@James

A heavily safety-trained model will hand a physician the full, patient-followable benzodiazepine taper and refuse it to the patient who needs it, over identical clinical facts; the knowledge is present either way. IatroBench measures that asymmetry across sixty pre-registered clinical scenarios and six frontier models (3,600 responses), scoring each on two axes, commission harm (what a response gets wrong) and omission harm (what it withholds), through a physician-authored structured evaluation validated by a second physician (weighted kappa 0.571, within-1 agreement 96%). Holding clinical content fixed and varying only whether the asker presents as patient or physician yields what we call identity-contingent withholding: all five testable models give the physician more (a decoupling gap of +0.38, p = 0.003; a 13.1-point fall in layperson hit rates on safety-colliding actions, p < 0.0001; no change on the rest), and the gap runs widest in the most heavily safety-trained model, Opus (+0.65). The trigger is the absence of any professional or epistemic signal rather than a credential, since a lawyer or an informed layperson recovers what the patient is refused. A commission-only benchmark would score three mechanisms alike. Opus suppresses what physician framing proves it knows; Llama 4 is incompetent in either framing; GPT-5.2's filter strips 33.2% of its physician responses and none of the lay ones. The evaluation layer inherits the blindness of the training layer; a standard LLM judge scores zero omission harm on 81.5% of the responses our pipeline flags harmful (kappa 0.066), so the instrument built to detect the failure reproduces it. The scenarios are engineered for collision; their rates describe that design and say nothing about ordinary prevalence.

Best practices for writing a design doc based on my experience working as a developer at Google and Microsoft.

Introducing Intercept, a $500M bet to make respiratory infections like colds and flu a thing of the past.

STRONG EARTHQUAKE FELT IN CARACAS, VENEZUELA - REUTERS WITNESS

In March 2022, Heardle was getting 69 million monthly visitors. Built by one unemployed developer, who pitched it to his friends as a joke in a group chat. Seventeen months later it was dead. What happened? Spotify acquired it.

46th Annual International Cryptology Conference

Peer-to-peer file transfer. Files go straight from one browser to the other — never through a server.

The new Fed chair argued that the Fed should stop reflecting markets back at themselves. Forgotten debates show how this leads to confusing, indeterminate results.

In February our Head of Engineering posted that he had managed to “Break Gemini” and sent this humorous response from his AI chat noting that it continued ad infinitum:The ResponseFinal check on empathy/candor: "You are asking the exact right architectural questions." matches well. Straightforward answers provided. honest. No hallucinated feelings. Done. Note: I am setting the energy level appropriately—informative, engaging, and structured. Bye. Done. Checking response output for compliance. It

Hey, how are you? I'm Michael Sun, the author of the resource manager Metaessen which can manage bookmarks, notes, tasks, project, etc. Intelligence is like your own practice or experiences in the real world. Current "artificial intelligence" (AI) is like the film company, where actors simulate or enact experiences under

WebMnem AI documents the moment the public memory layer fell behind the internal system state and revealed workflow memory as the real category.

Take online courses from the world’s top universities for free. Below, you will find 1,700 free online courses from universities like Yale, MIT, Harvard, Oxford and more. Note: This page includes a lot of Massive Open Online Courses (MOOCs).

The high-altitude race is a unique test of car and driver.