eliminating barriers to mental health. cofounder @spring_health | faculty @Yale & @Yalepsych
Brooklyn, NY
- Tweets 4,126
- Following 854
- Followers 3,282
- Likes 15,725
Pinned Tweet
our new @TheLancetPsych paper was the biggest ever study of exercise and mental health. it caused quite a stir! here's my guided tour of the paper, highlighting some of our excitements and apprehensions along the way [thread] 1/n thelancet.com/journals/lanps…
80
2,109
3,732
unbelievably bad form to joke about breaches at any time, let alone now. this will age so badly when he inevitably actually has a breach
despite the unfortunate timing, we take disclosures seriously at @typesafeai
at approximately 6:49PM this Friday evening, we became aware of a massive unauthorized leak of ALL customer noul outputs
in effort to maximize transparency, here is the link: gist.github.com/diogo149/810…
1
181
10 months ago we published VERA-MH, the first open-source AI safety benchmark for mental health. Last week we added a second scenario: harm from others.
The rubric asks whether AI can spot risk, confirm it, and guide someone to help. Open for comment now: bit.ly/4yYGjwp
2
105
Last month, we flew the whole company to NYC to mark 10 years of Spring Health!
Grateful for the people in this company. 🌿
8
242
First attempt. Full scope. ISO 42001 certified across every AI system we've built for mental healthcare.
Really proud of what we built here.
Spring Health is now ISO/IEC 42001:2023 certified — the world's first international standard for AI management systems, independently audited across the full scope of our AI for mental healthcare.
AI-native since we started. Now independently verified.
bit.ly/4iJZNzY
1
4
397
Your model is only as good as the evals you're willing to fail.
At Spring Health, we ran our crisis agent against VERA-MH and scored 76. It exposed a weak handoff between detecting risk and reaching a human. We fixed it, hit 82, and published both.
The 76 matters more.
2
209
Ten years ago April and I started @spring_health. This week we're flying the whole company to New York to celebrate a decade of it.
The idea got us started. The people in that room are the reason it worked. I'm grateful, and already thinking about the next ten.
3
16
1,430
We keep framing AI safety and access as a tradeoff, as if protecting kids means fewer good tools. That's backwards. Every headline about a chatbot harming a kid makes the tools that actually help harder to deploy.
Safety isn't a tax on innovation.
1
156
More than 240 AI mental health bills are moving across 43 states, and many contradict each other. The same chatbot can be banned in one and allowed with a disclosure in another.
Underneath the patchwork, we still lack a shared way to measure what safe actually means.
1
2
480
Glad to see the CHATBOT Act advance with protections for young users that still preserve the tools that help, like connecting someone in crisis to real support.
Protecting kids and responsible AI are complementary goals. It's the same thinking behind VERA-MH.
1
144
This month, the State of California brought @spring_health to first responders through a new partnership with @CalHR_gov.
I could not be prouder to serve the people who spend their careers carrying everyone else's emergencies.
bit.ly/4w9DSpS
2
145
You'd never sign up for a 401(k) if the balance reset every time you changed jobs. Mental healthcare has asked people to do exactly that for decades.
It's why we built Spring Health + Alma to carry care across every job and life stage.
5
418
Congress is moving to make AI safer for kids, which raises a question we should answer honestly. How do we actually know whether an AI is safe in a mental health context?
Verifiable standards matter here, and it is why we built VERA-MH.
bit.ly/4yolT0a
1
169
Research out of Stanford found AI developers lean on experts to judge whether a mental health response is "safe," and those experts rarely agree.
That's exactly why we built VERA-MH, an open and clinically grounded standard for what safe means.
stanford.io/3QUavIT
180
when someone in crisis turns to AI, the answer they get matters
~50% of US adults turned to an AI chatbot for psychological support last year, and most of those systems were never built for it
🧵 on my convo w @EstherShein for @CACMmag about this: cacm.acm.org/news/can-ai-pre…
1
2
197
as AI moves deeper and deeper into how we deliver care, this is exactly the kind of foundation the field will need.
I had the opportunity to talk with @EstherShein for @CACMmag about this: cacm.acm.org/news/can-ai-pre…
1
2
96