- AI Horizons
- Posts
- Microsoft’s AI Diagnoses Patients 4x More Accurately Than Doctors
Microsoft’s AI Diagnoses Patients 4x More Accurately Than Doctors
PLUS: Cloudflare’s Bold Move to Protect Web Content, the Rise of Multi-Agent AI Systems, and the Best New Tool for Building AI Agents
Welcome back to AI Horizons, your weekly shortcut to staying ahead in AI, minus the hype, in under 5 minutes. Here's what’s on deck today:
Microsoft’s breakthrough in medical AI
Cloudflare’s fight against AI scraping
News traffic lost to ChatGPT
Ilya Sutskever’s new AI lab
Multi-agent AI course from IBM
Tool for AI workflows
FEATURED INSIGHT💡
Microsoft’s AI Diagnostician Just Outperformed Doctors by 4x

Microsoft has introduced a system called MAI-DxO that’s solving complex medical cases more accurately (and more efficiently) than experienced physicians. Tested against the famously tricky NEJM Case Records, MAI-DxO paired with OpenAI’s o3 model correctly diagnosed 85.5% of the cases. Human doctors? Just 20%.
What’s different here isn’t just accuracy, it’s reasoning. Rather than guessing from multiple-choice tests, MAI-DxO steps through real diagnostic workflows, ordering virtual tests, checking costs, and verifying its own conclusions. It mimics human thinking, orchestrating multiple models like a panel of experts.
This isn’t clinical deployment yet, but it’s a major signal. Diagnostic AI is moving past flashcard memorization and into real-world reasoning, potentially transforming how we approach medical care, cost-efficiency, and expertise at scale.
ON THE HORIZON 🌅
Cloudflare Just Changed the Rules for AI on the Open Web

Image Source: Reuters
In a quiet but seismic shift, Cloudflare has flipped the default on AI crawlers: no more free scraping. Now, AI bots must ask for permission before training on your content.
Why it matters: For decades, content has been indexed, not ingested. AI changed that. Crawlers now vacuum up data without attribution, killing traffic and incentive for creators. Cloudflare’s opt-in system forces AI companies to disclose their purpose and seek explicit approval.
With AI transforming how content is used, this marks a return to a model where original work is protected, and creators hold the leverage.
Keep an eye on this space. If AI companies comply, it could reshape licensing and data markets.
LATEST IMPORTANT NEWS 📰
AI Is Eating the News. Slowly, Then All at Once.
News sites are seeing a massive decline in organic search traffic, down from 2.3B to under 1.7B visits/year, as users get answers directly from AI models instead of clicking through.
ChatGPT referrals to news sites are growing (25x year-over-year), but they’re nowhere near enough to offset search declines. Publishers like Reuters and Business Insider are seeing modest gains, but others, like The New York Times, aren’t faring as well.
Meanwhile, OpenAI and Google are pushing tools that further entrench this shift, and some publishers are scrambling for solutions: paywalls, micropayments, newsletters, or just… layoffs.
This isn't just a traffic problem. It's an existential shift in how information is discovered and monetized.
Ilya Sutskever Becomes CEO of Safe Superintelligence
Following the departure of co-founder Daniel Gross, OpenAI co-founder Ilya Sutskever is taking the helm of Safe Superintelligence, a startup singularly focused on building “safe AGI.”
Why this matters: Meta allegedly tried to acquire the entire company. Now Sutskever leads it solo, promising to build superintelligence from scratch, without side products or distractions.
Some are skeptical: Why did Gross bail if they were so close to something huge? But if SSI really is the next moonshot, Sutskever’s leadership keeps it on track… or steers it into deeper water.
xAI’s Data Center Faces Scrutiny Over Generator Use
Elon Musk’s xAI has been operating dozens of natural gas generators at its Memphis data center, some without permits. Regulators have now granted approval for 15 units, but legal groups are preparing lawsuits, citing possible Clean Air Act violations and concerns over pollutant emissions, including formaldehyde.
Why it matters: As AI infrastructure expands, so does its energy demand. This situation highlights the growing friction between rapid AI development, environmental regulation, and the push for sustainable infrastructure.
FOR THE TECHNICALLY INCLINED 🛠️
How to Build Composable, Multi-Agent Systems with ACP
The Agent Communication Protocol (ACP), built in partnership with IBM Research, makes it possible to chain agents across frameworks without glue code or tangled dependencies. If you're building multi-agent systems, or want to start, this free course will show you how to:
Wrap agents with an ACP server
Discover and chain them into linear and hierarchical workflows
Create agent registries for team-wide sharing
It’s lightweight, open, and supports protocols like CrewAI, Smolagents, and more.
4 Titans of the Creator Economy. One Stage. One Day.
You know their names. You’ve followed their success. But you’ve never seen them all in one place—until now.
On July 16th, four of the most influential minds in the creator space are coming together for a live, virtual event: The Creator Business Summit.
Shaan Puri. Codie Sanchez. Chris Koerner. Tyler Denk.
They’ve built newsletters, brands, and 7-figure companies. And they’re breaking down how they did it. For free. For you.
From audience growth to monetization to long-term brand strategy, this is your chance to learn directly from the best.
Whether it’s day 1 or you’ve been building for a while, this is the one event you can’t afford to miss.
AI TOOL OF THE DAY 🚀
If you’re building workflows, n8n gives you more freedom than Zapier or Make, especially for AI-powered, multi-step automation.
Build with drag-and-drop, or dive deep with code. It’s flexible, open-source, and perfect for teams building complex logic with on-prem or cloud hosting.
AI Won’t Replace You… But Someone Using It Will.
Master prompts. Get smarter results. Work faster. Grab The AI Prompt Playbook — your guide to unlocking the full power of generative AI tools like ChatGPT, Gemini, Claude, Midjourney, and more.
Now on sale (50% off!) for a limited time only. Don't get left behind.
That's all for now!
We'll catch you in the next one.
Cheers,
The AI Horizons Team
P.S. If you missed our last issue, no worries, you can check out all previous issues here!
P.P.S We value your thoughts, feedback, and questions - feel free to respond directly to this email!
... and if you enjoyed this email and would like to support our work and help us keep bringing you cutting-edge AI insights, you can donate here. Every bit makes a difference—thank you for your support!
What did you think about today's email? |