HealthBench Professional is designed to be: - meaningful: higher scores should lead to improvements in real-world care - trustworthy: examples verified by multiple physicians - challenging: current frontier models have remaining headroom

It uses physician-authored conversations, physician-written rubrics, multi-stage adjudication, and careful data filtering. About a third of examples involve deliberate physician red teaming, and the dataset is enriched 3.5x for the most difficult conversations for our models.

Doctors are increasingly using ChatGPT to supercharge their workflows in delivering care. To support this high impact domain, we take steps to make ChatGPT more accessible, accurate, and measurable for clinical work. openai.com/index/making-c…

Homegirl was in town for the weekend and we might have gone too crazy last night. Anyway I’m just gonna go grab a cocktail by the beach with my girl. I should be back in the morn.

Yk what you’re right. Like who even cares.

God does not reside in Roysambu

rhaenyra targaryen

The Koda balcony

One of the best tweets of all time

Everytime someone says yoga is easy I challenge them to hold a downward facing dog for 4 minutes. And suddenly they remember they have wrists 😭

I thought Padel was gon be #1😂

It’s been a great week. Jah bless.

Used to clock 48 hours lakini siku hizi hata 24 hour shift feels like I’ve been hit by a truck. What is this stage called?

I just got stuck in the elevator of my apartment when the lights went off, no phone. Longest 10 minutes of my life. I wouldn’t survive solitary confinement.

Hungry Ang

Got my coffee so now I’m flow state.

This guy used AI to put himself in Game of Thrones and fix everything

I turn everything into a joke as a trauma response 😂