Rahul Arora
@rahularoradfs
Followers
211
Following
7
Media
2
Statuses
9
I’m especially grateful for the thought and contributions of the 262 physicians from all around the world who contributed to HealthBench. Working closely with them was one of the highlights of my past year.
0
1
7
Foivos showed me how good and clean code could be. Michael and Preston, our human data colleagues, made key intellectual contributions and pulled off the feat of running four parallel physician expert campaigns at one point as we were finalizing HealthBench data.
1
0
5
@jquinonero, back to an in-the-weeds technical role, lived up to his GitHub username DrSlowPlay: incredibly careful and catching many subtle issues throughout the process.
1
0
4
@KidsBalanced brought exceptional clinical judgment and thoughtfulness, managing her intense work on this project in parallel with her full-time clinical role;.
1
0
4
I’ve learned a lot from @_jasonwei’s finely-tuned sense of what matters and judgment about how to implement it, and had so much fun pairing with him in the past weeks.
1
0
4
@thekaransinghal is an exceptional researcher, who’s both deep in the technical details and also has a clear grand vision.
1
0
4
It’s been an absurd privilege to work on HealthBench with this team, which is the strongest I’ve worked with and the most serious about making the world better.
1
0
3
This is my favorite figure from the paper. Performance vs inference cost reveals a clear Pareto frontier. Particularly absurd is the progress that we’ve made with small models: GPT-4.1 nano performs better on HealthBench than Aug 2024’s GPT-4o.
1
0
4
Very proud to share #HeathBench with the world, alongside @thekaransinghal @_jasonwei’s @KidsBalanced @jquinonero and many others.
📣 Proud to share HealthBench, an open-source benchmark from our Health AI team at OpenAI, measuring LLM performance and safety across 5000 realistic health conversations. 🧵. Unlike previous narrow benchmarks, HealthBench enables meaningful open-ended evaluation through 48,562
6
7
35