Microsoft’s new AI system just diagnosed patients four times more accurately than human doctors in tests.
Key Points:
- Microsoft built a system called MAI-DxO that combines multiple top AI models, including GPT, Gemini, Claude, and others.
- In a test using 304 real medical cases, it reached 80% accuracy, while human doctors scored 20%.
- The AI also cut costs by 20% by picking cheaper tests and procedures.
Details:
Microsoft says its new medical AI system is a big step toward what it calls “medical superintelligence.” The system, MAI-DxO, works by using different AI models together, like a group of doctors debating a case. It analyzed case studies from the New England Journal of Medicine, breaking each case into steps similar to a real diagnostic process. The system outperformed doctors in accuracy and suggested more cost-effective options. The company has not yet decided if it will release this tool publicly but hinted at possible future use in Bing or as support for medical professionals. Experts say it’s an impressive step but warn that results from lab studies may not match real-life clinics. Microsoft plans to keep testing in real-world settings before any wide rollout.
Why It Matters:
This is actually pretty exciting. Imagine going to the doctor and getting answers faster and more accurately, without a bunch of extra tests or huge bills. That’s the big promise here. AI won’t replace doctors, but it could handle the tough analysis, so doctors can focus on real care and talking to patients. We’re not fully there yet since real-world trials still need to prove it works, but it feels like a real step toward making healthcare smarter and more personal at the same time.
Apple weighs using Anthropic or OpenAI to power Siri in major reversal
Apple is exploring using Anthropic or OpenAI tech to power a new Siri version.
Key Points:
- Apple is in talks with Anthropic and OpenAI to test their language models for Siri.
- The company hasn’t made a final decision yet, and talks are still early.
- Apple had delayed major Siri AI upgrades to 2026 and recently reshuffled leadership to fix setbacks.
Details:
Bloomberg reports that Apple might replace its own AI models for Siri with ones from Anthropic or OpenAI. Apple has asked both companies to train versions of their models that could run on Apple’s cloud for testing. While no final decision is made, this move would mark a major change in strategy. Earlier this year, Apple delayed planned AI improvements to Siri to 2026 and reshuffled leadership, with Mike Rockwell taking over after Tim Cook lost confidence in John Giannandrea. At its recent developer conference, Apple emphasized smaller AI updates like live call translations and opened its AI tools to developers, including offering OpenAI’s code completion tools.
Why It Matters:
Siri has felt behind for years. Apple is finally admitting it needs help to catch up with smarter assistants from OpenAI, Google and Amazon. By looking at Anthropic and OpenAI, Apple might actually make Siri useful again. But letting in outside tech is a big shift for Apple’s control-heavy style. If it works, we might finally get a Siri that understands us, though people will be watching closely to see if Apple keeps its privacy promises. Still, for a tech giant like Apple, it’s surprising they can’t get their act together.
Baidu goes open source with ERNIE 4.5
Baidu has open-sourced its new ERNIE 4.5 model family, changing its strategy to compete globally and support wider AI development.
Key Points:
- Baidu released ERNIE 4.5 under the Apache 2.0 license, reversing its closed-source approach.
- The lineup has ten model variants, including large Mixture-of-Experts versions and compact models, plus developer tools like ERNIEKit and FastDeploy.
- ERNIE 4.5 performs well against other Chinese models but doesn’t beat leaders like OpenAI or Google.
Details:
Baidu surprised the AI world by open-sourcing ERNIE 4.5, moving away from its past strategy of keeping models closed. This shift follows pressure from Deepseek’s open-source success. ERNIE 4.5 offers a wide range of models, from small 0.3 billion parameter versions to massive Mixture-of-Experts models with up to 424 billion total parameters. It also includes tools to help developers deploy and use these models easily. Benchmark tests show strong results, outperforming Deepseek-V3 on most metrics, though it doesn’t top global leaders like OpenAI’s o3 or Gemini 2.5 Pro. Baidu’s move reflects a growing trend of major AI labs opening up to the community, raising standards and encouraging more accessible innovation.
Why It Matters:
This move shows Baidu isn’t just reacting, it’s joining a big trend where more labs open their models to everyone. People are tired of being stuck with closed systems and high fees. By going open, Baidu helps push the whole industry forward and gives developers more freedom to build new stuff. It also pressures other big players to share more, which might lead to faster and cheaper AI for all of us.
