DSA Library
Daily papers
6 picked from 34 new papers
AI engineering 6
-
The Value of Thinking: Learning When to Reason in Hybrid Decision Models
Adaptive routing cut token cost by 80.1% on MiniWoB while preserving strong task performance.
-
Actor, Judge, or Teacher? Understanding Decision Models in Long-Horizon Agents
JEV trained a Qwen2.5-1.5B policy to higher success on ALFWorld, despite weak actor performance.
-
HiVe-Mem: Interleaving Horizontal Memory Expansion and Vertical Evidence Composition for Multimodal Agents
Graph-expanded retrieval and per-memory evidence selection cut token use about 90% across multimodal memory benchmarks.
-
Who Remembers What? Multi-Granularity Memory Extraction for Multi-User, Multi-Agent Interaction
Scoped memories improved pass rates while limiting cross-user leakage in mixed-user, multi-agent tasks.
-
Evidence-Guided Recovery for AI Agents with Ambiguous Tool Outcomes and Partial Execution
Planning advice certified more completions than evidence prompting under a shared shield for ambiguous tool outcomes.
-
DeepStress: Stress-Testing Deep Search Agents
Controlled evidence corruption exposed agents answering confidently from unreliable sources, including when retrieved documents were false.