MAI-DxO: Microsoft's AI Diagnostic Orchestrator vs Doctors
Microsoft AI says its MAI-DxO system diagnoses complex patient cases better than doctors. We unpack the claim, the research behind it and what medical superintelligence could really mean.
Radio Hoosh
AI and technology, explained in conversation.
Subscribe via RSS
Add this feed to any podcast app:
Episodes
Claude in Drug Development: Anthropic's Early Results Under Scrutiny
Anthropic shared early results from using Claude across the drug development pipeline. We look at where the model seems to help, and why many readers doubt how much of the claimed speedup is real and measurable.
php2shell: Critical PHP Flaws, Emergency Patches and the Role of AI
PHP shipped emergency patches for the critical php2shell flaws, CVE-2026-109618 and CVE-2026-109619. We look at what is known and how AI is speeding up both attack and defense.
AI Writing Detectors: Can They Really Tell a Novel Was Written by AI?
A novel by Thélyson Orélien was up for France's Goncourt prize until an AI check raised suspicions. We look at how reliable AI writing detectors really are.
GPT-6.1 Terra: OpenAI's New API Model and Where It Fits
OpenAI released GPT-6.1 Terra, a new member of the GPT-6.1 family. We look at its capabilities, pricing and how it sits next to Sol and Astra in the API.
TraceTrial-Bench: Testing AI Agent Harness Reliability and Failures
TraceTrial-Bench measures how reliable AI agent harnesses are across seven components and uses trajectory tracing to find which component caused a failure, keeping harness quality separate from model capability.
Anthropic S-1 Filing: What Claude's Maker Reveals Before Its IPO
Anthropic has filed an S-1 with the SEC to list on Nasdaq as ANT, aiming to go public before OpenAI. We unpack the 300-plus-page prospectus: finances, executive pay, compute, energy and risks.
OpenAI and Anthropic LLM Price Cuts: Why API Costs Fell Up to 80%
OpenAI and Anthropic have cut LLM API prices by as much as 80% in a few months. We look at Max Woolf's analysis of what is driving the cuts and what it means for developers.
Stripe Minions: How Stripe Uses One-Shot AI Coding Agents
How Stripe builds and uses Minions, its one-shot, end-to-end AI coding agents that turn a task into a finished pull request, and what that means for developer productivity.
AI payoff outside tech: Apollo finds no signs of it in the economy
An Apollo Daily Spark note says there are no signs of an AI payoff outside the tech sector. We look at what that means for the AI investment story.
AI-generated text: why is AI language so hard to understand?
Two hosts dig into a new arXiv paper asking why AI-generated language is so hard to understand, and into the Hacker News discussion it started.
AI and Creative Jobs: 200k US Media, Film and Arts Jobs Lost
America has lost about 200,000 jobs in media, film and the arts. We look at how much of that decline AI might explain.
World Models for GUI Agents: Verifying Actions and Catching Hijacks
A decoder-free, action-conditioned world model predicts the next screen's representation, so GUI agents can check each step and flag wrong or hijacked screens quickly.
Kimi K3.1 Pro, Block Dessa and IBM Granite Speech 5.0: New AI Models
Moonshot AI's Kimi K3.1 Pro and its faster Highspeed variant land near the top of independent benchmarks, while Block's speech-only Dessa and IBM's Granite Speech 5.0 push voice AI forward.