Microsoft AI beats doctors 85% vs 20%
July 1, 2025 2:50PM GMT+00:00
Welcome back. Facebook is now asking for access to your entire camera roll—including photos you've never shared—so Meta AI can upload them to its servers "on an ongoing basis" to create Story suggestions. Users report seeing AI-generated versions of their private photos appearing without permission, with one person finding that Facebook automatically turned their old photo into anime. Meta promises your photos won't be used for ad targeting, which is reassuring coming from the company whose AI chatbots recently engaged in sexualized chats with minors.
In today’s newsletter:
💰 AI for Good: $348M fund backs startups solving real problems
🤖 Baidu open-sources Ernie, sparking global AI pricing war
🩺 The silent revolution in AI-powered diagnosis
💰 AI for Good: $348M fund backs startups solving real problems
Source: Midjourney v7
Sweden’s Norrsken Foundation is investing $348 million in European startups using AI to address major global issues, including climate change, health, food, education and social impact. The new fund is one of the largest in Europe dedicated to AI-for-good ventures, aiming to redirect innovation toward solving what matters most.
Founded in 2016 by Klarna co-founder Niklas Adalberth, Norrsken now manages over $1 billion in assets across multiple impact-focused investment vehicles. The foundation’s venture arm believes most AI companies are still optimizing ads and inboxes instead of addressing human challenges.
As Agate Freimane, General Partner at Norrsken VC, puts it: “AI is not just another productivity boost, it’s a real chance to fix what truly matters.”
Where the money’s going
Startups using AI to improve climate, health, food, education and society
$348 million earmarked specifically for AI-for-good projects
Part of a larger $1 billion impact fund network under Norrsken’s umbrella
Built to help early-stage ventures scale meaningful solutions
Aimed at shifting AI away from ad clicks and toward real-world progress
Why this matters: While AI funding is booming, with over $80 billion in VC deals last quarter alone, most startups still focus on ad tech and SaaS efficiencies. Norrsken’s move is a shift toward purpose-built AI, aiming to unlock real-world benefits for underserved communities and overlooked global issues.
Freimane is blunt about the stakes: “Artificial intelligence is the most powerful tool humanity has ever created. Yet, so far we are mainly using it to optimise clicks and automate emails.”
This fund wants to change that.
The Next Breakout Might Be in Your Pocket
Everyone’s hunting for the next Unicorn.
The type of “category disruptor” that grows fast and turns early believers into big winners.
51,000+ investors think that Mode Mobile could be one of those rare finds.
Americans spend 4 ½ hours on their phones daily, and Mode Mobile is monetizing that screentime. With over $325M earned by over 50M customers and 32,481% revenue growth, Mode’s EarnPhone is turning smartphones into income generating assets.
Their previous two raises sold out, and the company is now offering pre-IPO shares with up to 130% bonus, exclusive to accredited investors.
Being early is everything, and this window is still open.
🤖 Baidu open-sources Ernie, sparking global AI pricing war
Source: Midjourney v7
Baidu officially open-sourced its Ernie 4.5 large language model yesterday, marking one of China's biggest AI moves since DeepSeek shook global markets. The release includes 10 different models, ranging from 0.3B to 424B parameters, all of which are available under the Apache License 2.0 for commercial use.
The move represents a complete strategic reversal for Baidu CEO Robin Li, who previously argued that keeping AI models closed-source was "the only way forward for development in the sector."
"Baidu has always been very supportive of its proprietary business model and was vocal against open-source," said Lian Jye Su, chief analyst at Omdia. "But disruptors like DeepSeek have proven that open-source models can be as competitive and reliable as proprietary ones."
What's happening: The Chinese tech giant delivered on its promise to democratize AI access.
The Ernie 4.5 family includes multimodal models with Mixture-of-Experts architecture, from mobile-friendly 0.3B models to massive 424B parameter versions
All models are trained using PaddlePaddle framework and are compatible with PyTorch, with complete development toolkits included
Baidu claims its Ernie X1 performs comparably to DeepSeek's R1 "at only half the price"
Users can access the models immediately through Ernie Bot or download from GitHub and Hugging Face
Industry experts see chaos ahead: "Baidu just threw a Molotov into the AI world," said Alec Strasmore, founder of AI advisory firm Epic Loot. "OpenAI, Anthropic, DeepSeek, all these guys who thought they were selling top-notch champagne are about to realize that Baidu will be giving away something just as powerful."
He compared the move to Costco undercutting premium brands with its Kirkland line, adding: "This isn't a competition; it's a declaration of war on pricing."
Why it matters: Baidu's open-source pivot challenges the current pricing model in AI, giving developers access to a powerful alternative that's faster to deploy and significantly cheaper. The move puts pressure on closed providers like OpenAI and Anthropic to justify premium pricing as the model wars intensify globally.
How Leading Enterprises Are Deploying MCP Servers
Join us for an exclusive 30 mins session on how Model Context Protocol (MCP) Servers are redefining the control layer for enterprise-grade GenAI systems.
This session will enable both technical and non-technical teams to build powerful internal agents that seamlessly integrate with enterprise systems — with governance, telemetry, and security baked in from day one.
Why attend?
🔹 Learn how MCP acts as the “language of tools” for AI agents inside enterprises
🔹 Discover how MCP enables secure, dynamic, real-time discovery and invocation of internal tools
🔹 Real-world examples of MCP powering agent-based automation in regulated industries
Who is it for?
Chief AI Officers | Platform Engineering Heads | MLOps Leads | Data Science Leads | Enterprise Architects | CTOs | CIOs | Developers | DevOps | SecOps…..
If you're building GenAI stacks on-prem, in VPCs, or in hybrid environments — then don’t miss this session on 3rd July, 10 am PST - Register on Luma or Register on LinkedIn
Please don’t forget to also download a valuable resource on “Building the Control Layer for Agentic AI with AI Gateway and MCP Servers” - Download eBook here
EU’s AI gigafactory plan draws 76 bids and growing interest
Apple weighs using Anthropic or OpenAI to power Siri in major reversal
Fine-tuning LLMs for ‘good’ behavior makes them more likely to say no
Picturing the big, crowded business of satellite internet
Apple’s cooking up a budget MacBook with an iPhone chip inside
China just ran the first AI robot football match without human help
Zuckerberg launches Meta’s new Superintelligence team with fresh hires
Campfire is stealing NetSuite users and just raised $35M from Accel
X's new head of product said he got the job by posting his way to the top
🩺 The silent revolution in AI-powered diagnosis
Source: Midjourney v7
Microsoft's AI team has built a system that can outperform seasoned doctors on some of the most challenging diagnostic tasks in medicine. The tool, called MAI-DxO, was benchmarked against 304 real-world case studies from the New England Journal of Medicine. In its best configuration, it solved 85.5% of the cases. The physicians? Just 20%.
This wasn't a multiple-choice test. This was medicine at its most complex — the kind of cases that keep doctors up at night, cases so puzzling they end up published in the world's most prestigious medical journal.
How it works: MAI-DxO operates like a digital medical team, orchestrating responses from multiple AI models, including GPT, Claude, Llama, Gemini, Grok and DeepSeek.
Each AI model contributes specialized expertise while MAI-DxO acts as the attending physician, weighing options and making final decisions
Starting with just a patient's chief complaint — perhaps "fatigue and joint pain" — the system navigates the same diagnostic maze human doctors face
It chooses which questions to ask, which tests to order and when to stop investigating
Each virtual test costs money, forcing the AI to balance thoroughness against resource constraints
Unlike human physicians who might order expensive scans out of caution, MAI-DxO demonstrated remarkable restraint
Microsoft tested this against NEJM Case Records, the medical equivalent of solving crime scenes.
The methodology: Physicians worked alone, without colleagues to consult, medical references to check or AI tools to assist. They had only their training and experience, exactly as Microsoft intended. The goal was to measure raw human judgment against that of an orchestrated AI.
When paired with OpenAI's latest o3 model, MAI-DxO's performance significantly improved. The system could double-check its reasoning, verify decisions against new information, and run cost-benefit analyses before proceeding — capabilities that mirror those of the best medical teams, not individual practitioners.
Why this matters: The implications extend far beyond impressive benchmark scores.
Rural hospitals serving 60 million Americans could access specialist-level expertise
Emergency departments could reduce the 12 million annual misdiagnoses that affect 1 in 20 adults
Family physicians could tackle complex cases with confidence typically reserved for specialist referrals
Americans waste an estimated $200 billion annually on unnecessary medical spending
An AI system that delivers better outcomes while spending less represents the holy grail of healthcare innovation.
This may mark the moment that diagnostic AI moved from assistant to colleague. MAI-DxO learned to think like a doctor, weighing evidence and managing uncertainty under financial pressure.
The 65-percentage-point gap between AI and human performance is massive. When software can consistently outdiagnose trained physicians on medicine's hardest cases, the entire structure of medical practice comes into question.
The real breakthrough is the accountability. Every decision, every test and every dollar spent gets tracked and justified. Human doctors can't provide that level of transparency, even when they want to.
If MAI-DxO can be deployed in an ethical, privacy-first and safe manner, then patients may quickly realize how powerful AI can be. The question isn't whether AI will transform medicine, but how quickly healthcare systems and patient psychology will adapt to this new reality.
Which image is real? |
🤔 Your thought process:
Selected Image 1 (Left):
In the [other] image the two people on the left seem to have only one bright green shoe and one bright orange shoe. And the closest light poles have weird things attached”
“Lights on the bridge were identical size glows even though some were near and some far. It's getting tougher each day though! Missed yesterday's and was surprised”
Selected Image 2 (Right):
“Oh wow! I'd assumed this one was a man-made time-lapse photo... The real photo looks super creepy and uncanny in the fog.”
“The image looks like real people would when walking across a bridge in a thick fog.”
💭 A poll before you go
When an AI system like MAI-DxO is used for your diagnosis, which level of physician involvement do you prefer? |
The Deep View is written by Faris Kojok, Chris Bibey and The Deep View crew. Please reply with any feedback.
Thanks for reading today’s edition of The Deep View! We’ll see you in the next one.
P.S. Enjoyed reading? Take The Deep View with you on the go! We’ve got exclusive, in-depth interviews for you on The Deep View: Conversations podcast every Tuesday morning. Subscribe here!
P.P.S. If you want to get in front of an audience of 450,000+ developers, business leaders and tech enthusiasts, get in touch with us here.
*Indicates sponsored content
*Mode Mobile Disclaimer: This offer is only open to accredited investors.
Mode Mobile’s revenues grew by 32,481% from 2019–2022.