For medical information, general AI frontier models (Google, OpenAI, Anthropic) outperformed specialized @EvidenceOpen and @UpToDate as assessed by 12 US clinicians, randomized and blinded to which model and extensive testing/benchmarks. This was not anticipated. @NatureMedicinenature.com/articles/s41591-0…
Medicine discovers the bitter lesson: frontier LLMs (here GPT 5.2, Opus 4.6, Gemini 3.1) outperform specialized "clinical AI" (e.g. OpenEvidence) in a blind test.
Even funnier that hospital IT are more likely to approve the *specialized* versions despite them being worse.
For medical information, general AI frontier models (Google, OpenAI, Anthropic) outperformed specialized @EvidenceOpen and @UpToDate as assessed by 12 US clinicians, randomized and blinded to which model and extensive testing/benchmarks. This was not anticipated. @NatureMedicinenature.com/articles/s41591-0…
Today I'm publishing a new essay, Policy on the AI Exponential. AI is progressing extremely fast—much faster than the policy process was built to handle. The essay lays out where I think the technology is now, and the action needed to close the gap: darioamodei.com/post/policy-…
Most people think NHCX is a database for health insurance claims.
It isn't. And that distinction matters more than you'd expect.
A quick thread on what NHCX actually does and why it's built the way it is. 🧵
#NHCX#CaladriusHealthAI
The innovation pipeline is moving too.
In March 2026, the NHCX Hackathon at IIT Hyderabad drew over a hundred FHIR-aligned claims submissions.
The developer community isn't waiting for mandates. It's already building.
And when something goes wrong, the system returns specific, standardized error codes rather than vague rejection notices.
That changes how billing teams respond. Less guesswork, more actionable next steps.
One of the less discussed shifts: claims status visibility.
Rather than waiting for a final response with no updates in between, providers can look up transaction status at any point through a self-service function built into the protocol.
Every NHCX transaction gets a unique correlation ID, similar to a package tracking number.
If a request fails to reach its recipient, the gateway retries automatically. If it still fails, the ID is deactivated, preventing duplicate or ghost transactions from circulating.
Right now, eligibility checks involve reaching out to each payer through their own process.
Under NHCX, providers send a single structured digital request. The gateway routes it. The payer responds through the same channel. Every time, in the same format.
#CaladriusHealthAI
Standardization rarely gets credit for the operational work it does.
But in a system where every payer communicates differently, a common protocol changes more than just the technology. It changes how work gets done every day.
Here is what NHCX introduces in practice 🧵#NHCX
All data exchange is FHIR Bundles: eligibility, pre-auth, claims, responses.
And the gateway is protocol-strict. Wrong status code, reused Correlation ID, expired token, any of these terminates the transaction.
Connecting to NHCX isn't just an API integration.
It's a shift in how systems communicate: asynchronous, FHIR-native, and protocol-strict.
A quick breakdown. 🧵#NHCX#CaladriusHealthAI
83 payers. 43278 providers. 23 million claims.
NHCX is active infrastructure, and the global playbook offers a clear roadmap for what comes next.
linkedin.com/pulse/beyond-nh…
This is an ecosystem maturing toward greater efficiency, transparency, and better outcomes for policyholders.
Every participant has a role in shaping what that looks like.
The investor signal worth watching.
India's insurtech sector has attracted significant cumulative venture funding, concentrated largely in distribution.
As NHCX matures, attention is shifting toward claims technology, data analytics, and clinical services.