This week I interviewed @c_asawa, founding engineer @Glean on solving the context gap in enterprise AI We covered why context beats model power, how organizational memory compounds, the horizontal vs vertical tradeoff, and why “good enough” breaks at enterprise scale 🔗👇

5 / Often people pick 2 of mission, technology, and business. With Abridge it’s all 3, from training our own models to rapid growth – with a small team at scale. Reach out if you are hungry to improve health outcomes, costs, and lives!

4/ @AbridgeHQ is at the center of this moment. We’re building products clinicians love at systems like Kaiser, Mayo, and 200+ more. ( https://t.co/ywtHikCRrk)

3/ Today: healthcare is the leading AI adoption vertical, clinicians want these tools, and the technology is improving at an incredible pace. This convergence is rare.

2/ Years ago, I started a seminar on AI in Healthcare ( https://t.co/AV5Mi1lvA2). The tech was impressive – yet over the years almost none of it made it into real clinical practice. That gap is finally closing.

1/ Excited to join Abridge to lead Engineering for AI Clinical Decision Support! AI in healthcare is finally crossing the gap from “impressive demos” to real clinical impact.

Training our advisors was too hard, so we tried to train black-box models like GPT-5 instead. Check out our work: Advisor Models, a training framework that adapts frontier models behind an API to your specific environment, users, or tasks using a smaller, advisor model (1/n)!

Instruct-tuned models are getting better at following instructions and ‘reasoning’ every day, but they’re shockingly poor at generating diverse responses. Diversity is crucial to many tasks like synthetic data generation. We tackle this with a new approach, BARE 🐻! (1/n)

I wonder if MAA problem writers know that the 30 problems they wrote for this week and the next are not just for USA(J)MO qualification but will benchmark AI progress for the next year (“AIME25”) 🙃

Today we're open-sourcing TensorZero: a platform that helps LLM applications graduate from API wrappers into defensible AI products. 1. Integrate our model gateway 2. Send metrics or feedback 3. Unlock compounding improvements in quality, cost, and latency

@jainarvind @AnthropicAI @awscloud @rosenthaltweet @carterbusse @Workato @billcom @belril Now, let’s take a look into the future of the Glean Platform with @c_asawa!

Thank you @vkhosla for your precious time in speaking in this series again!! And further thanks for helping promote the videos the organizers worked so hard for! Students are always very impressed when you speak and your thoughts on really pushing the edge!

Learning concepts through sensorimotor interactions is important for general AI. Check out our AAAI18 paper and blog on this topic:

Also many other interesting talks from @LakeBrenden, Tomas Mikolov, John Langford and others... https://t.co/Vy2ssflzjc #hlai2018

When I posted this earlier the presentation slides were broken. They have now fixed this problem, and it should make for a more enjoyable viewing.

Check out @AndrewYNg's incredibly insightful opening talk at Stanford's CS 522: AI in Healthcare ( https://t.co/8SZBmZ78Ms)! Notes: