Tagged “llm agents”
Research
ARIA's Runtime Repair Proposal Targets the Error-Propagation Gap in Multi-Agent AI
A new arXiv paper proposes fixing multi-agent AI runs while they execute. The abstract reports a large repair dataset but no numbers and no enterprise validation.
Research
Can a Research Agent Tell a Good Plan from a Plausible One?
Sci$\Pi$vot asks whether an AI agent can pick a human-corrected research plan over its own. Four of five frontier models fell below chance on the selected cases.
Research
AutoOnboarder Reports Up to 29× Lower Costs, but the Abstract Leaves Key Details Open
An arXiv abstract says automatically onboarded agents matched a Gemini 3.5 Flash baseline at up to 29× lower cost on one task. Scores, run counts and cost accounting are missing.