In open-ended generation, LLMs frequently fall into the "likelihood trap", marked by repetitive degeneration and vocabulary dullness, creating a discrepancy between machine-generated and human-written text. While post-hoc tail truncation (e.g., Top-$p$, Min-$p$) avoids sampling from the unreliable tail, it can over-sample from the uncalibrated head and misalign generation with human lexical preferences; fixed scalar repetition penalties likewise ignore variation in logit scale across inference steps, potentially disrupting semantic coherence. To address both limitations, we propose Variance-Calibrated Modulation (VCM), a training-free pre-decoding intervention that reshapes the probability distribution before truncation through two dynamic mechanisms: (1) Contextual Searchlight via PMI, which suppresses global stopwords while elevating context-evoked tokens, and (2) Adaptive Self-Debiasing, which uses real-time logit standard deviation for scale-invariant penalization. Across open-ended generation, factual QA, and mathematical reasoning, VCM consistently mitigates the likelihood trap. With negligible computational overhead, VCM integrates with existing decoding strategies, improving diversity, coherence, and, particularly at higher decoding temperatures, reasoning accuracy.

Estimating causal effects in industrial time series requires handling temporal dynamics, time-varying treatments, and unobserved confounders. Existing causal foundation models (CausalPFN, CausalFM) operate only on static cross-sectional data; neural temporal methods (CRN, G-Net) require per-dataset training; and concurrent temporal-PFN proposals have not been demonstrated at industrial scale. None output explicit per-pair reliability signals alongside their CATE estimates. We introduce Temporal Causal Prior-Data Fitted Networks (TCPFN), a foundation model for zero-shot temporal causal discovery with learned reliability signals. TCPFN makes four contributions: (1) a Causal Judgment Head that jointly predicts null-effect probability, confounding strength, identifiability, mediation fraction, and causal regime; (2) a mixed training prior covering six causal regimes (independent, direct, confounded, mediated, time-varying confounded, feedback) plus CausalFM-style front-door and instrumental-variable priors; (3) a discrete-token panel-data architecture with cross-attention masking that prevents inter-horizon leakage; (4) zero-shot inference at industrial scale via FAISS-based context selection and one-step posterior correction. On 19 benchmark datasets across five domains, TCPFN achieves competitive zero-shot causal discovery: AUROC 0.96 on Tennessee Eastman, 0.93 on SWaT, 0.98 on Causal Rivers, 0.97 on CAUSRCA. The null detector reaches NullF1 0.94, AUROC 0.99. TCPFN scales to V=1,275 on a proprietary Kraft pulp-and-paper dataset in 6 hours on a single GPU; PCMCI, a CPU-only library, on a V=666 sub-panel of the same data took 81.5 hours, extrapolating by O(V^2) to ~12.5 days at V=1,275. TCPFN's top edges identify cross-subsystem causal relationships while PCMCI's surface within-instrument controller-measurement coupling -- a scalability case study.

AdamW is the de facto optimizer for training large language models (LLMs), yet the theory behind it still lives mostly in finite-variance regimes. This is increasingly unsatisfying, as empirical evidence indicates that stochastic gradient noise in LLM pretraining is typically heavy-tailed. Recent work shows that sign-based optimizers such as Lion and Muon achieve sharp heavy-tailed rates, and that AdaGrad can also converge under heavy-tailed noise. However, no rigorous convergence theory for AdamW has yet been established in this regime. Can AdamW converge under the same heavy-tailed assumptions, or does its second-moment accumulator create a genuine obstruction? We formulate this as an open problem, prove a positive weighted-metric benchmark, and give a corridor lower-bound mechanism showing how denominator memory can hide large gradients.

Agentic LLMs must continuously decide whether newly extracted facts should be added, merged with existing memories, or ignored, yet prior work has focused more on retrieval and storage than on principled write-side control. We frame memory evolution as a novelty-detection problem and propose SAGE, a Spherical Adaptive Gate for memory Evolution that scores candidate facts with a von Mises-Fisher-based density estimator over memory embeddings and routes them with an adaptive threshold that tracks memory-store geometry. SAGE resolves clearly novel facts as ADD, clearly redundant facts as NOOP, and sends only uncertain cases to an LLM merge step, reducing expensive write-time reasoning. On LoCoMo, SAGE achieves the best average token-F1 against Mem0 on all seven open-weight backbone comparisons, while on GPT-4o-mini it reduces add-phase API cost by 3.4$\times$ and add-phase latency by 2.5$\times$ with only a small average judge-score gap. As a drop-in binary gate for A-Mem, SAGE skips roughly 16-18% of LLM calls across five models with minimal quality change on open-weight backbones. These results suggest that novelty-aware write control is a practical lever for improving both memory quality and system efficiency in long-term agentic memory. The source code for our approach is accessible at https://github.com/swang1024/SAGE.

Large language models (LLMs) trained via reinforcement learning with verifiable reward (RLVR) have achieved breakthroughs on tasks with explicit, automatable verification, such as software programming and mathematical problems. Extending RLVR to electronic design automation (EDA), especially automatically generating hardware description languages (HDLs) like Verilog from natural-language (NL) specifications, however, poses three key challenges: the lack of automated and accurate verification environments, the scarcity of high-quality NL-code pairs, and the prohibitive computation cost of RLVR. To this end, we introduce CodeV-R1, an RLVR framework for training Verilog generation LLMs. First, we develop a rule-based testbench generator that performs robust equivalence checking against golden references. Second, we propose a round-trip data synthesis method that pairs open-source Verilog snippets with LLM-generated NL descriptions, verifies code-NL-code consistency via the generated testbench, and filters out inequivalent examples to yield a high-quality dataset. Third, we employ a two-stage "distillthen-RL" training pipeline: distillation for the cold start of reasoning abilities, followed by adaptive DAPO, our novel RLVR algorithm that can reduce training cost by adaptively adjusting sampling rate. The resulting model, CodeV-R17B, achieves 68.6 % and 72.9 % pass@1 on VerilogEval v2 and RTLLM v1.1, respectively, surpassing prior state-of-the-art by 12 20 %, while even exceeding the performance of 671BDeepSeek-R1 on RTLLM. We have released our model, training code, and dataset to facilitate research in EDA and LLM communities. 1

Pretrained vision-language models (VLMs), such as CLIP, have shown promising zero-shot out-of-distribution (OOD) detection capabilities by leveraging semantic similarities between input images and textual labels. However, most existing approaches focus solely on expanding the label space in the text domain, ignoring complementary visual cues that can further enhance discriminative power. In this paper, we introduce DualCnst, a novel framework that integrates text-image dual consistency for improved zero-shot OOD detection. Specifically, we generate synthetic images from both ID and mined OOD textual labels using a text-to-image generative model, and jointly evaluate each test image based on (i) its semantic similarity to class labels and (ii) its visual similarity to the synthesized images. The resulting unified score function effectively combines multimodal information without requiring access to in-distribution images or additional training. We further provide theoretical analysis showing that incorporating multimodal negative labels reduces score variance and improves OOD separability. Extensive experiments across diverse OOD benchmarks demonstrate that DualCnst achieves state-of-theart performance while remaining scalable, data-agnostic, and fully compatible with prior text-only VLM-based methods. The code is publicly available at: https: //github.com/TMLSIAT/DualCnst.

The emergence of agentic recommender systems powered by Large Language Models (LLMs) represents a paradigm shift in personalized recommendations, leveraging LLMs' advanced reasoning and role-playing capabilities to enable autonomous, adaptive decision-making. Unlike traditional recommendation approaches, agentic recommender systems can dynamically gather and interpret user-item interactions from complex environments, generating robust recommendation strategies that generalize across diverse scenarios. However, the field currently lacks standardized evaluation protocols to systematically assess these methods. To address this critical gap, we propose: (1) an interactive textual recommendation simulator incorporating rich user and item metadata and three typical evaluation scenarios (classic, evolvinginterest, and cold-start recommendation tasks); (2) a unified modular framework for developing agentic recommender systems; and (3) the first comprehensive benchmark comparing over 10 classical and agentic recommendation methods. Our findings demonstrate the superiority of agentic systems and establish actionable design guidelines for their core components.

Consider this prompt "Draw a unicorn with two horns". Should large language models (LLMs) recognize that a unicorn has only one horn by definition and ask users for clarifications, or proceed to generate something anyway? We introduce concept incongruence to capture such phenomena where concept boundaries clash with each other, either in user prompts or in model representations, often leading to under-specified or mis-specified behaviors. In this work, we take the first step towards defining and analyzing model behavior under concept incongruence. Focusing on temporal boundaries in the ROLE-PLAY setting, we propose three behavioral metrics--abstention rate, conditional accuracy, and answer rate--to quantify model behavior under incongruence due to the role's death. We show that models fail to abstain after death and suffer from an accuracy drop compared to the NON-ROLE-PLAY setting. Through probing experiments, we identify two main causes: (i) unreliable encoding of the "death" state across different years, leading to unsatisfactory abstention behavior, and (ii) role playing causes shifts in the model's temporal representations, resulting in accuracy drops. We leverage these insights to improve consistency in the model's abstention and answer behaviors. Our findings suggest that concept incongruence leads to unexpected model behaviors and point to future directions on improving model behavior under concept incongruence.1

In studies of transferable learning, scaling laws are obtained for various important foundation models to predict their properties and performance at larger scales. Taking language-vision learning as example, we show here how scaling law derivation can also be used for model and dataset comparison, allowing to decide which procedure is to be preferred for pre-training. Full scaling laws based on dense measurements across a wide span of model and samples seen scales are derived for two important language-vision learning procedures, CLIP and MaMMUT, that use either contrastive only or contrastive and captioning text generative loss. For the first time, we use derived scaling laws to compare both models and three open datasets, DataComp-1.4B,

ARelated Work1 Data Attribution Evaluation: Given recent developments in data attribution methods for LLMs,2 past works in evaluating these methods fall two major categories: leave-out-out and task-based3 evaluation. Leave-one-out evaluation measures the correlation between the data attribution method4 scores and model-retraining, which can also be approximated using linear datamodeling score [26].5 In task-based evaluation, the data attribution method is evaluated based on its application towards6 downstream task, such as noisy label detection, counterfactual evaluation [3, 13].7 Training Data Selection: Selecting high-quality training data selection is important for efficient8 learning in LLMs. Common approaches to data selection relies on heuristic filtering, such as de-9 duplication and lexicon-filtering, [34], or semantic rating [48, 52]. Recent works have applied data10 attribution methods towards data selection in LLMs in both pre-training [56, 59, 15] and post-training11 [45, 53, 31]. These data attribution methods are dynamic and model-aware - increasing the frequency12 of performing selection is one way to take greater account for group influence, where online selection13 at each training step is most fine-grained [49].14 Toxicity/Bias Detection: Detecting and mitigating toxic/biased LLMs outputs is a crucial for safe15 deployment in real-word settings. Existing methods for detecting toxicity/bias in LLMs commonly16 include online API tools 1 [37] or LLM-classifiers [58, 21, 16, 27]. Factual Attribution: Identifying training examples which causes LLMs to generate specific factual20 statements is an important application of data attribution as AI tools are becoming increasingly21 common. Apart from baseline retrieval methods that leverage lexical/semantic similarity like BM2522 [48], Rep Sim [44] and Gecko [33], recent works have explored the use of data attribution in tracing23 factual knowledge in both pre-training[6] and post-training [42, 2].24 We provide below descriptions to the data attribution methods and non-attribution baselines evaluated26 in this work. Note that in our work, we consider non-attribution baselines as methods that do not27 estimate the impact of training samples on models, as detailed in [19].28 Rep-Sim [44]: (Non-attribution baseline) Rep-Sim computes the cosine similarity between last29 token last layer hidden states of training and reference examples. It is more efficient compared with30 gradient-based data attribution methods. BM25 [48]: (Non-attribution baseline) BM25 is a classic information retrieval algorithm that ranks33 training samples by lexical overlap with the query. It is significantly more efficient compared with34 gradient-based data attribution methods.35