The authors formalized human-AI collaboration as a sequential decision problem in which human judgments are reliable but costly.
CARE learns from human review while preserving calibrated risk control
The framework updates an AI system from selectively acquired feedback and recalibrates after model changes, reducing human review across four safety-critical datasets.
Big Tech
Chenyu Zhang · Rachel Luo · Boyi Li · Anjali Parashar · Marco Pavone · Apoorva Sharma
MIT · NVIDIA · Stanford University
Research Digest··3 min read
Zhang et al.
Why this paper
From NVIDIA and 2 others
In one line
CARE maintains risk control during online learning from selective human feedback while cutting human-review queries by 25-81% across four safety-critical datasets.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§