Records found: 14
'Meta's DreamGym synthesizes environment interactions as text using a reasoning experience model and grounded replay memory, cutting real rollouts and boosting RL performance across web benchmarks.'
'Overview of six memory patterns for LLM agents across vector, graph, and event/log families, with practical tradeoffs for latency, hit rate, and failure modes.'
CMU researchers present PPP and UserVille, a multiobjective RL framework and interactive environment that teach LLM agents when to ask clarifying questions and how to adapt to user preferences, significantly improving productivity, proactivity, and personalization on benchmark tasks.
'DeepAgent merges thinking, tool search, calls, and memory compression into a single reasoning stream, enabling dynamic tool discovery across tens of thousands of APIs and improved long-horizon performance.'
'WALT converts latent website functionality into deterministic callable tools for LLM agents, boosting success rates on VisualWebArena and WebArena while cutting action counts.'
'Learn how LangChain's DeepAgents add planning, subagents, and persistent file memory to create robust, multi-step LLM workflows, with a practical example and code snippets.'
'Step-by-step tutorial demonstrating how to build a reliable insurance support agent with Parlant, including tool definitions, glossaries, journeys, and the server runner.'
'Stanford released MedAgentBench, the first large-scale FHIR-compliant benchmark that tests LLM agents in realistic EHR workflows, revealing strong retrieval skills but gaps in safe multi-step action execution.'
AgentSociety is an open-source framework enabling large-scale simulations of societal interactions using LLM-powered agents and realistic environment modeling, achieving faster-than-real-time performance.
Salesforce AI has introduced CRMArena-Pro, the first enterprise-grade benchmark testing LLM agents across complex multi-turn business tasks including sales, customer service, and confidentiality handling.
LifelongAgentBench introduces a novel benchmark for evaluating continuous learning in LLM-based agents, focusing on knowledge retention and adaptation across sequential tasks in dynamic environments.
Researchers introduce StarPO-S and RAGEN frameworks, significantly improving stability and reasoning capabilities in training autonomous large language model agents for multi-turn interactive tasks.
This article examines the critical role of AI agent protocols in enabling scalable and interoperable communication among large language model agents, highlighting key frameworks and future directions for the Internet of Agents.
AgentA/B leverages AI agents simulating real user behaviors to revolutionize traditional A/B testing, enabling faster, scalable, and resource-efficient interface evaluations on live web platforms.