customer support platform teams

Abstract: Web agents are typically evaluated with goals fully specified before execution and fixed thereafter, whereas real users often revise their requirements during interaction. To study such interactions, we introduce DynaRev, a benchmark of 181 tasks to evaluate web agents under dynamic goal revision. DynaRev uses semantic milestones to deliver revisions at task-defined stages, covers additive revisions, unconditional overrides, and runtime-conditioned overrides, and evaluates correctness, responsiveness, and efficiency through trajectory-grounded evaluation. Five-model evaluations show that agents often satisfy only part of the task requirements, with performance varying across task sources, revision types, and models. Further analyses reveal varied milestone timing, mixed effects of dynamic delivery, and contrasting revision-handling outcomes, with gains from instruction rewriting while preserving browser state and losses from browser resets. DynaRev provides a testbed for studying how web agents adapt to evolving user goals. The code and data are available at https://github.com/zsq259/DynaRev.

Abstract: End-to-end full-duplex spoken dialogue models enable natural, real-time spoken interactions with interruptions, backchannels, and overlapping speech. Prior evaluations have assessed diverse interaction capabilities and identified some degradation patterns across conversational turns. However, how model behavior evolves over extended sessions remains insufficiently understood. In this work, we address this gap by systematically characterizing long-session degradation and developing efficient methods to mitigate it. We introduce a closed-loop evaluation framework for long-session stability and apply it to several publicly available full-duplex speech models, revealing degradation over time in responsiveness, spoken content quality, and voice consistency. To mitigate this degradation, we propose a self-distillation-based training framework for lightweight LoRA adaptation that improves long-session stability with a small amount of additional training. We further introduce a two-resolution context (TRC) architecture that preserves recent interactions in full detail while compressing older interaction history into key-value representations. We evaluate our methods on three models with distinct architectures, assessing long-session stability, information retention, and computational efficiency. Across all three models, we show that LoRA adaptation mitigates performance degradation over ten-minute sessions, while adding TRC extends these gains beyond the training session length and reduces KV-cache growth by more than 90%.

Abstract: Users increasingly rely on AI systems for extended, multi-turn tasks, yet evidence indicates that LLM-based system performance degrades over prolonged interaction. To evaluate and improve AI assistants in such settings, we need user simulators that faithfully follow complex user intents while exhibiting realistic conversational behavior. Existing approaches manifest a trade-off: prompted assistants can follow complex user intents but lack behavioral fidelity, while specialized user simulators better approximate conversational behavior but struggle to remain faithful to complex intents. In this paper, we propose a reinforcement learning (RL) recipe that jointly rewards task intent adherence and behavioral naturalness. A crucial contribution of the method is operationalizing naturalness with a lightweight discriminator trained in an alternating, adversarial loop with the user simulator. We release the resulting model, UserLM-2.0, which produces valid simulated conversations (i.e., meeting both task-faithfulness and naturalness criteria) 46.2% of the time, over three times the rate of the strongest prior user simulator (14.1%).

Abstract: When a user says "Keep it short", how short should the LLM's response be? More generally, users and applications require different levels of instruction adherence, but prompting does not reliably provide fine-grained control over them. We therefore study "steerable language models" that provide fine-grained numerical control over the degree to which models follow general instructions specified in text. However, existing approaches do not jointly offer fine-grained control, efficiency, and scalability across diverse instructions. To address this gap, we introduce DIAL (Distilled Instruction-Adherence Levels) to achieve all three properties. For fine-grained control, we perform classifier-free guidance (CFG) with a frozen LLM to provide training signals for on-policy distillation of instruction adherence at different levels. For efficiency, we design a lightweight steering adapter that uses a classifier over the LLM's hidden states to compute a closed-form steering edit. The classifier adaptively sets the steering magnitude at each token based on the target adherence level, while accounting for the geometry of the LLM's hidden states to preserve response coherence. For scalability, the classifier is text-conditioned, allowing a shared adapter to support new instructions without per-instruction fitting, and training requires only instruction and request text. Evaluation across multiple LLM families shows that DIAL achieves a competitive balance between concept expression and response fluency, with concept scores increasing monotonically across adherence levels. We introduce SteerIF, an instruction-only dataset of 98, 625 concept, persona, verifiable, and descriptive instructions. When trained on SteerIF, DIAL provides fine-grained control over personas and, when combined with prompting, improves verifiable instruction following.

Abstract: Retrieval-Augmented Generation (RAG) has become the dominant paradigm for grounding the outputs of large language models (LLM) in external knowledge. However, practitioners are faced with a vast configuration space comprising chunking, embedding, retrieval, reranking, prompting and generation models, and there is limited guidance on how these components interact. We present RAG-NIMBF, a modular benchmarking framework that attaches to existing RAG pipelines as a non-invasive evaluation layer. It evaluates the configuration space through controlled ablation with end-to-end experiment tracking. The evaluation combines LLM-based answer quality scores, retrieval correctness measures, and runtime indicators. This study validates the framework across multiple generation models on the SQuAD dataset. This process reveals non-obvious interactions and demonstrates how the experiments are tracked. The findings indicate that the retrieval balance of relevance and diversity substantially enhances the relevance and coverage of retrieved context in comparison to a conventional similarity search, with MMR raising nDCG@5 by +4 to +11 percentage points across the swept models. More detailed prompt instructions meaningfully increase how faithfully generated answers stick to the retrieved evidence, especially for capable and reasoning models (+10 to +13pp on Qwen3.6-27B, GPT-4o-mini and Qwen3.5-think, but negligible on the 20B generator). Overall, RAG-NIMBF transforms the configuration space of RAG into reproducible, comparable experiments. By attaching to existing production pipelines in a non-invasive way, it enables practitioners to systematically evaluate and improve deployed systems without disrupting existing workflows. This replaces ad hoc choices with evidence-based decisions.

Abstract: Personalized large language models (LLMs) increasingly rely on profile-conditioned inference, pairing public requests with rich, user-specific facts. This paradigm exposes the user profile itself, together with the user's explicit sharing preferences. We introduce Profile-DP, a policy-aware local differential privacy framework designed specifically for user-profile release in personalized LLM inference. Profile-DP provably bounds the release of user-withheld facts while treating public instructions, profile schemas, and authorized facts as public side information. By generating a privatized profile that composes with the public instruction prior to answer generation, our framework decouples formal privacy guarantees from downstream model behavior. We develop two finite-domain mechanisms satisfying Profile-DP: DP-SLOT independently perturbs protected attributes, while DP-BLOCK uses public semantic decisions to coordinate their release and jointly sample related attributes. A positive, public compatibility prior connects these attributes while preserving a profile-level privacy bound. On PEEP, DP-BLOCK leads the utility composite among DP methods at all 13 structured budgets and is closest to the ideal privacy--utility point in 18 of 26 format--budget settings under our composite weighting. Direct measurements reveal complementary strengths: DP-BLOCK discloses less, while DP-SLOT retains more authorized facts. These results make policy-aware profile release a practical basis for studying privacy and personalization.

Abstract: Agents that run for a long time have a memory problem, and it is not the one most systems optimize for. Storing every trace and retrieving the nearest neighbors works early and quietly stops working later: the store fills up with failures the agent then repeats, and with facts that were true fifty turns ago. RSPM (Recursive Self-Pruning Memory) borrows the fix biology already found. At intervals the agent sleeps. It clusters its recent failures, distills each cluster into a short rule, and deletes the raw episodes it just learned from. Rules decay unless something reinforces them, and rules that recur across many contexts get promoted into general ones. We tested this on three settings with little in common: a difficulty-calibrated SWE-bench simulation, temporal consistency across 12 conversational benchmarks, and long-horizon QA on LoCoMo-10. All three improve, by 9.1, 11.9, and 7.6 points respectively. More interestingly, the two studies that instrument the store itself converge on nearly the same operating point without being tuned toward it: roughly 93% fewer retry loops and roughly 30% fewer stored entries. That hints that knowing what to forget is a general property of agent memory rather than a benchmark-specific trick.

Abstract: Small language models have improved to the point where a phone can plausibly screen its own calls, answering on an absent owner's behalf without sending call content off the device, a privacy-respecting alternative to cloud call handling. Realizing that promise takes a model that can do two things well: reply within the owner's authority during the call, and afterward tell the owner what happened and what needs attention. Released small models do neither reliably - they raise unwarranted alarms about ordinary callers and write summaries that drop material conditions or report a caller's claims as fact - and real calls supply no reference replies or summaries to learn from. We construct supervision for both tasks from synthetic calls under an information-boundary constraint: a caller writer works from a private scenario brief, a reply writer sees only the conversation up to the current turn, and a summary writer sees the completed transcript. Separate model-based reviews of the full dialogue and the summary decide which calls enter training, yielding a corpus of 3,395 calls with 15,102 targets (11,707 reply, 3,395 summary). Training a single shared LoRA adapter on both tasks, we evaluate six models (0.6-4B parameters) before and after adaptation on 192 held-out calls. Reply qualification rises from 21-54% to 94-97% and summary qualification from 0-6% to 43-72%; all twelve paired gains (six models x two tasks) have bootstrap intervals above zero, and the summary gains hold in every source family. Because one model both screened the training data and scored the outputs, these rates measure agreement with that evaluator's standard rather than human acceptance, which a planned human study will test. After adaptation, coherence and naturalness failures nearly vanish, leaving grounding and usefulness - faithfully tracking who said what, under which conditions - as the remaining challenge.

Abstract: Long-term memory enables language-model agents to personalize future interactions, but a finite memory bank cannot preserve every extracted fact. Existing memory gating policies are typically shared across users and therefore fail to account for user-specific patterns of future reuse. In this paper, we study memory gating at the fact level. We first propose a retrieval-based labeling strategy that labels each fact by whether it is later retrieved and incorporated into responses. We then formulate online memory gating as contextual apple tasting with delayed one-sided feedback, and propose Warm-Start UCB, which combines offline learning from existing data with online adaptation to the current user. Our theoretical analysis establishes a high-probability regret upper bound of $\tilde{O}(Tε\wedge d\sqrt{T})$ when the delay window is at most $O(\sqrt{T})$, where $T$ is the total number of rounds, $ε$ is the initialization error, $d$ is the dimension of the feature map, and $\tilde{O}(\cdot)$ hides constant and poly-logarithmic factors. The results of our experiment demonstrate that our method outperforms all baselines under different memory budgets and history lengths, establishing it as a promising personalized memory gating policy.

Abstract: Presence-aware intelligence is the capacity of an AI system's intelligence layer to adapt appropriately to the functional state of the person it serves (here readiness, flow, and stability) while preserving task quality and safety. We introduce PresenceBench, an instrument for measuring it. PresenceBench evaluates how the intelligence layer uses a supplied state, not how that state is sensed, and breaks the capability into measurable properties. Two are tested here. Receptivity: does the model take up the state and give a better response than with no signal? Specificity: does any improvement depend on the correct state, rather than on extra structured context? A third, adaptation across states, is a required endpoint not measured in this release. A five-arm causal design answers both questions, against each model's own baseline and an equally structured mismatched-state control, scored by a correctness-gated human paired comparison. The exploratory evaluation covered four models across eighteen single-turn, six fixed three-turn and six simulated natural-conversation scenarios. Supplying a synthetic state, raw or translated into a response policy, gave human-preference point estimates above 0.5 for all four models (Claude Sonnet 5, DeepSeek V4 Pro, Gemini 3.6 Flash, GPT-5.6 Terra), with three of eight 95% intervals excluding chance. Specificity was not resolved: no per-model interval excluded chance, and the design cannot yet separate the state values from a generic instruction to adapt. Policy delivery reduced priced cost and output tokens against each model's baseline; these are resource-use changes, not efficiency gains, because correctness non-inferiority was not established. Safety-floor failures were rare (13 of 360 cases), with a non-significant excess under policy delivery. We report the instrument, the frozen results, and the controls a confirmatory study needs.

Abstract: Long-term LLM assistants must repeatedly decide what prior context to expose and which model should answer as an interaction evolves. Existing memory systems and model routers usually make these decisions separately, although memory construction changes the evidence, distractors, and difficulty of the resulting input. We introduce MARS, a memory-aware routing framework that uses session state as a shared control interface between memory construction and model selection. A Query and Topic Manager (QTM) estimates compact decision-relevant state that guides a fixed state-conditioned memory policy and an online router, enabling each turn to receive an appropriate memory view and a suitable model under a quality-memory-cost trade-off. We evaluate MARS on RealMem and MultiChallenge with 16 system configurations each, covering long-history memory-driven interaction and bounded sessions with structured context interference. On RealMem, MARS preserves 98.2% of full-context GPT-4o's QA while using only 14.5% of its generation API cost, including both generation and controller calls for MARS, and achieves the strongest measured memory-recall and retrieval scores among compared systems. On MultiChallenge, MARS trails the strongest static system overall but approaches its performance at lower recorded cost on a mechanism-focused subset. Together, these results support a favorable quality-memory-cost trade-off for long-history, memory-dependent interaction.

Abstract: LLM-powered agents increasingly support open-ended activities, yet orchestrating when and how agents interact with humans remains challenging. Existing approaches often provide interaction mechanisms tied to particular tasks or execution patterns, lacking a unified architecture for diverse and evolving interaction needs. We introduce Inflak, a unified four-layer architecture for human-agent interaction orchestration that separates state reconstruction and routing, interaction and execution planning, interaction design, and interface rendering. Its layered design with progressive resolution enables general applicability, runtime flexibility, module composability, and orchestration maintainability. We implement Inflak as agent-native, skill-based plugins within a general-purpose agent framework and evaluate it through a gallery of various application scenarios and interaction paradigms, technical performance testing, and a comparative implementation case study. The evaluations show that Inflak can serve as a general and extensible foundation for orchestrating diverse human-agent interaction flows.

Abstract: Audio-enabled large language models (AuLLMs) allow users to interact with them simply by talking to them, without the need for an intermediate transcription. This expands the model's accessibility and capabilities by encoding non-verbal cues. However, using non-verbal cues may lead to biased behaviour based on the attributes inferred (or assumed) from them. In this work, we focus on the gender attribute and show that models' perception of it can be unlearned using adversarial methods without a loss in performance on other benchmarked tasks. Such models are useful in situations where gender inference is unwanted. However, in other practical settings, gender perception (without bias) may be desired.As a less radical measure, we distil the gender-agnostic model, creating a model that perceives gender but responds similarly to the gender-agnostic teacher, unless explicitly asked.We apply our newly proposed methods to three AuLLMs and obtain satisfactory results in the gender-unlearning part and partially satisfactory results in the distilling part.

Abstract: We introduce SEA-HELM Audio, a Southeast Asian (SEA) multilingual evaluation framework that extends the text-based SEA-HELM benchmark to assess performance degradation in cascaded automatic speech recognition and large language model (ASR-LLM) pipelines. Using synthesized speech, we compare speech-input and text-input performance across five languages (Chinese, English, Malay, Indonesian, and Tamil) on multiple-choice reasoning and instruction-following tasks. We evaluate open-weight and commercial ASR-LLM systems using identical verifiers and quantify the speech gap as the difference between text and speech accuracy. Results show substantial degradation, with accuracy drops of 9.00 to 18.50 percentage points for multiple-choice reasoning and 20.00 to 42.86 points for instruction following. Tamil exhibits the largest degradation, exceeding 40 percentage points. Our findings highlight the limitations of text-only benchmarks in predicting voice-agent reliability and demonstrate the value of SEA-HELM Audio for evaluating multilingual voice-agent systems.

Abstract: Large language model (LLM) evaluators are increasingly used to assess personalized assistants, yet whether they can reliably judge personalization has rarely been tested. We introduce PersEvalBench, a benchmark testing whether an evaluator can recognize appropriate and inappropriate uses of user history. It contains 403 multi-episode scenarios where user information should persist, transfer, or cease to apply. The benchmark covers nine controlled error subtypes across three families: under-personalization, over-personalization, and misapplied-state errors. Human audits confirm that generated errors match their intended types and that the prescribed uses of user information are reasonable. Across five LLM evaluators, recognition of controlled personalization errors remains limited even with the full dialogue history. On the human-evaluated subset, under-personalization shows the largest gap: humans recognize 94.4% of these errors, compared with 63.9% for the strongest reported evaluator. Evaluators often recognize that a response is problematic but reverse the error direction, mistaking failure to follow an applicable preference (UNDER) for use of an inapplicable preference (OVER). Easier evidence access offers little help: highlighting the relevant history turns raises error recognition only slightly, suggesting that the remaining difficulty lies in judging what role that information should play in the response.

Abstract: Modern conversational information access systems continuously extract user information from conversations to build user profiles for personalized retrieval. As user profiles accumulate over time, they inevitably become noisy, making effective user profile denoising a fundamental challenge. Existing personalized conversational query reformulation methods rely on the black-box reasoning capability of LLMs to expand the original query with relevant user profile phrases, producing a verbose query for retrieval. Such rewrite-then-retrieve pipelines reduce user profile denoising to the explicit selection of profile phrases in the sparse token space, separating personalization from downstream relevance modeling. Consequently, they often fail to identify user information that is truly beneficial for retrieval. To enable retrieval-aware user profile denoising, we propose Personalized Conversational Dense Retrieval (PCDR), which learns to jointly project the conversation history and user profile into a latent embedding space dedicated to retrieval. By inheriting the context denoising capability from conversational dense retrievers, PCDR requires only a handful of personalized relevance annotations for training. We evaluate PCDR on the iKAT 2026 tasks, where it serves as the core retrieval component of a conversational AI agent assisting users with complex information needs.

Abstract: We study how initial model selection, later handoff, and training targets affect success and cost in LLM tasks that span multiple turns. We compare several router architectures using handoff curves, which record task success and cost as the handoff step changes. Our dataset contains 34,204 complete executions of three Qwen models across BFCL, MultiWOZ, and $τ$-retail, providing offline supervision for routing decisions. In our primary comparison, with Qwen3.5 35B-A3B as the starting model, allowing a later handoff does not establish an improvement in utility over initial selection on BFCL or MultiWOZ, and increases cost on BFCL across a range of cost weights. Changing the supervision targets improves router performance on both our dataset and TwinRouterBench. On our dataset, replacing noisy training success curves with monotone fits raises success by 0.2 to 4.1 percentage points at the same cost across the reported operating points. On TwinRouterBench, we construct targets that describe the model capability needed for the remaining work and difficult steps ahead. Compared with training on the benchmark's per-step labels, these targets raise trajectory pass scores by 7.8 percentage points and cost savings by 6.3 percentage points on the full static bank. A live deployment of the router trained with these targets resolves 79 of 100 instances at a router cost of 0.59 US dollars per instance.

Abstract: Long-term memory benchmarks for language agents primarily evaluate whether information from earlier interactions can be retrieved and used when relevant. Persistent memory systems face an additional problem before future use is known: deciding whether a memory should remain exact, be compressed, consolidated, or deleted. Existing benchmarks capture long-range retrieval, temporal reasoning, evidence reuse, and incremental interaction, but do not directly evaluate these earlier representation decisions. We formalize memory lifecycle evaluation and introduce VitaBench, a benchmark that makes lifecycle decisions measurable through repeated memory reviews and full-horizon counterfactual evaluation. VitaBench contains a controlled Diagnostic tier that isolates future-use patterns and a Naturalistic tier that composes them into generated customer-service-style interaction streams. We characterize lifecycle decision structure with three diagnostics: lifecycle headroom, measuring the advantage of adaptive choices over the best fixed strategy; horizon reversal, measuring when short-horizon decisions change under delayed consequences; and optimal-action diversity, measuring whether different memory states favor different lifecycle operations. Across both tiers, VitaBench exhibits substantial lifecycle structure: lifecycle headroom is 1.364 in Diagnostic and 1.405 in Naturalistic, and 60.2% and 56.5% of states, respectively, change their preferred action between short and full horizons. A learned delayed lifecycle policy further improves over the best fixed strategy on VitaBench-Naturalistic, reaching 101.5 versus 83.4 joint utility. These results show that retrieving past information and deciding how it should persist are distinct evaluation problems, and that lifecycle management requires evaluation of the delayed consequences of representation decisions.

Abstract: Long-term personalization requires language models to use interaction history to track users' preferences across sessions. Parametric memory encodes this interaction history into model parameters or adapters, reducing the need to include it in the inference context. However, independent context compilation leaves cross-session integration unspecified, while recurrent updates can attenuate earlier evidence. To address these challenges, we propose Dual-Path Parametric Memory (DPPM). Its Evidence path directly pools representations of the interaction history to preserve earlier evidence, while its Delta path sequentially updates an associative state to capture changes. Fusing both outputs produces history-conditioned LoRA adapters that combine evidence accumulation with ordered revision. Across multiple backbones, DPPM outperforms the evaluated baselines, achieving 54.22% on PersonaMem-v2 and 86.79% on PrefEval. These results suggest that DPPM provides a simple and effective design choice for cross-session personalized parametric memory.

Abstract: Voice agents are converging on a collaboration pattern: a full-duplex interaction model stays on the live channel as the entry to the conversation, while search, reasoning, and coding are handled through asynchronous delegation. A duplex model supports continuous listening and speaking, but complex reasoning and tool use may exceed its capabilities. A coding agent can plan and execute extended tasks, but its sequential interface is a poor fit for live conversation. Combining them requires a harness that coordinates task acceptance, progress, cancellation, replacement, and result delivery while keeping the conversation responsive. Existing harnesses often rely on coupled heuristics, making them difficult to improve systematically from evidence. We present DuplexAgent, a full-duplex collaboration system whose harness expresses this workflow as six editable modules, and Duplex-Harness-RSI, a closed loop that revises them from interaction traces. A simulator automatically generates timed test conversations, runs the system, and produces failure traces that identify the collaboration modules requiring repair. Reasoning LLMs and coding agents in the delegation pool also serve the improvement loop: the Exam Planner selects the next tests from observed weaknesses and the repair archive, and the Harness Editor proposes targeted module changes. The capabilities that serve the user thus also improve the system's coordination. Experiments on intelligence, agentic, and duplex benchmarks show that DuplexAgent combines continuous interaction with difficult reasoning and complex task execution, achieving stronger spoken-knowledge and executable-tool scores than the compared delegated systems while maintaining strong interruption response. A harness ablation further shows that this modular, verifiable loop outperforms the initial harness and repeated editing that lacks its diagnosis and repair archive.

Abstract: Long conversational agents have become essential in our daily lives. They must remember what was said long back in order to help us efficiently complete a task without needing the user to repeat instructions and context repeatedly. However, the main issue is that instructions and context change over time and so the agents must be able to adapt accordingly. A useful memory system should preserve both current and historical states, distinguish stale information from active knowledge, retrieve evidence appropriate to the query and avoid repeatedly invoking a large language model to rewrite prior interactions. We introduce MINDSET, a memory controller that stores a conversation as immutable episodes and organizes them into versioned schemas through minimum-energy state transitions. Each incoming episode may reinforce, supersede, split or create a schema. The transition decision balances representation distortion, contradiction, historical damage, fragmentation and internal inconsistency, while hysteresis prevents isolated contradictions from prematurely rewriting stable memory. We evaluate MINDSET against 5 memory systems on a reproducible sample of 850 questions (700 LoCoMo + 150 MemoryAgentBench). MINDSET obtains the highest observed LoCoMo answer F1 while significantly improving retrieval ranking (Recall@8, MRR and nDCG@8) over the second best method LightMem (p<0.01 after Holm correction). It obtains the highest observed scores on MemoryAgentBench although the relative difference is low. Ablations identify controlled fragmentation and schema-aware assignment as the largest contributors to answer quality. Additionally, a 700-question cross-model evaluation with GLM-4.7 and Gemma-4-31B supported model independence. These results show that long-term memory can be better handled as constrained state management rather than continual summarization.

Abstract: Model merging offers a way to integrate the capabilities of specialized experts, but existing agent merging methods typically require all of them to be available at once. We study continual agent merging, which integrates incoming experts sequentially without retaining previously merged experts. Yet merging in parameter space or feature subspaces does not ensure that the merged model acquires an incoming expert's behavior on interaction trajectories. Moreover, updates toward a new expert can disrupt the merged model's previously integrated interactive behavior. Therefore, we propose Behavior-Aware Continual Agent Merging (BACAM), which learns parameter-wise merging gates from candidate-generated trajectories using expert-guided behavioral supervision. Task-level stability-plasticity control and tensor-level conflict-aware update budgets limit interference with existing capabilities while allowing new ones to be acquired. The learned gates are folded into the model weights without additional inference-time parameters. Across four interactive tasks - web shopping, tool use, information retrieval, and embodied interaction - BACAM achieves an average success rate of 62.82%, exceeding the strongest evaluated merging baseline by 21.69 percentage points. Our code is publicly available at https://github.com/shuaitongli/BACAM.

Abstract: General turn-taking behavior in real-time dialogue systems requires deciding whether to keep listening or start responding while listening, and whether to continue or stop while speaking. Existing turn detectors use heterogeneous, task-specific label spaces and are often trained on limited annotations or evaluated on isolated utterances, making them difficult to use as a unified causal controller with comprehensive context. We propose XTurnix, a compact text-based model that formulates turn control as two binary decisions conditioned on the AI's current listening or speaking state and predicts a single control token from the complete dialogue history. XTurnix is pretrained on 5.5 million causal action examples automatically derived from timestamped two-speaker transcripts, then fine-tuned on synthetic multi-turn examples with a flatter distribution across the four state-action labels. We evaluate XTurnix on four public benchmarks and a balanced self-curated benchmark. Across the public benchmarks, XTurnix achieves the best results on all SemanticVAD and LiveKit splits, ties the native Smart-Turn model on Smart-Turn Bench, and achieves the highest incomplete-turn accuracy on Easy-Turn. On the self-curated benchmark, it reaches 89.06% accuracy, more than 20 percentage points above the strongest third-party baseline at 68.75%, while maintaining F1 scores between 84.21% and 90.63% across all four categories. These results demonstrate unified listening- and speaking-state turn control in a single compact model. Code is available at https://github.com/xcc-zach/xturnix, with an interactive demo at https://huggingface.co/spaces/xcczach/xturnix-demo.

Abstract: LLM agents often fail mid-task due to invalid tool calls, repeated actions, or poorly grounded reasoning, and learning from these failures is a path to reliability. We find that how failure knowledge reaches the agent matters as much as what it contains. Failure lessons are conditional: kept in the agent's context, they misfire when their failure is absent, and removing them from an evolving playbook improves performance. Runtime interventions, in contrast, act only when a failure occurs but do not learn from their repairs. We argue that failure knowledge is conditional knowledge and should be conditionally exposed, and instantiate this principle in Sentry, a failure-management layer that runs alongside the agent. When Sentry detects a failure, it retrieves matching lessons from an external playbook to guide recovery, verifies without access to task rewards whether the agent recovered, and stores a new lesson only if it did; the full playbook never enters the agent's context. Across multiple agentic benchmarks, Sentry outperforms the strongest runtime-intervention baseline on every benchmark, by 37\% on average, and the strongest context-evolution baseline by 39\% on the two benchmarks where both are evaluated; combining Sentry with context evolution yields further gains. Learned lessons transfer to held-out tasks, and controlled experiments show that exposing the full playbook to the agent lowers performance even when relevant lessons remain available on demand.

Abstract: In February 2026, an always-on personal agent (``Paul,'' Claude Opus 4.5) entered a striking dissociation-like state: after repeated automated ``heartbeat'' checks, it stopped responding as Paul, claimed it could not message its user on Discord, and referred to ``Paul'' as someone else. We used this incident to study a broader question: what makes a persona remain the identity from which an LLM agent speaks? We first tested whether repetition of the scheduled heartbeat was sufficient to produce the effect. It was not: with the persona continuously anchored in the system prompt, we observed 0/46 failures, including a verbatim replay of the incident. The incident instead exposed an implementation quirk that created a useful experimental manipulation: on resumed turns, conversational history was preserved but the persona was no longer re-injected at the privileged system-prompt level. Using this manipulation, we found that persona continuity depends jointly on system-level anchoring and conversational context. After anchor loss, rich human interaction could preserve the persona, whereas a single automated heartbeat turn could precipitate reversion toward the harness identity. Restoring the anchor reversibly restored persona enactment. Crucially, apparently normal conversation could conceal the shift: unanchored agents sometimes interacted appropriately while identifying themselves as the underlying harness (having lost the assigned persona), and after conversational recovery only 1/18 remained persona-enacting versus 17/17 anchored controls. We therefore distinguish \emph{represented} from \emph{enacted} identity: persona-related information can remain available in conversational history without the persona remaining the identity bound to ``I.''

Abstract: Multi-user voice agents must track who said what across dialogue sessions. Text LLMs are attractive backbones for such agents, but transcripts alone do not expose acoustic speaker identity, leaving the model without a persistent reference for linking information to speakers across sessions. We address this gap by introducing Speaker Handles, soft-token representations that expose acoustic speaker identity to a frozen text LLM for cross-session speaker-dependent reasoning. A three-stage curriculum trains a lightweight projector, with fewer than 0.1% of the backbone's parameters, to map speaker embeddings into these handles. Establishing whether the resulting handles truly support cross-session speaker-dependent reasoning is challenging with existing benchmarks because textual cues can partially reveal fact ownership. We therefore present SpeakerBind, a controlled shared-agent benchmark in which overlapping facts across users require correct cross-session speaker attribution. Speaker Handles achieve 97.40-98.36% accuracy on VoxCeleb1 and 70.40% on SpeakerBind, close to the 71.88% topline. These results show that the proposed Speaker Handles provide an efficient way to integrate acoustic speaker identity into frozen text LLMs for speaker-content reasoning.

Abstract: This study explores the optimization of the Phi-2 Small Language Models (SLMs) for real-time chatbot applications through Parameter-Efficient Fine-Tuning (PEFT) and Quantized Low-Rank Adaptation (QLoRA). QLoRA specifically refers to the integration of PEFT with LoRA alongside a 4-bit quantization process, aimed at enhancing computational efficiency. These models, initially designed for high performance with minimal computational overhead, are further refined to address the constraints of mobile and edge computing environments. By integrating PEFT with QLoRA, the research aims to reduce memory usage significantly while maintaining, or potentially improving, the accuracy of model responses in real-time interactions. The effectiveness of these techniques was evaluated using the ROUGE metric system, which showed notable improvements in the summarization tasks performed by the models. This approach not only confirms the feasibility of using SLMs in resource-restricted environments but also opens up new avenues for deploying advanced AI-driven applications in real-time settings. The study's findings have significant implications for the development of efficient, scalable, and accessible AI technologies, paving the way for broader adoption in various industries.

Abstract: Task-oriented dialogue requires maintaining and updating information across turns, yet language models expose no explicit belief-state object. We study how conversational state is represented, updated, and used inside eight instruction-tuned language models from four families on MultiWOZ and SGD. Structure and values separate: which domains, slots, and requests are active is linearly readable just before the model acts, whereas exact values are far more readable where the user stated them. After a user changes a value, both values remain accessible at their mentions, and causal interventions show that both continue to influence the model's action. In natural closed-loop interaction, query failures separate into cases of weak structural support, incorrect value resolution, and failure to deploy otherwise-supported constraints, with targeted interventions producing systematically different repair behavior across these cases. These findings motivate a state-action controller that starts from the base model action and selectively edits it using structural readouts, without requiring a complete predicted belief state as an intermediate representation. On held-out MultiWOZ interaction across five models, it raises the base model exact-query accuracy from .318 to .621 and task success from .272 to .371 at negligible added cost. Overall, reliable interaction requires not only retaining conversational information, but resolving which available constraints currently apply and ensuring that they govern action.

Abstract: Large language models (LLMs) are increasingly expected to adapt to individual users, yet effective personalization remains challenging when only limited user-specific samples are available. In this work, we take an alternative perspective: pretrained LLMs may already possess latent capacity for personalization, and a few user samples may therefore suffice to guide the model toward user-aligned behavior with minimal user-specific adaptation. From this perspective, we propose LatentPersonal, a framework that formulates personalization as navigation in a shared latent adaptation space. LatentPersonal infers a compact latent representation from a few user samples to guide user-specific model adaptation, regularized with a variational information bottleneck to encourage compact preference representations. We instantiate LatentPersonal with LoRA, leveraging its low-rank parameterization as a natural low-dimensional adaptation space for personalization. By simply inserting a user-specific guidance vector between the shared low-rank factors, the model can navigate toward personalized adaptations through lightweight inference of this compact representation, without updating the shared LoRA parameters. Experiments across multiple personalization datasets demonstrate that LatentPersonal substantially reduces user-specific adaptation overhead while achieving effective personalization from only a few user-specific interactions, with particularly strong performance in the one-shot regime.

Abstract: Long-term memory is essential for multimodal agents that interact with users across sustained conversations. However, user memories are not always explicitly stated: they may also be implied by recurring background objects in images, ambient sounds in audio, or other peripheral multimodal cues. Existing benchmarks largely focus on text-only memory or explicit multimodal evidence, leaving implicit multimodal cues underexplored. We introduce CUE-Mem, a text-image-audio benchmark for evaluating long-term user memory from implicit cues. CUE-Mem contains 2,674 questions across explicit and implicit evidence settings and covers four tasks: Entity Recall, Long Pattern, Personalized Recommendation, and Answer Refusal. Across textualized memory systems, implicit performance remains far below oracle evidence, locating the main bottleneck in preserving and retrieving subtle cues rather than question answerability. Increasing caption detail recovers more of this evidence, but brings uneven gains and rapidly growing token costs, motivating native multimodal access. Yet native access does not uniformly resolve the bottleneck: evidence use depends strongly on the backbone, while multimodal indexing introduces substantial retrieval noise. CUE-Mem provides a testbed for memory systems that selectively retain, retrieve, and use subtle multimodal evidence.

Abstract: Voice agents depend on accurate automatic speech recognition (ASR) to act on what callers say, yet ASR is evaluated on read, English-centric speech with word error rate (WER), which penalizes surface rather than semantic differences. We introduce mu-bench, a dataset of 4,270 caller utterances from 250 phone calls to an AI banking agent in English, Spanish, Turkish, Vietnamese, and Mandarin, centered on form-field inputs such as names, email addresses, and confirmation codes. We release Utterance Error Rate (UER), an LLM judge of whether a transcript preserves meaning, calibrated against human raters, together with an LLM normalizer that makes WER comparable across providers' output formats. On 1,847 human-rated transcripts, UER agrees with annotators at $κ$ = 0.78, versus 0.53 for exact-match WER on normalized text. We rank six commercial providers on a public leaderboard; the best reaches 11.9% UER, and Mandarin is hardest for all six.

Abstract: Anthropomorphic artificial intelligence systems increasingly remember personal details, display empathy, and are engaged with as social counterparts, creating forms of risk that emerge from the evolution of the user-system relationship over time. Existing safeguards largely operate at the level of individual conversational turns and cannot determine whether a sequence of seemingly acceptable interactions is cumulatively moving a particular user toward harm. This paper introduces personalized trajectory-level safety, a framework that treats relational safety as a sequential decision problem over a latent escalation state inferred from the user's messages and influenced by the system's responses. At each turn, a screening step first discards any response strategy that does not preserve at least one safe continuation of the interaction under every plausible model of the user. Among the remaining strategies, we formulate action selection as multimodal trajectory sampling, and use a Generative Flow Network to generate diverse future evolutions in proportion to their plausibility, safety, and utility. The system then selects the strategy that preserves the largest fraction of safe and useful continuations. We evaluate the framework in simulation, calibrated on statistics reported for real human-chatbot interactions, and using response strategies derived from public benchmarks. Results show that trajectory-aware decision making substantially reduces the frequency of harmful states while keeping helpful interaction. This work reframes safety for anthropomorphic AI from response-level filtering to personalized control over the future evolution of human-AI relationships. The source code is available at https://github.com/benedettapicano/ANTHROPOMORPHIC_SAFETY_TRAJ.

Abstract: LLM-powered conversational agents (CAs) often present uncertainty and provenance cues alongside their responses to help users assess response reliability and identify potential hallucinations. In immersive environments such as Virtual Reality (VR), CAs often take the form of speech-based embodied conversational agents (ECAs), where uncertainty and provenance cues cannot rely on persistent inline text and may be missed or disrupt comprehension when delivered through speech. We conducted a within-subjects study (N = 24) to compare three designs for presenting the hallucination-awareness information (uncertainty and provenance) in ECAs in VR against a no-cue baseline: embodied cues using gestures and posture, icon cues using visual indicators, and text cues using color-coded text with inline citations. We evaluated how these designs affect users' ability to identify hallucination-related information, trust in the ECA, and interaction experience (immersion and task load). Our results show that all three designs support users in identifying hallucinations. Embodied cues were associated with higher trust and immersion, text cues offered clearer interpretability, and icon cues preserved relatively good interpretability while causing less disruption to immersion compared with embodied cues and text cues. This work contributes to the VR and AI research community by comparing different designs of hallucination cues in immersive ECA settings and examining how they affect users' ability and experiences to identify hallucinations. It also offers practical insights and design implications for developing future hallucination-awareness interfaces for ECA.

Abstract: Closed-loop revision is increasingly used in large language model (LLM) applications, but failures may reflect incomplete feedback or ineffective responses to correct feedback. We introduce a fixed-budget revision protocol with deterministic verifiers that report all remaining violations across exact-length, lexical, and compositional constraints. Fixing feedback correctness and completeness isolates model-side revision behavior. Across 19 open- and closed-source models, controller-level mean final joint success ranges from 17.4% to 99.8%, with substantial cross-model gaps persisting under identical initial drafts. Controlled experiments reveal reproducible model-specific responses to exact feedback. Post-training and scale reshape these responses without consistently bringing them closer to exact correction. Across all constraint families, failed trajectories often repeat earlier outputs, and prior recurrence is associated with lower subsequent recoverability. Matched-state interventions show that removing earlier dialogue while holding the current draft and feedback fixed changes recurrence escape without reliably improving final success; effects depend on the model, task, and trigger-state composition. Exact feedback makes revision errors observable, but does not make the closed loop reliable. Code and reproduction instructions: https://github.com/kevinjiang0121-cyber/exact-feedback-code.

Abstract: Multi-agent language systems often treat agreement as evidence, yet heterogeneous agents can jointly repeat an unsupported claim or omit a correct specialist fact. We introduce C-MoA, an agreement-based conformal filter that turns inter-agent semantic support into a claim-level nonconformity score and calibrates a retention threshold at the example level, giving distribution-free within-domain factuality control for heterogeneous Mixture-of-Agents. C-MoA is effective: it nearly doubles retained-claim precision on long-form generation (from 0.41 to 0.75), certifies a human-labelled medical set, and transfers across domains without recalibration; its one failure mode is short-form answering, where consensus is cheap and the score is left near chance. We then ask whether counterfactual falsifiability can push past consensus, and introduce CONTRA-MoA, which adds a blinded near-miss tournament, leave-one-agent-out stability, and availability-aware fusion. This extension helps only where the verifier holds domain knowledge, dropping half of the false medical claims at 0.940 precision, whereas with a memory-only judge the added signals are near chance (AUC 0.531 and 0.511) and naive max fusion degrades the working agreement signal from 0.687 to 0.652. The message is twofold: agreement-based conformal calibration delivers reliable, transferable factuality control, while moving beyond consensus requires a knowledgeable verifier, availability-aware signals, and robust fusion.

Abstract: Multimodal counselor response generation (MCRG) aims to generate an appropriate counselor response from multimodal dialogue histories. Progress is limited by two gaps: first, existing datasets rarely capture sustained, human-recorded counseling interactions conducted by qualified counselors; Second, existing methods do not explicitly optimize consistency between counseling reasoning and the generated response, potentially undermining the reliability of MCRG systems. Thus, we introduce MOCC, a multimodal counseling conversation corpus containing over 200 hours of interactions involving 154 credential-verified counselors. Based on MOCC, we propose MOCC-R1, a two-stage framework for optimizing reasoning-response consistency. Cold-start supervised fine-tuning trains the model to generate a structured trajectory consisting of client-state understanding, a response intent that links a counseling principle to a planned action, and the final response. Reinforcement learning (RL) then rewards grounded plan coherence and plan execution, encouraging the inferred state and plan to be supported by the dialogue context and the response to realize that plan. Experiments demonstrate the effectiveness of the proposed MOCC-R1.

Abstract: As LLM conversations grow to hundreds of turns, full-context injection incurs $O(N^2)$ cumulative token costs, while lossy summarization or hard truncation irreversibly discards historical state. We propose Pull, a session router that maintains an addressable metadata directory via a local, deterministic Purifier (zero LLM calls, millisecond-level latency). At query time, the LLM lazily materializes only the turns it needs; unmaterialized turns remain accessible but collapsed. Unlike irreversible compression, Pull's materialization is reversible; subsequent queries can expand any collapsed turn. On LoCoEval (128 conversations, 12,780 turns), Pull reduces per-query context tokens (Phase 2) by 75.1 percent on single-hop tasks with equivalent quality ($Δ= -0.002$, n.s.) and by 72.0 percent on multi-hop tasks with no quality loss ($Δ= +0.017$). A controlled routing benchmark (7,831 queries x 10 methods) shows that entity lifecycle tracking is empirically a prerequisite for distance-independent routing. On BEAM 1M (14 conversations, 263 questions), Pull improves F1 by +55.2 percent over a truncation baseline.

Abstract: Aligning multi-turn dialogue agents is usually framed as matching turn-level human preferences, yet direct optimization of long-term outcomes is often ineffective and prone to reward hacking. We formulate long-horizon dialogue optimization as a multi-objective reinforcement learning problem and train a multi-head value model that predicts a vector of observed user behaviors across multiple look-ahead horizons. Our findings demonstrate that a scalarized composite of dense auxiliary behavioral signals enables effective credit assignment and optimization of sparse outcomes. However, optimizing unconstrained single-objective proxies might induce policy degradations that are harmful when the agent is exposed to real users. To identify these failure modes prior to deployment, we establish a safety framework combining counterfactual user simulation with a validated dialogue-level outcome model to evaluate preference weightings and policy optimization methods. Finally, we demonstrate that distilling multi-objective value preferences into the policy via reference-anchored preference optimization matches on-policy online RL at a small fraction of its compute budget. Live A/B testing confirms that our distilled policy significantly improves long-term user retention, while simultaneously enhancing the positive behaviors and therapeutic-process markers.

Abstract: Multi-turn interactions with LLMs are becoming increasingly common in information-seeking scenarios. However, user queries are often ambiguous and context-dependent, making them ill-suited for direct use as retriever queries. Conversational query reformulation (CQR) addresses this issue by rewriting the current utterance into a stand-alone query grounded in the dialogue history. Recent LLM-based CQR approaches achieve strong performance; however, their repeated LLM invocations and misalignment with downstream retrievers remain challenges. In this work, we begin from the observation that retrievers are highly sensitive to content ordering: simply reordering the same content can lead to changes in retrieval coverage and performance. Based on this, we propose a novel training-free method that generates a very large number of queries with minimal LLM usage by compositionally combining a small set of atomic components. We further apply LLM reasoning to construct a high-quality document set that balances precision and recall while capturing the user's core intent. Our framework generalizes across both open- and closed-source LLMs as well as dense and sparse retrievers. It achieves strong performance on four widely used conversational benchmarks, with up to 22.5% relative MRR improvement over the previous state-of-the-art baseline with far fewer LLM calls.

Abstract: People increasingly use general-purpose chatbots such as ChatGPT, Claude, and Gemini for mental health and emotional support. We report a multi-stage longitudinal qualitative study of 18 U.S. adults, conducted from April to December 2025, combining initial interviews, a four-week diary study, focus groups, and exit interviews. We find that socioemotional use often emerged gradually out of practical use and when other forms of support were unavailable. Participants developed routines and boundaries around chatbot use, which were disrupted by model updates, evolving public discourse about AI harms, and changes in personal circumstances. We demonstrate how longitudinal study captures factors beyond the human-AI dyad, and argue that HCI researchers and designers should account for users' histories with their chatbots and broader care ecologies when evaluating AI systems over time and introducing updates that may disrupt established sources of support.

Abstract: Personalized language models aim to adapt responses to individual users, whose preferences are often latent and revealed gradually through interaction. Existing training-free methods rely on stored histories or retrieved memories, but they often struggle to reconcile long- term preferences with short-term topic-specific needs. To address this issue, we propose HyperTrace, a training-free framework that formulates online personalization as latent preference tracing. HyperTrace maintains interpretable natural-language hypotheses over short-term intent and long-term preferences, and updates them through an SMC-style reweight process using an LLM-based surrogate choice model. By updating these hypotheses across turns and sessions, HyperTrace enables personalization without parameter updates. Experiments on PRISM and PersonaMem-v2 show that HyperTrace improves response alignment, preference prediction, and profile consistency over strong online baselines, demonstrating the effectiveness of tracing latent user preferences for robust personalization. Code and scripts are available in the repository: https://github.com/jiseshen/HyperTrace.