Follow the Gradient: Tracing Safety Degradation Back to Training Data
Paper • 2610.37481 • Published • 32
ML Foundations
non-profit
AI & ML interests
None defined yet.
Recent Activity
View all activity
pratyushmaini
authored 9
papers 6 days ago
Natively Unlearnable Large Language Models
Paper • 2606.13873 • Published
LLM Dataset Inference: Did you train on my dataset?
Paper • 2406.06443 • Published
ÜberWeb: Insights from Multilingual Curation for a 20-Trillion-Token Dataset
Paper • 2602.15210 • Published • 1
The Finetuner's Fallacy: When to Pretrain with Your Finetuning Data
Paper • 2603.16177 • Published
20/20 Vision Language Models: A Prescription for Better VLMs through Data Curation Alone
Paper • 2605.11405 • Published
DatBench: Discriminative, Faithful, and Efficient VLM Evaluations
Paper • 2601.02316 • Published • 10
Luxical: High-Speed Lexical-Dense Text Embeddings
Paper • 2512.09015 • Published
Safety Pretraining: Toward the Next Generation of Safe AI
Paper • 2504.16980 • Published
pratyushmaini
submitted a
paper to Daily Papers 6 days ago
rulins
authored a
paper 28 days ago
vishaal27
updated a
dataset 29 days ago
rulins
submitted a
paper to Daily Papers about 1 month ago
yuhuizhang
authored a
paper about 2 months ago
ryanmarten
authored 3
papers about 2 months ago
Harbor Adapters and Harbor-Index: Infrastructure and a Curated Meta-Dataset for Large-Scale Agentic Evaluation
Paper • 2609.04298 • Published • 1
OpenThoughts-Agent: Data Recipes for Agentic Models
Paper • 2606.24855 • Published • 49
Terminal-Bench: Benchmarking Agents on Hard, Realistic Tasks in Command Line Interfaces
Paper • 2601.11868 • Published • 38
libeanim
authored 3
papers 2 months ago
LAION-BVD: A 10-Million-Hour Open Video Dataset for Multimodal Pre-training
Paper • 2608.24845 • Published • 16
DataComp-VLM: Improved Open Datasets for Vision-Language Models
Paper • 2606.28551 • Published • 52
Solving Spatial Supersensing Without Spatial Supersensing
Paper • 2511.16655 • Published