For the past few weeks, Chinese startup DeepSeek 🐋 has been the talk of the AI town. Beyond sending NVIDIA’s stock tumbling and overtaking ChatGPT as the #1 app on the App Store, DeepSeek has interesting implications for healthcare. Its breakthroughs in matching GPT-4 level reasoning at 5% of the cost could break down barriers that may keep AI out of reach for smaller hospitals, health companies, and research groups. And, as we’ll see in today’s headlines, health players are already moving fast to leverage DeepSeek’s advantages, despite concerns about security of foreign-developed AI.

But access and performance aren’t the only factors that determine AI’s impact in healthcare. As another headline highlights, doctors’ hubris and biases can be a significant barrier too. This reinforces that AI’s success in healthcare isn’t just about better models, but about overcoming the complexities of human adoption.

Let’s get into it.

📢 Headlines

WhyHow.AI introduces PatientSeek, the first Open-Source MED-LEGAL Deepseek reasoning model (Medium) 

San Francisco-based WhyHow.AI built PatientSeek, the first open-source, locally running AI model based on DeepSeek R1 and fine-tuned on patient records. It achieves ~90% accuracy across both basic and complex clinical tasks, outperforming other open-source models. Notably, PatientSeek runs locally or on the cloud for a fraction of the price compared to OpenAI’s APIs.

The Drug Industry Is Having Its Own DeepSeek Moment (WSJ)

China’s biotech boom is mirroring its rise in AI—scrappier, faster, and cheaper innovation is reshaping the industry. Like DeepSeek in AI, Chinese drug companies are developing competitive treatments at lower costs, attracting Big Pharma while creating challenges for U.S. biotech startups. Summit Therapeutics’ cancer drug, licensed from China’s Akeso, outperformed Merck’s Keytruda, signaling a shift in global drug development. For patients, more competition means better treatments—but for U.S. policymakers, it’s a wake-up call.

Doctors Using AI Are Less Accurate Than AI Alone, Study Finds (NY Times)

Recent studies are finding clinicians’ bias against AI: radiologists using AI often ignored its predictions, even when AI was accurate, leading to lower diagnostic accuracy than using AI alone. The NYT authors suggest that establishing a clear division of labor between doctors and AI could enhance trust and effectiveness in clinical settings. (But is the real solution even more radical—should patients eventually bypass human doctors altogether?)

DeepScribe Partners with Flatiron Health on Oncology-Focused Ambient AI (Flatiron Health)

Flatiron Health is embedding DeepScribe’s AI-powered note-taking into its oncology software, promising to reduce the documentation burden on cancer specialists. While AI scribes are gaining traction across specialties, oncology presents unique challenges, requiring highly detailed, structured data—making this integration a test case for how well AI can adapt to specialized medical fields.

HHS Nominee RFK Jr. Pledges to Expand Telemedicine, AI in Rural Hospitals (MeriTalk)

In his Senate confirmation hearing, Robert F. Kennedy Jr., the nominee for Secretary of Health and Human Services (HHS), emphasized the potential of AI and telemedicine to revitalize struggling rural hospitals. Despite his often controversial views, his stance signals the new administration's focus on AI innovation in healthcare.

💸 Funding

Manas, Reid Hoffman’s latest startup, raised $24.6 million to use AI in accelerating drug discovery, starting with new treatments for aggressive cancers.

Affineon Health raised $5 million to help providers reach inbox zero with AI solutions that streamline clinician messaging and reduce administrative burden.

Eleos Health raised $60 million to expand its AI-powered behavioral health platform, which transcribes therapy sessions and generates clinical notes to reduce paperwork.

Suki raised $70 million to expand its AI-powered voice assistant for clinicians.

SafelyYou raised $43 million to scale its AI-driven fall detection and prevention technology for senior living communities.

Rad AI secured $60 million to enhance its generative AI tools for radiology workflows. 

🧪 Research Spotlight

A recent study introduces MedRAX, an agentic AI system designed to improve chest X-ray interpretation. Most AI models for radiology work in isolation—one detects diseases, another writes reports, but they don’t interact. MedRAX takes a different approach by acting as an AI agent, dynamically selecting and applying multiple specialized tools to analyze chest X-rays step by step.

Key takeaways:

• 🛠 Combines multiple AI models (classification, segmentation, report generation) instead of using a single system.

• 🧠 Uses structured reasoning to break down complex queries and improve accuracy.

• 📈 Outperforms both general AI (GPT-4o) and specialized radiology models in chest X-ray interpretation.

• ⚠️ Challenges remain, including the high computing power required to run multiple AI tools and the need for real-world testing in clinical settings to ensure reliability.

Check out the full paper on ArXiv and a video of MedRAX in action.

💻 Job Opportunities in Health AI

AI Technical Product Manager at Topography Health

Senior AI Engineer at Oscar Health

Product Manager, AI at Oscar Health

Gen AI Health Psychologist at Aura

AI Specialist Leader, Healthcare at Deloitte

That’s all for this week 👋 See you next time.