AI keeps getting smarter

The past week has been packed with announcements that show just how quickly AI is reshaping our world. Here’s a roundup of the key stories I’ve been following:

YC Demo Day: Health Tech Trends of 2024 Part 2

Y Combinator’s latest Demo Day gives us another preview of the latest trends in health tech. And it’s no surprise- Healthcare startups have been 100% AI focused again. This time, they are digging a little deeper into the daily fabric of medical practice.

Side note: There was only one physician in this batch of Y Combinator founders. See if you can guess which startup he worked on.

AI-Driven Clinics and Care Models

Startups like OpenClinic and HealthSpark are building AI-native care delivery systems that promise to extend provider capacity while improving patient outcomes.

• OpenClinic is helping doctors launch new clinics that are 99% AI and 1% doctor. They are helping doctors launch cash based virtual, text-only practices, and expand from there. Their goal is for the AI to do all the history taking, triaging, treatment suggestions, and follow-ups.

• HealthSpark has a similar concept but for physical therapists. They are empowering physical therapists to launch independent practices with AI automating everything from scheduling to billing.

These companies highlight a shift from traditional clinic models to hybrid AI-human workflows, where providers can scale care without scaling costs.

Automation of Administrative Workflows

Administrative bottlenecks remain a costly burden in healthcare, but startups like Helpcare AI and Docflow are tackling these issues head-on.

• Helpcare AI builds autonomous agents that schedule patients, follow up on care plans, and handle coordination tasks—all in 29 languages.

• Docflow simplifies complex scheduling for hospitals and residency programs, saving hundreds of hours of admin time.

By focusing on reducing manual tasks, these companies are not only driving efficiency but also improving the financial health of practices. Most of the value in healthcare will be created in this space.

Smart Decision-Making with AI

Companies like Astrix Health and Reticular are leveraging AI to enhance operational and clinical decision-making.

• Astrix Health helps healthcare providers optimize supply purchasing, saving clinics and health systems up to 20% on recurring expenses.

• Reticular applies explainable AI to drug discovery, allowing precise control over protein design, potentially speeding up the development of new therapeutics.

Takeaway: 

AI is no longer just a tool. It’s becoming the foundation for scalable, leaner healthcare systems. Even small practices can adopt these innovations to stay competitive.

Also, Dr. Syed worked on OpenClinic.

12 Days of OpenAI: A Glimpse into the Future

OpenAI is redefining what’s possible with its “12 Days of OpenAI,” unveiling a new product every weekday leading up to Christmas.

We’re now on Day 7, and the releases have been impressive. For doctors, the first two announcements were particularly striking. OpenAI launched the full o1 model, an AI capable of advanced reasoning, and an even more powerful o1 Pro mode. They also introduced the ability to train o1 on custom data, enabling specialization in areas like specific medical fields. The Pro mode comes at a hefty $200/month.

So far, I’ve found its medical reasoning underwhelming. However, a friend is using the Pro model to replace a team of thermal and mechanical engineers in his startup, which is impressive.

Another major release is Sora, which lets anyone create lifelike video clips with just a prompt. Free users are limited to 5-second clips, while Pro subscribers can create up to 200 seconds. This tech hints at a future where full-length AI-generated films become a reality.

One of the most exciting updates is the addition of vision capabilities to the ChatGPT app. With your phone’s camera, the AI can now “see.” This opens up possibilities like checking your form during workouts, offering home repair guidance, or even assisting with ER triage.

I tested it with some CT scans. It identified the body part and modality, but refused to analyze because “it’s not a doctor.” That’s a relief, for now.

To catch up on the first seven days of announcements, check out OpenAI’s youtube channel and tune in for Monday’s reveal.

Google AI announcements

Not to be outdone, Google rolled out its own AI updates this week, including a model with vision capabilities that can use your camera or screen.

Of course, I also tested it with some CT scans. Unlike OpenAI’s model, Google’s AI was willing to play the role of a Radiologist. However, it was barely at a medical student level. And the kind of medical student that isn’t interested in Radiology. The medical student that uses Radiology as a vacation rotation or nap time.

I also had the model watch me play chess, but despite AI’s well-known mastery of the game, it left me floundering on my own without any helpful suggestions to beat a superior opponent. This is to say that the model’s vision feature is more of a demo, capable of seeing but not yet paired with the right engines for specialized tasks, like a chess engine or medical AI trained in Radiology.

The bigger picture? Medicine is at a tipping point. AI isn’t just targeting visual specialties like Radiology, Dermatology, and Pathology. It’s capable of transforming core inputs that make up a doctor’s skillsets:

1. Perception Inputs (e.g., vision for imaging or patterns)

2. Linguistic Inputs (e.g., patient conversations, notes)

3. Data Inputs (e.g., labs, imaging results)

4. Domain Knowledge (the expertise we’ve spent years mastering)

AI can do all 4 amazingly well. In fact, vision is probably the last domino to fall, not the first. It’s not a question of if AI will transform your specialty. It’s when.

Replit Agents: AI for Coders and Non-Coders Alike

Replit is taking coding accessibility to new heights with its Agents and Assistants. For $15 a month, you get a semi-competent coding team to build apps based on simple descriptions.

It’s far from perfect, but with basic coding skills, you can keep things on track. The speed of improvement is incredible.

Case in point: I recently used Replit Agent to prototype a new feature for my app when my CTO said it’d take weeks to clear the backlog. Using the AI, I had it ready in hours.

Replit is bridging the gap between ideas and execution. When I brainstorm an idea, instead of writing it down in a wish list, I can quickly ask it to make a prototype for me even as I’m walking to my car.

I share this story because:
1. It’s crazy.
2. It’s a reminder of the risks associated with powerful AI tools, especially when released without proper guardrails. While the possibilities of AI are exciting, the potential for harm is just as real, and often neglected in the chase for profits. Something we can’t risk in healthcare.

We’re living in a moment of extraordinary innovation, where AI is simultaneously breaking barriers and exposing risks. From YC’s health tech startups to OpenAI’s groundbreaking tools, the potential is undeniable.

But as the last story shows, these tools are only as good as the people building them… and their guardrails.

Best,
Mohammed