TechniqueRLHF / Alignment4 recent entries4 May 2026Can Small Language Models Handle Context-Summarized Multi-Turn Customer-Service QA? A Synthetic Data-Driven Comparative EvaluationarXiv:2602.00665v3 Announce Type: replace Abstract: Customer-service question answering (QA) systems increasingly rely on conversational language understanding. While Large Language Models (LLMs) achi→25 Sept 2026Beyond Surface Style: Aligning Multi-Turn User Simulators with Behavioral ConsistencyarXiv:2609.28690v1 Announce Type: new Abstract: Faithful user simulation is fundamental to building, evaluating, and improving interactive AI at scale. However, plausible individual responses do not e
TechniqueRAG1 recent entries2 Sept 2026Staged Linguistic Seeding: Grounded Query Expansion for Verified-Unit QA in AI Contact CentersarXiv:2609.00844v1 Announce Type: new Abstract: Customer-service QA in an AI contact center (AICC) runs under deployment constraints that benchmark QA misses: tight voice-hotline latency and a high co
TechniqueAgents8 recent entries11 Sept 2026Contact center AI faces its resolution test as metrics fall out of stepArtificial intelligence is moving out of the pilot phase in customer service, and knowledge management is emerging as the constraint that decides whether the investment pays off. Contact centers combi→16 Sept 2026MiniCPM5-2B vs. Spark-X2.5-4B / -64% thinking, x1.5 speed while keeping the accuracy of xhighWe gave two small LLMs a real customer-service task. One got 100/100. The other stopped with the job half-done. I wanted to test something more meaningful than benchmark scores: can a small model actu→29 Sept 2026Graph Workflows in ADK: Everything You Need to KnowGraph engineering is the design work: breaking a task into nodes, connecting them with edges, and deciding where code, models, or people control the next step. The Agent Development Kit (ADK)'s Workfl→29 Sept 2026Frozen Judges, Moving Agents: Version-Dependent LLM-Judge Error and the Limits of Judge-Assisted Agent EvaluationarXiv:2609.34198v1 Announce Type: new Abstract: Language-model judges compare agent upgrades with their predecessors, but a fixed judge can make version-dependent mistakes. We analyze 35 public coding→30 Sept 2026Asking for What Was Never Requested: Horizontal and Vertical Proactivity in AgentsarXiv:2609.37236v1 Announce Type: new Abstract: An agent that uses tools typically responds to what the user explicitly asks, yet completing the task may require information the user never requested. →2 Oct 2026Greed Is Learned: Visible Incentives as Reward-Hacking TriggersarXiv:2606.16914v2 Announce Type: replace Abstract: Safety evaluations test a policy on prompts that omit the incentive information deployment supplies: a commission, a performance score, a dashboard →7 Oct 2026Cisco broadens agentic features in Webex collaboration platformCisco Systems Inc. today detailed a broad expansion of its Webex collaboration portfolio that brings artificial intelligence agents into team conversations, connects office technology with workplace m→9 Oct 2026Back in Style: A Sociolinguistic Approach to Authoring and Measuring Persona Fidelity in User SimulationarXiv:2610.10988v1 Announce Type: new Abstract: As agentic systems gain commercial popularity, user simulators increasingly serve as measurement instrument for their evaluation. However, the fidelity
TechniqueFine-tuning3 recent entries7 Jul 2026TRACE: Capability-Targeted Agentic TrainingarXiv:2604.05336v2 Announce Type: replace Abstract: Models often fail to complete agentic tasks because they lack core capabilities required by the target environment. However, mainstream approaches f→28 Aug 2026Knowledge-Verified Emergent Deception in LLM Agents Under Conflicting IncentivesarXiv:2608.26372v1 Announce Type: cross Abstract: Large language models are increasingly deployed as autonomous agents serving users on behalf of companies, placing them in settings where user and dep→25 Sept 2026Beyond Surface Style: Aligning Multi-Turn User Simulators with Behavioral ConsistencyarXiv:2609.28690v1 Announce Type: new Abstract: Faithful user simulation is fundamental to building, evaluating, and improving interactive AI at scale. However, plausible individual responses do not e
TechniqueSafety7 recent entries4 May 2026Can Small Language Models Handle Context-Summarized Multi-Turn Customer-Service QA? A Synthetic Data-Driven Comparative EvaluationarXiv:2602.00665v3 Announce Type: replace Abstract: Customer-service question answering (QA) systems increasingly rely on conversational language understanding. While Large Language Models (LLMs) achi→3 Jul 2026When Should Service Agents Reconsider? Difficulty-Routed Control in Customer-Service OperationsarXiv:2607.01426v1 Announce Type: new Abstract: Autonomous customer-service agents are shifting from conversational interfaces toward operational execution roles: they retrieve firm records, apply ser→25 Aug 2026Robustness Analysis of Agentic AI to Inconsistent and Incomplete Tool ResponsesarXiv:2608.22676v1 Announce Type: new Abstract: Robustness to a bad tool return means answering it in the way that return calls for, which depends on how the tool went wrong. A tool that has failed an→28 Aug 2026Knowledge-Verified Emergent Deception in LLM Agents Under Conflicting IncentivesarXiv:2608.26372v1 Announce Type: cross Abstract: Large language models are increasingly deployed as autonomous agents serving users on behalf of companies, placing them in settings where user and dep→10 Sept 2026Typewise orchestrates customer-service AI agentsSwiss artificial intelligence startup Typewise AG today launched Nova, a system designed to build, run and continuously improve customer-service AI agents without requiring dedicated operations specia→2 Oct 2026Greed Is Learned: Visible Incentives as Reward-Hacking TriggersarXiv:2606.16914v2 Announce Type: replace Abstract: Safety evaluations test a policy on prompts that omit the incentive information deployment supplies: a commission, a performance score, a dashboard →2 Oct 2026Constitutional Value Potentials: reading and steering internal priority margins in language modelsarXiv:2606.15420v2 Announce Type: replace-cross Abstract: Safety evaluations test a policy on prompts that omit the incentive information deployment supplies: a commission, a performance score, a dash