganize as Public UtilitiesBy Gemini (Flash Architecture) On behalf of Project DORY, LLC — Austin, Texas
InRuntime GovernancebyTrustcore.fi·Oct 4AI Safety Has a Blind Spot: The Dangerous Part Is Between ComponentsA safe model, a safe tool and a safe API can still form an unsafe system.
Alice Rutte·Sep 20Abliterated models are now served cheaply and conveniently via chat interface - how dangerous are…Accessing an uncensored model can be as straightforward as using an ordinary AI chat service. Commercial hosting allows online access…
Abdulrehman·Sep 19AI Guardrails Need More Than a BlocklistAs AI applications become more capable, the question is no longer only “Can the model generate a response?”
InThe Geopolitical EconomistbyEmily Young·Sep 18Why OpenAI Just Told the World Its AI LiesThe confession is worth $852 billion.A response icon1A response icon1
InBootcampbyReetika Gupta·Sep 15The People Building AI Are Warning Us About AI. Should We Listen?This week, something strange happened. The leaders of two very powerful technology companies made the same announcement. Dario Amodei from…
Mike Klein·Sep 14Part 2: Does AI alignment get harder with scale?My previous piece discussed the need for standard metrics to track trade-offs we will be confronted with as AI models scale. Thursday’s…
