Autonomous Continual Learning of Computer-Use Agents for Environment Adaptation
AI & ML interests
Natural language processing, language models, language agents
Recent Activity
View all activity
Papers
When Actions Go Off-Task: Detecting and Correcting Misaligned Actions in Computer-Use Agents
When Benign Inputs Lead to Severe Harms: Eliciting Unsafe Unintended Behaviors of Computer-Use Agents
Beyond Clicking: A step towards generalist grounding via text dragging
Is Your LLM Secretly a World Model of the Internet? Model-Based Planning for Web Agents
SAEs for vision models like CLIP or DINOv2
Generative models to produce GCG-like adversarial suffixes
-
osunlp/AmpleGCG-llama2-sourced-llama2-7b-chat
Text Generation • 7B • Updated • 57 • 4 -
osunlp/AmpleGCG-llama2-sourced-vicuna-7b
Text Generation • 7B • Updated • 1 -
osunlp/AmpleGCG-llama2-sourced-vicuna-7b13b-guanaco-7b13b
Text Generation • 7B • Updated • 20 • 1 -
osunlp/AmpleGCG-plus-llama2-sourced-llama2-7b-chat
Text Generation • 7B • Updated • 187 • 2
When Benign Inputs Lead to Severe Harms: Eliciting Unsafe Unintended Behaviors of Computer-Use Agents
-
osunlp/AutoElicit-Seed
Viewer • Updated • 361 • 129 • 1 -
osunlp/AutoElicit-Bench
Viewer • Updated • 117 • 39 • 1 -
osunlp/AutoElicit-Exec
Viewer • Updated • 132 • 6.17k • 1 -
When Benign Inputs Lead to Severe Harms: Eliciting Unsafe Unintended Behaviors of Computer-Use Agents
Paper • 2602.08235 • Published • 1
Evaluating Agentic Search with Agent-as-a-Judge
Towards Generalist Agents for the Web (NeurIPS'23 Spotlight)
Navigating GUIs as Humans Do: Universal Visual Grounding for GUI Agents (ICLR'25 Oral)
LLMs tuned on the SMolInstruct dataset for chemistry tasks.
Autonomous Continual Learning of Computer-Use Agents for Environment Adaptation
When Benign Inputs Lead to Severe Harms: Eliciting Unsafe Unintended Behaviors of Computer-Use Agents
-
osunlp/AutoElicit-Seed
Viewer • Updated • 361 • 129 • 1 -
osunlp/AutoElicit-Bench
Viewer • Updated • 117 • 39 • 1 -
osunlp/AutoElicit-Exec
Viewer • Updated • 132 • 6.17k • 1 -
When Benign Inputs Lead to Severe Harms: Eliciting Unsafe Unintended Behaviors of Computer-Use Agents
Paper • 2602.08235 • Published • 1
Beyond Clicking: A step towards generalist grounding via text dragging
Evaluating Agentic Search with Agent-as-a-Judge
Is Your LLM Secretly a World Model of the Internet? Model-Based Planning for Web Agents
Towards Generalist Agents for the Web (NeurIPS'23 Spotlight)
SAEs for vision models like CLIP or DINOv2
Navigating GUIs as Humans Do: Universal Visual Grounding for GUI Agents (ICLR'25 Oral)
Generative models to produce GCG-like adversarial suffixes
-
osunlp/AmpleGCG-llama2-sourced-llama2-7b-chat
Text Generation • 7B • Updated • 57 • 4 -
osunlp/AmpleGCG-llama2-sourced-vicuna-7b
Text Generation • 7B • Updated • 1 -
osunlp/AmpleGCG-llama2-sourced-vicuna-7b13b-guanaco-7b13b
Text Generation • 7B • Updated • 20 • 1 -
osunlp/AmpleGCG-plus-llama2-sourced-llama2-7b-chat
Text Generation • 7B • Updated • 187 • 2
LLMs tuned on the SMolInstruct dataset for chemistry tasks.