Privasis: the Largest "Public" Private Dataset
➡️TLDR: we synthesize data with private information and train a open PII scrubber model🤫
⏰Thu, Jul 9, 2026 • 2:30 PM – 4:15 PM KST
📍HALL A Poster Location: #3105
nitter.cf/hyunw_kim/status/20191…
🚨New paper to level up your 🦞#Clawdbot ?!
Bots are now posting your sensitive info in real time. But privacy research is a desert with no data to train better models. That's about to change
Enter 🏝️Privasis, the oasis where you can train strong privacy-forward AI with scale✨
Cool approach to deal with fragmented PII snippets across context
🚨New paper to level up your 🦞#Clawdbot ?!
Bots are now posting your sensitive info in real time. But privacy research is a desert with no data to train better models. That's about to change
Enter 🏝️Privasis, the oasis where you can train strong privacy-forward AI with scale✨
I’m hopeful that bots posting sensitive data will receive more research attention such as below. I understand how useful they can appear to be but you are giving away sensitive data.
🚨New paper to level up your 🦞#Clawdbot ?!
Bots are now posting your sensitive info in real time. But privacy research is a desert with no data to train better models. That's about to change
Enter 🏝️Privasis, the oasis where you can train strong privacy-forward AI with scale✨
We trained a 4B model that's state-of-the-art at sanitizing sensitive data -- so you can run it locally before uploading potentially sensitive data to cloud services. Check out the work (and the underlying dataset), led by @hyunw_kim and @niloofar_mire, below!
🚨New paper to level up your 🦞#Clawdbot ?!
Bots are now posting your sensitive info in real time. But privacy research is a desert with no data to train better models. That's about to change
Enter 🏝️Privasis, the oasis where you can train strong privacy-forward AI with scale✨
Privacy research has a fundamental data problem: by definition, private data is "private." There is virtually no large scale public training data to work with.
🚀 Second release of the week
We show how to synthesize the Largest "Public" Private Dataset from scratch!
💡The algorithm requires zero seed data beyond a first name...
Check out our paper to learn how! amazing work led by @hyunw_kim
🚨New paper to level up your 🦞#Clawdbot ?!
Bots are now posting your sensitive info in real time. But privacy research is a desert with no data to train better models. That's about to change
Enter 🏝️Privasis, the oasis where you can train strong privacy-forward AI with scale✨
Data privacy has never been more important, esp with all the personalization and ADs👀 that LLMs are generating🚨 We introduce a sensitive data generation pipeline to empower further privacy research, and we release Privasis-Cleaner, a sanitation model outperforming SOTA llms🌟
🚨New paper to level up your 🦞#Clawdbot ?!
Bots are now posting your sensitive info in real time. But privacy research is a desert with no data to train better models. That's about to change
Enter 🏝️Privasis, the oasis where you can train strong privacy-forward AI with scale✨