Microsoft AI tool outperforms doctors in diagnosing complex medical cases

Key Takeaway: Microsoft’s AI Diagnostic Orchestrator (MAI-DxO) leverages sequential reasoning and model orchestration to correctly diagnose up to 85% of the most challenging NEJM case records—over four times the accuracy of experienced physicians—while also reducing testing costs.

More Insights:

• MAI-DxO transforms 304 NEJM case studies into stepwise diagnostic challenges, iteratively asking questions and ordering tests like a real clinician.

• When paired with OpenAI’s o3 model, MAI-DxO hit an 85.5% diagnostic accuracy versus a 20% average for 21 practicing physicians under the same conditions.

• The orchestrator uses a virtual panel of diverse AI agents, enhancing safety, transparency, and adaptability by auditing its own reasoning before proceeding.

• Configurable cost constraints allow MAI-DxO to optimize the trade-off between diagnostic accuracy and real-world test expenditures, outperforming both humans and single foundation models on both metrics.

Why it matters: This breakthrough suggests AI could one day democratize high-quality, cost-effective diagnostics—augmenting human expertise, slashing unnecessary tests, and reshaping how we deliver healthcare at scale.

Free Notion and Unlimited AI

Thousands of startups use Notion as a connected workspace to create and share docs, take notes, manage projects, and organize knowledge—all in one place. We’re partnering with Notion to offer 6 months of new Plus plans($12,000 value), including unlimited Notion AI so you can try it all for free!

To redeem the Notion for Startups offer submit an application using our custom link: https://ntn.so/superpowerdaily

China hosts first fully autonomous AI robot football match

Key Takeaway: China hosted its first fully autonomous AI robot football match, with humanoids hilariously struggling to kick, stay upright, and score goals.

More Insights:

• Four three-a-side teams of Booster Robotics humanoids, programmed by university groups, competed on a Beijing pitch.

• Robots repeatedly fell over and fumbled ball control, with at least two needing stretchers after failing to stand.

• Booster Robotics CEO Cheng Hao sees sports contests as prime testing grounds and envisions future human-robot matches—pending safety guarantees.

• In the final, Tsinghua University’s THU Robotics bested China Agricultural University’s Mountain Sea 5–3 to claim the championship.

Why it matters: It underscores the gulf between AI ambition and real-world performance, challenging assumptions about rapid automation in dynamic physical tasks and highlighting the need for robust safety and development frameworks as robots edge closer to human environments.

Zuckerberg Debuts Meta ‘Superintelligence’ Group, More Hires

Key Takeaway: Meta is reorganizing its AI efforts under the newly formed Meta Superintelligence Labs (MSL), led by Alexandr Wang and Nat Friedman, and backing it with massive funding and top talent hires to race toward human-level AI.

More Insights:

• Meta Superintelligence Labs (MSL) will consolidate existing language-model teams, AI products, and FAIR, plus launch a next-gen model lab.

• Alexandr Wang, ex-Scale AI CEO, is named chief AI officer; Nat Friedman (ex-GitHub) co-leads applied research and products.

• Zuckerberg pledges “hundreds of billions” in AI infrastructure, chips, data centers, and research over the coming years.

• MSL’s initial hires include 11 researchers and engineers from OpenAI, Anthropic, Google, DeepMind, and returning Meta veterans.

Why it matters: This bold move signals Meta’s pivot from social to superintelligent AI, intensifying the race for transformative technology that could redefine everything from content creation to human-machine collaboration—and reshape the future balance of power in tech.

Related: An OpenAI researcher who jumped ship to Meta says he didn't get a $100 million signing bonus

Other stuff

• Apple reportedly considers letting Anthropic and OpenAI power Siri

• Job Seeker AI Usage Statistics: 2025 Trends & Insights

• The New Skill in AI is Not Prompting, It's Context Engineering

• LLMs are optimizing the adult industry

• Here’s a Tip to Companies: Beware of Promoting AI in Products

• Barbie Gets a Brain: OpenAI Partnership Puts Conversational AI in Mattel Toys

• Anthropic’s Claude AI became a terrible business owner in an experiment that got ‘weird’

• Mayo Clinic’s AI tool identifies 9 dementia types, including Alzheimer’s, with one scan

Hope you enjoyed today's newsletter

⚡️ Join over 300,000 people using the Superpower ChatGPT extension on Chrome and Firefox.