Spoken dialogue state tracking recovers slot-value pairs from speech, where ASR errors concentrate in entity values and persist across turns, making it both a generation and an editing problem. A strong per-turn text editor corrects much of this but, operating on the transcript alone, leaves three recoverable errors: a value predicted inconsistently across turns, an omitted slot, and a value the audio does not support. We present AVERT, which scores each candidate value by combining cross-turn agreement with a trained audio-conditioned verifier and resolves the three error types with three operators, vote, add, and swap, each restricted to the slots where its error is common. On SpokenWOZ, a base speech-LLM reaches 33.04 JGA, a text editor 38.34, and AVERT 40.13, without retraining either. This is in the range of a 1B end-to-end system that consumes the full spoken history (39.32), though AVERT uses two 1B decoders rather than one. The audio verifier contributes a statistically significant gain, and restricting each operator to a selected slot subset matters: removing it lets unrestricted voting overwrite correct categorical values and fall below the editor.
Evaluating task-oriented dialogue agents requires judging not merely whether a reply reads well but whether each turn advances the underlying workflow state correctly--a distinction conventional holistic LLM judges can miss because they evaluate the available context as a single unit and require one or more full-model calls per turn. We propose SAGE (State-Grounded Abstention-Aware Evaluation), which compiles a workflow specification and per-turn state diff into atomic, schema-grounded criteria and routes each through a cascade of symbolic and encoder/NLI verifiers that abstain rather than guess, aggregating criterion verdicts into a turn-level decision with an evidence trace. Its recommended operating point, SAGE-Core, decides 81--91% of criteria with only the compiler, symbolic rules, and on-device encoders--at zero paid LLM cost--while SAGE-LLM adds an optional focused-LLM fallback for open-class criteria. Across four slices spanning MultiWOZ, Schema-Guided Dialogue, and ABCD, no evaluated LLM-as-a-judge baseline--including a state-aware GPT-4.1 judge and cheaper GPT-4.1-mini variants--significantly exceeds SAGE-Core on any slice, even though the GPT-4.1 G-Eval judge costs $4.7--8.0 per 1,000 turns to SAGE-Core's $0. A two-annotator human audit (n=200, $κ$=0.94) confirms strong label fidelity on the transcript-visible failure classes--where, excluding the weak-salience IUV class, SAGE-Core is statistically tied with the strongest LLM judge--and honestly scopes ignored-user-value as a state-consistency signal with weak broad-human salience. We analyze construct-validity limits from injected failures and partial symbolic circularity.
Video world models are increasingly used as simulators, yet visual fidelity alone does not show that a model maintains the hidden state of the world. We examine this gap with an action-conditioned video Shell Game, a visual analog of $S_5$ state tracking that decouples visual rendering from compositing the hidden state underneath. Bidirectional and autoregressive Transformers, Mamba, and linear attention restricted to nonnegative transition eigenvalues all fit the training horizon of 5 swaps and then fall toward chance on longer swap chains (extrapolation) while still rendering plausible video with additional denoising steps providing no benefit. The pixel-based diffusion target never supervises the unseen hidden state, so the generated frames cannot carry it and the state has to live inside the architecture rather than in the tokens. For a Transformer, that architectural state is only an append-only KV cache, so the model has to re-derive the hidden arrangement from the whole history at every chunk. We find two mechanisms that do extrapolate, and both carry a state across chunks and revise it in place. Linear attention succeeds once its transition eigenvalues may be negative, and TTT with a nonlinear fast weight succeeds by updating the feature map through which it reads its own state. We further examine harder cases in dynamic world exploration tasks, and discuss the broader implications for building stateful video world models.
In this paper, we propose \textbf{Mahalanobis-Based Multi-Head Attention} (MHA-CSP), a novel attention mechanism that replaces the standard dot-product with a \textbf{Mahalanobis distance-based RBF kernel}, which effectively computes attention in an infinite-dimensional feature space without increasing the parameter count. Crucially, the positive definiteness of the Mahalanobis distance enables a \textbf{direct construction of Tree Attention}: attention scores are built directly from accumulated distances, with a LogSumExp correction that rectifies the raw distance by subtracting the log-sum of edge exponentials. Moreover, the multi-head Mahalanobis distance matrices are themselves repurposed to construct an \textbf{attention meshing mechanism}, enabling cross-head kernel collaboration that simultaneously boosts accuracy and training efficiency. Extensive experiments demonstrate that MHA-CSP, with only 119K parameters and \textbf{teacher forcing applied exclusively at the final hidden state}, consistently outperforms Transformer and GCN baselines trained from scratch under identical conditions on long-sequence state tracking tasks. While these baselines rely on dense attention or graph propagation, MHA-CSP achieves robust structured reasoning via synthetic distance rectification---powered by Mahalanobis-based attention---and efficient information bypass inherited from the CSP backbone. This result highlights the effectiveness of complex-valued state propagation with collaborative multi-head rectification in capturing symbolic structures, establishing a new efficiency-performance trade-off for structured reasoning.
Reliable multi-turn tool use requires an agent to preserve an evolving task state and ensure that each action remains consistent with it. However, direct function-calling and ReAct-style policies learn state tracking and action generation within the same autoregressive trajectory. This coupling creates state-action competition: the pressure to produce the next call can overwrite or ignore information accumulated earlier in the interaction. Inspired by Boyd's Observe-Orient-Decide-Act cycle, we introduce OODA-Tool, a typed closed-loop policy designed to mitigate this competition by separating state preservation from action realization. Rather than generating an action directly from the interaction history, OODA-Tool routes each decision through controller-checked intermediate states, ensuring that the final output remains grounded in the current task state. Specifically, Observe reconstructs the task state, Orient determines whether execution is warranted, Decide forms an admissible action structure, and Act realizes the external output. We evaluate OODA-Tool against direct function-calling and ReAct policies using Qwen3 models ranging from 0.6B to 14B across multi-turn, multi-tool, and incomplete-information settings. OODA-Tool consistently improves task success across model sizes, with larger gains on smaller models and on tasks whose actions depend strongly on information accumulated across turns and prior tool results. Controlled variants, stage-level ablations, and transfer evaluations further demonstrate the robustness of these improvements.
As LLM-based agents are deployed for longer and higher-stakes tasks, their memory systems continue to have crucial gaps. While existing memory benchmarks focus largely on recall-shaped tasks, we argue an effective memory system must track the evolving state of the world; as facts, constraints, and decisions are revised over a long interaction, answers must reflect the current state and not a superseded one. We define this capability as state tracking and instantiate it in StateMemBench, a benchmark of 234 multi-session scenarios spanning two conversation-length regimes. Its closed-pool grading scores whether an answer reflects the current state, the superseded state, or fails otherwise, separating state-tracking failures from other errors by construction. Our analysis shows that this task is challenging for existing memory systems, retrieval-augmented baselines, and long-context baselines. We then present StateMem, a state-first memory method that explicitly tracks supersession and relational dependencies, and show it improves current-state accuracy over the strongest same-backbone baseline by 1.8x (0.205 -> 0.363) on DeepSeek-V4-Flash and over the strongest memory system by 1.6x (0.149 -> 0.233) on Qwen-3.5-9B, while remaining competitive with the long-context baselines. Finally, we show the same state approach can be applied as a lightweight single-call wrapper over existing memory systems, lifting current-state accuracy by +32 to +67 points on StateMemBench across six memory and retrieval backends. A length- and cost-matched control attributes +15 to +32 of those points to state structure rather than added context.
User-centric multi-turn agents must act on an evolving task situation shaped by changing user intents, accumulated tool-grounded facts, missing information, and execution constraints. Existing context-management methods improve the use of past interaction history, but rarely maintain an explicit situation state that separates grounded facts from task-state judgments. As a result, agents often need to infer fine-grained attributes, task dependencies, and constraint satisfaction implicitly from dialogue traces. We propose Intent-Driven Situation States (IDSS), a training-free framework that maintains an explicit situation state alongside the dialogue. IDSS parses tool returns into provenance-aware entities and attributes, tracks user intents, required variables, constraints, and execution status, and propagates new facts to task constraints to update action executability. This allows agents to avoid infeasible actions, advance dependent goals, and reuse relevant information without repeatedly searching raw history. Experiments on three interactive benchmarks across eight LLMs show that IDSS improves task completion, preference elicitation, and interaction efficiency, with clear gains on tasks involving multi-entity coordination, evolving user constraints, and constraint-aware replanning. Ablations and error analyses show that these improvements come from the interaction between fact persistence, intent-centered state tracking, and constraint modeling. These results suggest that explicit situation tracking offers an effective alternative to history-centric context management for reliable user-centric multi-turn agents.
Transformer-based architectures have dominated sequence modeling, largely due to the expressive power of attention mechanisms. However, for a class of deterministic state tracking tasks---such as parity checking, modular counting, and parenthesis matching---attention may be overkill. In this paper, we show that \textbf{state propagation alone is sufficient}. We propose the \textbf{Complex State Propagator (CSP)}, a minimalistic recurrent architecture that \textbf{only propagates hidden states} across layers without output projections at intermediate steps. The state is represented as a complex-valued vector, updated via input-dependent rotations in the complex domain. To enable deep propagation without gradient vanishing or degradation, we introduce a \textbf{block-level skip connection} alongside element-wise complex normalization and SiLU activation at sequence boundaries. Applied with Focal Loss, CSP achieves \textbf{100\% accuracy} with perfect F1 scores across canonical tasks.
Long-horizon tasks remain uncommon in large language model (LLM) evaluation, and for a reason: when each step depends on the last, per-step accuracy that looks excellent in isolation decays catastrophically, as errors cascade and the end-to-end failure probability grows sharply with length. Existing agentic benchmarks report end-to-end success but confound this state-tracking difficulty with instruction interpretation, give no control group that isolates it, and are vulnerable to shortcuts such as a hallucinated final answer, so they cannot say why a long run fails. Whether an LLM can carry exact intermediate state across many tool calls at all is itself not well established. We test this cleanly by having the model compute a cryptographic hash, MD5, step by step: a sequence of $196$ dependent tool calls over $64$ rounds while it carries four $32$-bit words $(a,b,c,d)$ in its own context from one call to the next. Interpretation is trivial and, because we implement MD5 from scratch (RFC~1321), we align every call to the ground-truth trace and check the digest to the bit, so any failure is pure bookkeeping. gpt-oss-120b, a mixture-of-experts model with only $\sim$5.5B active parameters per token, at temperature $0$ with a short fixed prompt, carries the full state across all $196$ calls and returns the correct digest on a majority of completed runs. In the strongest setting we replace every primitive tool with a second LLM, so a driver and a worker compute the whole hash from scratch with no exact-arithmetic oracle in the loop. Two ingredients decide success and neither changes the weights: keeping the model's own reasoning in its context each turn, and voting over a thinking-enabled worker to remove its modular-arithmetic slips. We localize the residual failures by origin, separating state-carrying from arithmetic and from serving.
Giuseppe Soriano, Nicola Tonellotto, Alberto Gottacs.AI
Forecasting under real-world conditions is inherently non-stationary, as the conditional distribution of future observations evolves over time. Recent test-time adaptive sequence models address this challenge by updating internal states during inference, but tie adaptation to instantaneous prediction errors or surprise. This coupling can conflate persistent distribution shift with stochastic innovations, leading to unnecessary updates and inefficient adaptation. We introduce Black-Mamba, a test-time adaptive forecasting architecture that formulates online adaptation as evidence-gated state tracking under distribution drift. The model augments a base predictor with a dynamic memory updated when temporally accumulated surprisal provides sufficient evidence of a regime change. This turns adaptation into a selective, event-driven process rather than a continuous one. Across multiple forecasting benchmarks with non-stationary dynamics, Black-Mamba achieves competitive or improved predictive performance compared to existing test-time adaptation methods while significantly reducing the number of memory updates during inference. Together with mathematical analysis and biological evidence, these results suggest that accumulated surprisal provides a principled signal for distinguishing persistent drift from transient noise, yielding more efficient and robust adaptation.
Wenyi Wu, Sibo Zhu, Kun Zhou +3cs.AI cs.CL cs.LG cs.MA
Recent advances in large language models (LLMs) and vision-language models (VLMs) have enabled increasingly capable digital agents for computer use. However, real-world tasks are often long-horizon and involve evolving contexts containing accumulated observations, intermediate edits, failed attempts, and partially completed executions. Existing agents typically operate over raw interaction history, making task progress difficult to interpret, verify, and recover, which ultimately limits reliable long-horizon execution. In this paper, we argue that addressing this challenge requires explicitly structuring both the agent's state and workflow around a unified causal representation of task progress. We present \textbf{StructAgent}, a state-centered framework that introduces a unified state for maintaining compact, verifiable task progress and a structured workflow that regulates progress through verifier-backed state transitions. Building on this design, StructAgent further enables explicit progress checkpointing, evidence-driven task completion, targeted failure recovery, and tool-supported execution, while ensuring that all progress updates remain grounded in verification. Extensive experiments demonstrate that StructAgent consistently improves a wide range of LLM and VLM backbones on long-horizon computer-use tasks. On OSWorld-Verified, it improves Qwen3.5-9B from 27.0\% to 46.9\% success rate and Qwen3.5-27B from 31.6\% to 62.2\%, while achieving a new open-source state of the art of 78.9\% with MiniMax-M3. Moreover, the same framework generalizes beyond desktop environments to Minecraft, demonstrating the generality of our design.
A striking feature of the human visual system is that it ingests visual information through a series of local foveated glimpses, rather than a single global computation. This makes human vision distinctly different from most popular computer vision models in use today, which input images globally and in a single shot. A natural question therefore is whether local, sequential vision models may provide any fundamental computational benefits in addition to being biologically more plausible than global models. In this work, we investigate this question from the perspective of visual state tracking and length generalization. Inspired by recent studies of length generalization in language models, we study the behavior of vision models trained on simple vision tasks that require the aggregation of local information across an image. Our experiments reveal that, similar to language models, vision models can learn to exploit global shortcuts and thereby fail to generalize over task length or complexity. We also show that recurrent vision policies based on strictly local perception can mitigate these failures, thereby allowing models to generalize on these tasks. Our results show that local attention may be an essential overlooked requirement for robust compositional generalization.
Zitong Shi, Yixuan Tang, Anthony Kum Hoe Tungcs.AI
Long term memory lets LLM agents act as persistent assistants, but user facts change. A useful memory system must know what is true now, what used to be true, and what changed. We study \emph{ghost memory}, a state coordination failure in which old, current, and transition facts coexist in the memory bank, remain mixed during retrieval, and mislead the answer model. We argue that memory systems should be understood and optimized from three levels: bank maintenance, retrieval, and answer time resolution. We propose ATMA, a state aware overlay for existing memory systems. ATMA keeps superseded and transition records in the bank, builds evidence packets for the query's requested state view, and exposes current, historical, and transition labels to QA. We further call for decoupled evaluation of bank, retrieval, and answer level failures, since final QA accuracy can hide where ghost memory occurs. To make this failure measurable, we build LTP (LoCoMo Temporal Plus), a conflict heavy benchmark for ghost memory, and evaluate on LoCoMo for long conversation generalization. On LTP, Graphiti+ATMA improves conflict accuracy by 0.240 absolute over Graphiti. On LoCoMo, Graphiti+ATMA raises temporal F1 from 0.0295 to 0.1705. The gains are host dependent, but they indicate that explicit state roles can reduce memory failures hidden by final QA accuracy.
Benjamin Shih, John Winnicki, Eric Darvecs.LG cs.CL
A central hope behind process supervision is that models can expose intermediate variables that matter for their later behavior. For this to help with alignment, a scratchpad must be tied to the computation: when the model writes a state, later steps should compute from that state. To test this requirement, we use a controlled state-tracking task with a known update rule, comparing models trained to report only the final state with models trained to write intermediate states before giving the final answer. At evaluation, we edit the internal representation of one written state while leaving the visible scratchpad text fixed. Because the transition rule is known, the edit has a single correct downstream consequence. In Qwen2.5-Coder-7B, the state-writing model predicts the next phase bit implied by the edited state on 80% and 91% of held-out examples across the two task variants, while pretrained and final-answer-only controls remain near baseline. Additional controls rule out generic next-token steering and copying another continuation: the prediction depends on both the edited state and the current move. The same causal-use pattern replicates across model families. Together, these results suggest a sharper goal for scratchpad oversight: not just to make intermediate reasoning legible, but to train written states that the model uses as part of its computation.
Hybrid language models that mix attention and recurrent layers have shown promise: theoretically, recurrent layers ameliorate the limitations of pure transformers on state tracking, and empirically, hybrids can outperform pure transformers in loss and downstream evaluations \citep{waleffe2024empirical,merrill2026olmohybrid}. Yet it remains unclear which data or capabilities drive these gains, and to what degree they reflect the theoretical advantages motivating hybrid models. We address this question using the open weights from Olmo 3 \citep{olmo2025olmo3} and Olmo Hybrid \citep{merrill2026olmohybrid}: we compare the loss of a matched transformer and hybrid at the same target tokens under the same prefixes, stratifying the results by natural token tags, copy features, delimiter structure, and controlled synthetic probes. The hybrid has lower loss on most tag families, but the gains are not uniform: they are largest for open-class content words and smaller for many closed-class function words. Across prose, code, and markup, the hybrid's loss advantage is larger on opening delimiters than on the corresponding closing delimiters, and nearly vanishes on repeated $n$-grams. Synthetic probes show the same split: the hybrid is favored on pronoun-memory and entity-tracking tasks, whereas the transformer is favored on bracket-matching tasks that require choosing closing delimiters. These patterns suggest that the recurrent layers in hybrids improve predictions that leverage the semantic state of a document, whereas attention helps on tokens predictable by $n$-gram copying or syntactic bracket matching. We conclude with proof-of-concept filtered evaluations showing how token-level decompositions can sharpen pretraining diagnostics for hybrid architectures.
Linear attention replaces softmax attention's growing KV cache with a fixed recurrent state, but this compression limits exact state tracking and long-context memory. We introduce \emph{Semidirect Fourier Delta Attention} (SFDA), a phase-controlled generalization of Kimi Delta Attention that replaces real diagonal decay with block-rotational Fourier control: \[ S_t=(I-β_t k_tk_t^*)Λ_tS_{t-1}+β_tk_tv_t^*, \qquad Λ_t=\diag(α_t\odot e^{iθ_t}). \] Our main result is a constructive chunk-WY factorization for products \(A_t=Λ_t-u_tr_t^*\), giving \[ A_t\cdots A_1=Γ_t-Y_tM_tW_t^* \] with rank growth bounded inside fixed chunks. This yields an exact affine chunk transfer, formal stability and complexity bounds, and a compact characterization of phase-plus-low-rank memory. We verify the algebra numerically and show in toy state-tracking experiments that SFDA learns cyclic memory where the phase-disabled KDA baseline remains near chance. Fused kernels and large-scale language-model comparisons are left to future work.
Anamaria-Roberta Hartl, Levente Zólyomi, David Stap +6cs.LG
Transformers dominate modern sequence modeling, but their quadratic attention incurs substantial computational cost. Subquadratic architectures offer a scalable alternative. However, it remains unclear which designs yield the most effective sequence models. We compare three leading approaches: xLSTM, Mamba-2, and Gated DeltaNet. We evaluate these models on tasks with complex dependencies: (1) code-model pre-training, (2) distillation of code models from large language models, and (3) pre-training of time-series foundation models. Across these settings, xLSTM delivers the strongest overall performance. To explain xLSTM's advantage, we present a unified formulation and analyze the underlying architectural mechanisms, focusing on state tracking and memory dynamics. Our results show that xLSTM enables more flexible and stable memory correction via its gating scheme. We corroborate these findings on controlled synthetic length-generalization tasks. Overall, our findings indicate that xLSTM's gains on complex tasks stem from robust state tracking and accumulation.
To interpret context correctly and retrieve relevant information, large language models must bind entities to their attributes and update these bindings as state changes. We analyze how LLMs implement this binding process in a dynamic state tracking. Using causal interventions, we identify a retrieval conditioned rebinding mechanism, a compact attention head circuit that encodes swap relevant binding information and reinstates it at readout. Across Gemma and Llama models, this circuit supports rebinding behavior, but the representational signature of the mechanism differs across model families. In Gemma models, the binding signature is clearly expressed in the query/key subspaces of the relevant attention heads, whereas in Llama models, the binding information is carried primarily in key vectors. Overall, our results reveal an interpretable mechanism for context dependent state tracking in LLMs.
State tracking exposes a sharp limitation of sequence models: the relevant signal is often not a summary of observed tokens, but an ordered latent state that evolves through non-commutative transformations. We introduce a held-out transition-pair falsifier for finite non-Abelian group tracking. The protocol forbids selected ordered generator pairs during training and requires the same local patterns during evaluation, blocking one direct local-transition memorization pathway. In a controlled $S_3 \times S_3$ benchmark, a projected recurrent state model trained only on length-8 sequences produces error-free final-state predictions (perfect 250/250 per horizon) through evaluation horizons up to 1,048,576 tokens across five seeds. Matched native-readout baselines, including bag, GRU, and a single-configuration structured state-space model, remain near floor under the same protocol. Projection-matched GRU, structured SSM, and bag baselines equipped with analogous finite-group prototype readouts also remain near chance under the same split. Mechanism diagnostics show that hard projection coincides with low homomorphism error, low state-consistency drift, and non-trivial commutator separation, while softened projection collapses final-state accuracy. Clean-split audits verify zero verbatim reduced-word overlap and zero structural-template overlap between training and evaluation partitions. The evidence is scoped to this controlled finite-group falsifier rather than to a general architecture ranking. Within that regime, explicit projected non-commutative state composition acts as a useful inductive bias for long-horizon hidden-state tracking.
Understanding a video requires more than recognizing isolated moments, as humans continuously track entities, states, and events over time. This capacity for visual state tracking is fundamental to video understanding, yet remains underexplored in current evaluations of Multimodal Large Language Models (MLLMs). We introduce Visual STAte Tracking benchmark (VSTAT), a video-based benchmark designed to diagnose visual state tracking in MLLMs. VSTAT consists of 834 clips drawn from both synthetic and real-world videos, paired with 1,500 questions that cannot be answered from any single frame or short segment, requiring continuous perception and integration of events across the entire video stream. Despite their strong performance on existing video benchmarks, we find that state-of-the-art MLLMs perform far below humans and only modestly above answer-prior baselines. To analyze this gap, we compare MLLMs' thinking traces with the underlying video stream to understand why and when MLLMs fail on VSTAT. We find that MLLMs reason and track correctly in text, but fail at visually perceiving the events they need to track. Finally, our preliminary evaluation suggests that recent agentic approaches, including MLLM-based video agents and coding agents, do not readily resolve these failures, still falling short on VSTAT.