POISE: Position-Aware Undetectable Skill Injection on LLM Agents
Paper • 2606.07943 • Published • 4
AI & ML interests
None defined yet.
Recent Activity
View all activity
Papers
POISE: Position-Aware Undetectable Skill Injection on LLM Agents
Latent Thoughts Tuning: Bridging Context and Reasoning with Fused Information in Latent Tokens
liofoilÂ
submitted a
paper to Daily Papers about 10 hours ago
liofoilÂ
authored a
paper about 11 hours ago
Post
51
We are happy to share our new survey:
Scaling LLM Agent Learning with Data Synthesis: A Comprehensive Survey
LLM agents are moving from passive chatbots to interactive systems that use memory, tools, planning, and external environments. Scaling agent learning requires more than input-output pairs: agents need synthetic tasks, trajectories, feedback signals, and environments that can support long-horizon interaction.
In this survey, we organize data synthesis for LLM agent learning around four core artifacts:
Task-level synthesis
Trajectory-level synthesis
Feedback-level synthesis
Environment-level synthesis
We also review quality control, learning frameworks, synthetic evaluation data, and applications across software engineering, agentic search, AI for science, social good, and AI safety/security.
Links:
ResearchGate: [https://www.researchgate.net/publication/406488336_Scaling_LLM_Agent_Learning_with_Data_Synthesis_A_Comprehensive_Survey](https://www.researchgate.net/publication/406488336_Scaling_LLM_Agent_Learning_with_Data_Synthesis_A_Comprehensive_Survey)
OpenReview: [https://openreview.net/forum?id=pQYwkpYmLy](https://openreview.net/forum?id=pQYwkpYmLy)
arXiv: under moderation, link coming soon
#LLMAgents #DataSynthesis #AgentLearning #LLM #Survey
Scaling LLM Agent Learning with Data Synthesis: A Comprehensive Survey
LLM agents are moving from passive chatbots to interactive systems that use memory, tools, planning, and external environments. Scaling agent learning requires more than input-output pairs: agents need synthetic tasks, trajectories, feedback signals, and environments that can support long-horizon interaction.
In this survey, we organize data synthesis for LLM agent learning around four core artifacts:
Task-level synthesis
Trajectory-level synthesis
Feedback-level synthesis
Environment-level synthesis
We also review quality control, learning frameworks, synthetic evaluation data, and applications across software engineering, agentic search, AI for science, social good, and AI safety/security.
Links:
ResearchGate: [https://www.researchgate.net/publication/406488336_Scaling_LLM_Agent_Learning_with_Data_Synthesis_A_Comprehensive_Survey](https://www.researchgate.net/publication/406488336_Scaling_LLM_Agent_Learning_with_Data_Synthesis_A_Comprehensive_Survey)
OpenReview: [https://openreview.net/forum?id=pQYwkpYmLy](https://openreview.net/forum?id=pQYwkpYmLy)
arXiv: under moderation, link coming soon
#LLMAgents #DataSynthesis #AgentLearning #LLM #Survey
ShichengfÂ
authored a
paper 10 days ago
ZhishanQÂ
authored a
paper 10 days ago
ShichengfÂ
authored 2
papers 10 days ago
MOSAIC: Module Discovery via Sparse Additive Identifiable Causal Learning for Scientific Time Series
Paper • 2605.05524 • Published • 1
Verifiable Rewards Beyond Math and Code: Lightweight Corpus-Grounded Process Supervision for Factual Question Answering
Paper • 2605.29648 • Published • 10
ShichengfÂ
authored a
paper 13 days ago
ZhishanQÂ
submitted a
paper to Daily Papers 14 days ago
ZhishanQÂ
submitted a
paper to Daily Papers 24 days ago
ShichengfÂ
updated a
model about 2 months ago
ShichengfÂ
published a
model about 2 months ago
ShichengfÂ
published a
dataset about 2 months ago
ShichengfÂ
updated a
dataset 3 months ago
ZhishanQÂ
submitted a
paper to Daily Papers 4 months ago
ZhishanQÂ
submitted a
paper to Daily Papers 5 months ago
ZhishanQÂ
authored a
paper 6 months ago
ZhishanQÂ
submitted a
paper to Daily Papers 6 months ago
ZhishanQÂ
authored a
paper 7 months ago
ZhishanQÂ
authored a
paper over 1 year ago