Centaur.AI

Superhuman Data Annotation powered by Collective Intelligence

Also see @DiagnosUs

#AI #annotation #datalabeling #modeltraining #healthcare #medtwitter

[Centaur.AI @CentaurLabs]

Building medical AI?

Your model is only as good as your training data. We put together the complete guide to medical AI labeling in 2026. 👇

[Centaur.AI @CentaurLabs]

PM: "can we just make the model faster AND more accurate?"

ML Engineer:

[Centaur.AI @CentaurLabs]

Hot take: most ML teams don't have a model problem, they have a 'we rushed annotation to hit a deadline and now we're paying for it in every sprint' problem.

This is a safe space.

[Centaur.AI @CentaurLabs]

stop guessing if your LLM improved.

rubric‑based human evals (pairwise, safety, MQM‑style) with agreement tracking + gold sets leads to signals you can actually ship on.

[Centaur.AI @CentaurLabs]

The future of AI is in the hand of centaurs — humans and machines working in symbiosis.

And we just proved that the best way to train AI...
...is to make humans want to compete to do it.

Learn more: https://t.co/49CswZSLgx

/end 🧵

[Centaur.AI @CentaurLabs]

🧵 A crowd of gamers is beating board-certified doctors at medical diagnosis.

And Big Pharma is paying them to do it.

This is the wild story of Centaur — the company that turned data labeling into a competitive sport:

[Centaur.AI @CentaurLabs]

The use cases are expanding beyond medical imaging:

→ Drug discovery acceleration → LLM hallucination detection
→ Clinical trial data extraction → Insurance claims processing → EHR data classification

34 million text labels processed since 2021 alone.

[Centaur.AI @CentaurLabs]

what's the real cost of bad annotations?

months of model development wasted. deployment delays that cost millions.

data quality is essential.

[Centaur.AI @CentaurLabs]

@scale_AI real
https://t.co/DlUB61vLZ6

[Centaur.AI @CentaurLabs]

edge cases that kill manufacturing AI:

new shift operator holds parts differently
supplier changes material slightly
machine vibration increases 0.5%

your AI sees apocalypse. humans see thursday.

[Centaur.AI @CentaurLabs]

@parmita POV at your lab

[Centaur.AI @CentaurLabs]

the fix is more annotated edge case data.

capture every weird thing for 6 months
have QC experts label why it's weird but OK
retrain quarterly

boring? yes. works? absolutely.

what edge case is killing your AI?

[Centaur.AI @CentaurLabs]

manufacturing AI works perfectly in the lab, fails instantly on the factory floor

it's not about the model. it's about edge cases.

here's what actually breaks:

[Centaur.AI @CentaurLabs]

edge cases that kill manufacturing AI:

new shift operator holds parts differently
supplier changes material slightly
machine vibration increases 0.5%

your AI sees apocalypse. humans see thursday.

[Centaur.AI @CentaurLabs]

@awwstn the era of great launch videos continues