hand & microsurgeon. dad. building a future for patients, physicians, and agents. bringing top 1% healthcare to everyone. https://teapawt.net/t.co/goyUm3km7l | https://teapawt.net/t.co/eE2wzDX31t
San Francisco, CA
Joined November 2024
- Tweets596
- Following376
- Followers1.3K
- Likes767
Pinned Tweet
Thrilled to be one of the 4 Voice Hack Night finalists with Surgical Triage!
I built a voice-first AI workflow that handles emergency hand surgery transfer requests, patient details, images, and next steps in one conversation. As a hand & microsurgeon, I look forward to having this agent available one day!
Huge thanks to @OpenAIDevs and @cerebral_valley
Vote for Surgical Triage → cerebralvalley.ai/e/openai-v…
What’s one way voice AI could transform your clinical workflow?
🩺 Surgical Triage
A voice-first workflow for triage: transfer requests, patient details, images, and next steps in one place.
cerebralvalley.ai/e/openai-v…
Here’s a great example of the long tail of edge cases in AI.
This Waymo ended up on a crowded, blocked-off Washington Street in SF last night during Halloween.
The car could safely stop and wait for the Waymo team to figure out a rescue. But in healthcare, stopping and waiting isn’t always an option.
Fully autonomous healthcare will eventually arrive, but in an industry where patients and physicians will demand many 9s of reliability, I suspect humans will need to remain in the loop for quite some time.
Great interview, @AriX and @JamesZmSun.
I share your excitement for computer use, particularly with WebMCP in the Codex IAB.
However, it doesn’t yet work well with Voice. I made a feature request in the post below.
Would love to see Voice + WebMCP work even better within Codex!
teapawt.net/pridgenmd/status/21028…
OpenAI's computer-using agents can now operate almost any app the way a person does, clicking and typing through software that never got an API, and the team building them says they're close to beating expert humans on speed.
Ari Weinstein works on OpenAI's consumer computer use products, and James Sun works on the enterprise side. Both were startup founders before joining OpenAI.
They join a16z's Seema Amble and Fabrizio Serafini to discuss what OpenAI's Astra model changed for computer use, the controls enterprises need for agents they can no longer watch, why computer use works as a universal API for legacy ERPs and finance systems, and why they think cheap software will push UI to get better and more personal.
0:50 What Astra changed for computer use
3:23 Consumer vs. enterprise computer use
5:43 How James runs his workday with Dots
8:06 When computer use becomes superhuman
10:01 How much autonomy to give agents
11:36 What agents mean for ad-supported sites
13:51 Computer use as the universal API
17:21 Getting enterprises to trust agents
20:22 Why the best use cases are the boring ones
25:37 Why UI isn't going away
32:38 Advice for founders building on agents
37:31 Making computer use feel coherent
40:57 Computer use's scaling moment
YouTube: youtu.be/9v42ZOdiBJY
@AriX @JamesZmSun @OpenAI @seema_amble @fabserafini
I’m excited to share the next step for my Anatomy Atlas: fracture fixation!
I built an interactive 3D model of a proximal phalanx fracture and lag-screw fixation.
Tools like this may improve surgical education and help patients better understand their surgery.
Built with AI starting from open-source 3D anatomy models.
Try it: anatomy-atlas.pridgenmd.chat…
Let me know what you want to see next.
Check out the full anatomy atlas here
anatomy-atlas.pridgenmd.chat…
I made an interactive 3D model of reduction of an anterior shoulder dislocation.
An EM doc (thanks @reverendofdoubt!) asked if I could build a teaching module for dislocations based on my anatomy atlas and recent kinematics release.
I was skeptical it would work, but I ended up pretty impressed by what Codex pulled off! @OpenAIDevs
The animation shows the humeral head clearing the anterior glenoid rim and returning to the socket during an external-rotation reduction.
Shoulder reductions are outside my hand surgery practice, so I’d welcome feedback from EM and ortho colleagues. Would this be useful for teaching? What should I correct or build next?
anatomy-atlas.pridgenmd.chat…
New update to my Anatomy Atlas: joint motion!
Flex an elbow, oppose a thumb, or turn the head. Watch the joints move and the muscles involved light up in real time.
Open the "Motion" menu to try it.
anatomy-atlas.pridgenmd.chat…
I love @hark_labs proactivity!
It noticed my CVS demo (quote-posted below) and sent me a brand new JACS study on a similar problem. I never asked it to. Can’t even find the study on Google!
And I love the one tap UI to pull the paper!
pubmed.ncbi.nlm.nih.gov/4286…
How well can a decision model recognize the three visual safety criteria used in gallbladder surgery?
I fine-tuned @CloudflareDev’s Clef-flash decision model to find out.
Clef-CVS scored among the leading results on Endoscapes. Here’s what worked and where it struggled. 🧵
Thanks, @hark_labs 🍕
Funny story… I got the wrong order! But it was the restaurant's mistake.
Usually a huge pain to fix. Instead, I gave Hark a photo and told it “wrong order”. I had a refund in 20 minutes!
We're giving away $1,000,000 of free pizza right now
Download and order from the Hark app: hark.com/download
Type "order a free pizza" in chat. It's first-come, first-served, delivered only in the U.S. (review details below)
Ask and you shall receive! Thanks for exposing this, @OpenAIDevs !
In addition to the annotations feature mentioned in the original post, you under-emphasized what I think is an equally (or more) useful functionality. Being able to press a button on a site to then launch a prompt removes friction for the user and lets the site deliver the intended prompt!
Can't wait to implement this.
teapawt.net/pridgenmd/status/20991…
Annotations paired with WebMCP give your agents more context to act on.
Check out how @kagigz uses custom annotation controls and suggested prompts to help users edit scenes and update expenses in ChatGPT’s browser.
How well can a decision model recognize the three visual safety criteria used in gallbladder surgery?
I fine-tuned @CloudflareDev’s Clef-flash decision model to find out.
Clef-CVS scored among the leading results on Endoscapes. Here’s what worked and where it struggled. 🧵
Clef-CVS still struggles when the hepatocystic triangle occupies only a small part of the image.
On validation, when two surgeons agreed the triangle was cleared, the model recognized clearance only 33% of the time, compared with 61% for the third surgeon.
Adding zoom improved the overall benchmark score but barely helped this weakness.
That's a limitation I'd want resolved before considering clinical use.
Thanks to @CammaLab and the Endoscapes team for organizing and sharing this dataset, and to @p_masca @Laparoscopes @Oz_Meireles and the SAGES CVS Challenge community for their work advancing CVS assessment.
I'm a plastic and hand surgeon so I haven't touched a gallbladder for quite some time. But I'm interested in AI and computer vision, and Endoscapes was a fantastic dataset for exploring how these new decision models do on a demanding visual task. I definitely welcome commentary and feedback from those who are more expert in gallbladder surgery than I.
There may be better architectures for clinical use, especially smaller models that use video context. This experiment was about exploring decision models.
Clef-CVS is my fine-tuned version of Cloudflare's Clef-flash (Apache-2.0).
huggingface.co/Cloudflare/cl…
Dataset: Endoscapes-CVS201, CAMMA / University of Strasbourg
github.com/CAMMA-public/Endo…
Adapted imagery/video: CC BY-NC-SA 4.0
creativecommons.org/licenses…
Cropping, overlays and animation: Brian Pridgen.
Outlines: fine-tuned Meta SAM 3 (SAM License).
github.com/facebookresearch/…
Thanks, @hark_labs !!
An airline incorrectly split up my kids on an upcoming flight. I tried fixing it. No success and loads of wasted time.
Today I put Hark on it. I enjoyed the Blue Angels and Fleet Week with my kids. Hark slogged through layers of customer service and fixed it!
To be fair, other agents may have pulled this off, too. But it was really easy to watch Hark's computer while it worked. And its responses to me were succinct, helpful, and really easy to read.