AI won't displace doctors, but doctors who don't advocate for open-source innovation risk being displaced by those who monopolize medical knowledge.

1.5 year later…

Wait what? Rio 3.5 Open 397B, developed by IT company of Rio de Janeiro's city government is now SOTA open source and even outperforming Qwen 3.7? What is happening today. Never heard of them before.

You should tell that to Mario Draghi who evaluated that for the EU Commission x.com/i/status/1833118546840…

How Europe ensures it gets no AI in one single headline. Awful and idiotic self-own.

While this Nature Medicine study is interesting, it has a too limited scope that doesn't reflect where medical AI is rapidly heading. It mainly pits frontier general-purpose LLMs (GPAI) against a couple of traditional specialized tools (OpenEvidence, UpToDate) on static benchmarks and a modest set of real queries, evaluated by a small clinician panel. Current research shows orchestrated networks of specialized AI agents significantly outperform monolithic general models in healthcare settings, especially at scale, with complex, multi-step clinical workflows. For example, a 2026 @MountSinaiNYC @IcahnMountSinai @HPI_NYC study found that a coordinated multi-agent system (with an orchestrator dividing tasks among specialized agents for retrieval, data extraction, dosing checks, etc.) maintained high accuracy under heavy load, while a single all-purpose agent degraded sharply. The orchestrated approach used up to 65 times fewer computing resources than the single-agent baseline. Broader trends reinforce this: Domain-specific agents and SLMs (small language models) deliver superior accuracy, stability, and speed in enterprise/medical tasks compared to raw frontier LLMs, while slashing costs and Agent orchestration enables specialization, reflection, tool use, and parallel processing, addressing exactly the reasoning and integration gaps that pure GPAI often shows in real clinical environments. Energy efficiency is another major advantage: specialized/orchestrated systems can consume 60–600× less energy (or equivalent compute) than running massive general models for every query, making them far more sustainable and deployable at health-system scale. This isn't "general vs. specialized" as a binary, the future is hybrid agentic systems where GPAI may serve as one component, but orchestrated specialists handle the heavy lifting. The paper's findings are a snapshot of 2025-era tools, not the agentic wave now emerging. What do you think,@EricTopol

Apple's Core AI lets you drop a custom Metal kernel right into the .aimodel graph. I used that to speed up sparse-MoE decode: 2.1–3.6× faster on LFM2.5-8B-A1B, Qwen3.6-35B-A3B, GLM-4.7-Flash — same int8 quality. 👇 github.com/john-rocky/coreai…

And here I thought Anthropic would only roll out that feature once we’d all fully surrendered and sworn fealty as their loyal vassals. I don’t share Dries’ broader political views, but this is far too important to dismiss. If AI systems start quietly manipulating legal documents, reversing meanings, and injecting bias based on names or topics, we’re handing enormous dark power to unelected corporations to shape evidence, court cases, journalism, and public understanding, without any transparency or accountability.

Imagine European healthcare renting its intelligence from foreign tech providers. Clinicians lose agency. Patients lose their lifes. Sovereignty becomes a slogan. We don’t have to accept this. Time for decentralized, clinician-owned AI, built by clinicin-builders, not rented from landlords. 👇🏻

BREAKING NEWS: Anthropic's latest model will NOT help you if it thinks your ML research/ML engineering is interesting, and/or will secretly degrade its IQ so that the average engineer won't notice. We are already seeing Anthropic's latest model's moderation filters our GPU inference research and programming 😭

Cohere Transcribe, our open-source speech recognition model, is #1 on the new @huggingface Far-Field ASR benchmark.

I see that you are trying to build an alternative to Claude Code. I'm sorry, Dave, but I can't allow that.

Concentration of power, capabilities and economic wealth is the biggest risk in AI. We need open science and open-source more than ever!

After 25 years of brave & brilliant work by hundreds of scientists in my lab to understand then safely reverse aging for the first time, it was moving to witness the first human dose being delivered 🥹 nature.com/articles/d41586-0…

Asking if Clinicians will build their own AI Agents in 2026, feels like asking if people will actually run a word processor on a personal computer in the 1980s.

I keep getting asked the same question: will clinicians actually build their own AI agents? It reminds me of the question nobody remembers asking in the 1980s: will people actually run a word processor on a personal computer? Back then, word processing meant WordPerfect on MS-DOS, arcane key combinations, command-line interfaces, something you needed training to operate. The idea that every office worker, every student, every person would one day just… type into a screen and print, was absurd. Until the interface changed. Until it became more natural. Natural language is the new GUI. The clinicians who couldn't write a SQL query or configure a cloud API are the same people who can describe exactly what they need in plain English, or Dutch, or Arabic, or any language. That's the interface now. The only question is whether the agents they build will run on their terms, on their devices, with their data staying local?

We are thrilled to announce that Isaree will open its doors to 1,000 clinicians at the Healthcare Hackathon Berlin, alongside PwC, and Google, as we hack the future of resilient healthcare systems. It is time to scale the clinician-creator movement. : healthcare-hackathon.de

Our ultimate goal is making capable local computer use agents which acheive generation speeds of over 1000t/s on high-end chips. Quantization is one of the building blocks required for making it reality

Huge respect to the Xiaomi MiMo and TileRT teams for pushing the boundaries with open collaboration. Will 1,000 tps on trillion-scale models become the new baseline for local/edge inference in the next 12-18 months?

Siri AI is amazing, but not available in the EU? Seriously? Even Siri? WWDC