in your head, man!!!
Joined January 2007
- Tweets5.2K
- Following1.1K
- Followers1.9K
- Likes9K
kestrin pantera: KP2 retweeted
A small shift that’s changed how we build at Triangle: treat LLMs as functions, not chatbots or agents.
One call. Typed input. Typed output. No conversation, no loop.
We call them Functors.
kestrin pantera: KP2 retweeted
PSA: you can refer to any AI as "my buddy."
"My buddy coded it."
"I'll have my buddy drive us there."
"This is way too long, did your buddy write it?"
Try it, it'll stick.
kestrin pantera: KP2 retweeted
We are building this at Triangle Health, Patrick. I met you years ago when you came to Singapore and I was at Shopify. Let me tell you a quick story: I found out I had brain tumor through resting @prenuvo and after getting it removed I went deep on my genetics and my tumors genetics. That lead me to a treatment path where I am taking a personal cancer vaccine and an IDH inhibitor. During all this investigation I found out I have a rare genetic condition called Li fraumeni syndrome. Rare genetic disease that leads to cancer. I got luck catching my tumor early. That all said- we launched @trianglehealthx to be able to integrate ai with your health and right new we are beta testing a wgs addition to it which I have found crazy powerful.
kestrin pantera: KP2 retweeted
In the future, everyone will have a personal tool that understands your total health & what you can do next to be healthy. This is not only the best use of AI, but the single most consequential technology product that will ever exist. That's what we're building at Triangle.
kestrin pantera: KP2 retweeted
I'm a healthy member of the global elite with excellent doctors & insurance, but I still don't have full clarity on my own health or a good plan for the future. 100m Americans don't have a doctor, and 4b humans have no access to health care at all. We all need something new.
kestrin pantera: KP2 retweeted
The other half of the problem is medical knowledge.
The default assumption is that models will absorb it through training.
Clinical research, treatments, and emerging therapies don’t behave that way in practice.
They are too large, too dynamic, and too context-dependent.
Treating them as static knowledge is what leads to hallucinations and missed options.
Systems need access to live knowledge, not just trained weights.
kestrin pantera: KP2 retweeted
The future of engineering is context engineering.
Models are getting more powerful, but context is the real bottleneck.
Healthcare is a clear example.
Data is scattered across institutions.
At the same time, the full medical record for a single patient is often far beyond any model’s context window.
So only a thin slice of context gets rebuilt from scratch, and AI stays shallow.
In practice, the patient is expected to hold this context in their head and re-enter it on forms every time they see a doctor.
The fix is simple, but surprisingly hard in today’s medical system:
Context has to be accumulated over time, organized so systems can use it, and most importantly owned by the patient.
kestrin pantera: KP2 retweeted
I wonder if the middle aged dad jam band was wondering why their show was just like
Thanks to all who watched #PrettyProblems and made it the most searched film on @hulu this past month!
