Electrocardiogram (ECG) foundation models pretrained on typical diagnostic 10-second ECG segments, have demonstrated strong transferability across a range of clinical applications. However, many real-world applications produce recordings that are typically longer, and are varied in duration during inference time. These 10-second models have no built-in way to combine information across time. Extending them to longer horizons introduces two challenges: structural incompatibilities arising from input-length disparities, and semantic challenges that limit meaningful temporal aggregation. We propose a parameter-efficient framework that extends pretrained ECG foundation models to longer and variable-length ECGs without retraining the backbone. Guided by a frozen pretrained 10-second model, we introduce a lightweight plug-in module that extends the model in two complementary ways: (i) structurally compatible long-sequence processing and (ii) semantically informed temporal modeling. Experiments on multiple long-horizon ECG tasks, datasets, and foundation model backbones demonstrate that our method enables robust long-horizon extension from pretrained snapshot models, consistently outperfo
Reasoning-trained language models often spend more tokens on harder problems, but longer chains of thought do not show whether a model is merely computing for more steps or following a different internal trajectory. We study this distinction through hidden-state trajectories during chain-of-thought generation across competitive programming, mathematics, and Boolean satisfiability. Raw trajectory geometry is strongly shaped by generation length: longer generations mechanically alter path statistics, so difficulty-dependent comparisons are misleading without adjustment. After residualizing trajectory statistics on length, difficulty remains systematically coupled to corrected trajectory geometry across all domains studied. The clearest reasoning-specific separation appears in the code domain, where harder problems show more direct corrected trajectories and less heterogeneous local curvature in reasoning-trained models than in matched instruction-tuned baselines. Corrected difficulty-geometry coupling is weaker, but still present, in mathematics and Boolean satisfiability. Prompt-stage linear probes do not mirror the code-domain separation, and behavioral annotations show that strong
Regularization is a technique to improve generalization of machine learning (ML) models. A common form of regularization in the ML literature is to train on data where similar inputs map to different outputs. This improves generalization by preventing ML models from becoming overconfident in their predictions. This paper shows how using longer timesteps when modelling chaotic Earth systems naturally leads to more of this regularization. We show this in two domains. We explain how using longer model timesteps can improve results and demonstrate that increased regularization is one of the causes. We explain why longer model timesteps lead to improved regularization in these systems and present a procedure to pick the model timestep. We also carry out a benchmarking exercise on ORAS5 ocean reanalysis data to show that a longer model timestep (28 days) than is typically used gives realistic simulations. We suggest that there will be many opportunities to use this type of regularization in Earth system problems because the Earth system is chaotic and the regularization is so easy to implement.
In 2007, Erika Bohm-Vitense published a provocative figure suggesting that the solar rotation period and activity cycle made the Sun an outlier compared to the trends observed for stars in the Mount Wilson HK survey. A decade later, after the discovery of weakened magnetic braking (WMB), an evolutionary scenario was proposed that could account for the properties of the Sun if activity cycles grow longer and weaker in the WMB regime. Recent observations of the gradual onset of WMB suggest that the efficiency of the global stellar dynamo declines by at least two orders of magnitude as the stellar Rossby number approaches a critical point slightly above the solar value. A new sample of activity cycle data from the California Legacy Survey suggests that the Sun is not an outlier, and unambiguously confirms that activity cycles grow longer and weaker on stellar evolutionary timescales.
Longer-running scams, such as romance fraud and "pig-butchering" scams, exploit not only victims' emotions but also the design of digital platforms. Scammers commonly leverage features such as professional-looking profile verification, algorithmic recommendations that reinforce contact, integrated payment systems, and private chat affordances to gradually establish trust and dependency with victims. Prior work in HCI and criminology has examined online scams through the lenses of detection mechanisms, threat modeling, and user-level vulnerabilities. However, less attention has been paid to how platform design itself enables longer-running scams. To address this gap, we conducted in-depth interviews with 25 longer-running scam victims in China. Our findings show how scammers strategically use platform affordances to stage credibility, orchestrate intimacy, and sustain coercion with victims. By analyzing scams as socio-technical projects, we highlight how platform design can be exploited in longer-running scams, and point to redesigning future platforms to better protect users.
Modeling ultra-long user behavior sequences is critical for capturing both long- and short-term preferences in industrial recommender systems. Existing solutions typically rely on two-stage retrieval or indirect modeling paradigms, incuring upstream-downstream inconsistency and computational inefficiency. In this paper, we present LONGER, a Long-sequence Optimized traNsformer for GPU-Efficient Recommenders. LONGER incorporates (i) a global token mechanism for stabilizing attention over long contexts, (ii) a token merge module with lightweight InnerTransformers and hybrid attention strategy to reduce quadratic complexity, and (iii) a series of engineering optimizations, including training with mixed-precision and activation recomputation, KV cache serving, and the fully synchronous model training and serving framework for unified GPU-based dense and sparse parameter updates. LONGER consistently outperforms strong baselines in both offline metrics and online A/B testing in both advertising and e-commerce services at ByteDance, validating its consistent effectiveness and industrial-level scaling laws. Currently, LONGER has been fully deployed at more than 10 influential scenarios at B
Flag codes generalize constant dimension codes by considering sequences of nested subspaces with prescribed dimensions as codewords. A comprehensive construction, which unites cyclic orbit flag codes, yields two families of flag codes on $\mathbb{F}^n_q$ (where $n=sk+h$ with $s\geq 2$ and $0\leq h < k$): optimum distance flag codes of the longest possible type vector $(1, 2, \ldots, k, n-k, \ldots, n-1)$ and flag codes with longer type vectors $(1, 2, \ldots, k+h, 2k+h, \ldots, (s-2)k+h, n-k, \ldots, n-1)$. These flag codes achieve the same cardinality $\sum^{s-1}_{i=1}q^{ik+h}+1$.
As a pioneering vision-language model, CLIP (Contrastive Language-Image Pre-training) has achieved significant success across various domains and a wide range of downstream vision-language tasks. However, the text encoders in popular CLIP models are limited to processing only 77 text tokens, which constrains their ability to effectively handle longer, detail-rich captions. Additionally, CLIP models often struggle to effectively capture detailed visual and textual information, which hampers their performance on tasks that require fine-grained analysis. To address these limitations, we present a novel approach, \textbf{FineLIP}, that extends the capabilities of CLIP. FineLIP enhances cross-modal text-image mapping by incorporating \textbf{Fine}-grained alignment with \textbf{L}onger text input within the CL\textbf{IP}-style framework. FineLIP first extends the positional embeddings to handle longer text, followed by the dynamic aggregation of local image and text tokens. The aggregated results are then used to enforce fine-grained token-to-token cross-modal alignment. We validate our model on datasets with long, detailed captions across two tasks: zero-shot cross-modal retrieval and
Many centralized mechanisms for two-sided matching markets that enjoy strong theoretical properties assume that the planner solicits full information on the preferences of each participating agent. In particular, they expect that participants compile and communicate their complete preference lists over agents from the other side of the market. However, real-world markets are often very large and agents cannot always be expected to even produce a ranking of all options on the other side. It is therefore important to understand the impact of incomplete or truncated lists on the quality of the resultant matching. In this paper, we focus on the Serial Dictatorship mechanism in a model where each agent of the proposing side (students) has a random preference list of length $d$, sampled independently and uniformly at random from $n$ schools, each of which has one seat. Our main result shows that if the students primarily care about being matched to any school of their list (as opposed to ending up unmatched), then all students in position $i\leq n$ will prefer markets with longer lists, when $n$ is large enough. Schools on the other hand will always prefer longer lists in our model. We m
Large Vision-Language Models (LVLMs) have made significant progress in recent years but are also prone to hallucination issues. They exhibit more hallucinations in longer, free-form responses, often attributed to accumulated uncertainties. In this paper, we ask: Does increased hallucination result solely from length-induced errors, or is there a deeper underlying mechanism? After a series of preliminary experiments and findings, we suggest that the risk of hallucinations is not caused by length itself but by the increased reliance on context for coherence and completeness in longer responses. Building on these insights, we propose a novel "induce-detect-suppress" framework that actively induces hallucinations through deliberately designed contexts, leverages induced instances for early detection of high-risk cases, and ultimately suppresses potential object-level hallucinations during actual decoding. Our approach achieves consistent, significant improvements across all benchmarks, demonstrating its efficacy. The strong detection and improved hallucination mitigation not only validate our framework but, more importantly, re-validate our hypothesis on context. Rather than solely pur
Despite tremendous progress in the field of text-to-video (T2V) synthesis, open-sourced T2V diffusion models struggle to generate longer videos with dynamically varying and evolving content. They tend to synthesize quasi-static videos, ignoring the necessary visual change-over-time implied in the text prompt. At the same time, scaling these models to enable longer, more dynamic video synthesis often remains computationally intractable. To address this challenge, we introduce the concept of Generative Temporal Nursing (GTN), where we aim to alter the generative process on the fly during inference to improve control over the temporal dynamics and enable generation of longer videos. We propose a method for GTN, dubbed VSTAR, which consists of two key ingredients: 1) Video Synopsis Prompting (VSP) - automatic generation of a video synopsis based on the original single prompt leveraging LLMs, which gives accurate textual guidance to different visual states of longer videos, and 2) Temporal Attention Regularization (TAR) - a regularization technique to refine the temporal attention units of the pre-trained T2V diffusion models, which enables control over the video dynamics. We experiment
Large language models have demonstrated remarkable progress in mathematical reasoning, leveraging chain-of-thought and reinforcement learning. However, many open questions remain regarding the interplay between reasoning token usage and accuracy gains. In particular, when comparing models across generations, it is unclear whether improved performance results from longer reasoning chains or more efficient reasoning. We systematically analyze reasoning chain length across o1-mini and o3-mini variants on the Omni-MATH benchmark, finding that o3-mini (m) achieves superior accuracy without requiring longer reasoning chains than o1-mini. Moreover, we show that accuracy generally declines as reasoning chains grow across all models and compute settings, even when controlling for difficulty of the questions. This accuracy drop is significantly smaller in more proficient models, suggesting that new generations of reasoning models use test-time compute more effectively. Finally, we highlight that while o3-mini (h) achieves a marginal accuracy gain over o3-mini (m), it does so by allocating substantially more reasoning tokens across all problems, even the ones that o3-mini (m) can already solv
Writing longer prompts for an AI assistant to generate a story increases psychological ownership, a user's feeling that the writing belongs to them. To encourage users to write longer prompts, we evaluated two interaction techniques that modify the prompt entry interface of chat-based generative AI assistants: pressing and holding the prompt submission button, and continuously moving a slider up and down when submitting a short prompt. A within-subjects experiment investigated the effects of such techniques on prompt length and psychological ownership, and results showed that these techniques increased prompt length and led to higher psychological ownership than baseline techniques. A second experiment further augmented these techniques by showing AI-generated suggestions for how the prompts could be expanded. This further increased prompt length, but did not lead to improvements in psychological ownership. Our results show that simple interface modifications like these can elicit more writing from users and improve psychological ownership.
This paper presents a new method for training sequence-to-sequence models for speech recognition and translation tasks. Instead of the traditional approach of training models on short segments containing only lowercase or partial punctuation and capitalization (PnC) sentences, we propose training on longer utterances that include complete sentences with proper punctuation and capitalization. We achieve this by using the FastConformer architecture which allows training 1 Billion parameter models with sequences up to 60 seconds long with full attention. However, while training with PnC enhances the overall performance, we observed that accuracy plateaus when training on sequences longer than 40 seconds across various evaluation settings. Our proposed method significantly improves punctuation and capitalization accuracy, showing a 25% relative word error rate (WER) improvement on the Earnings-21 and Earnings-22 benchmarks. Additionally, training on longer audio segments increases the overall model accuracy across speech recognition and translation benchmarks. The model weights and training code are open-sourced though NVIDIA NeMo.
Direct reciprocity is a wide-spread mechanism for evolution of cooperation. In repeated interactions, players can condition their behavior on previous outcomes. A well known approach is given by reactive strategies, which respond to the co-player's previous move. Here we extend reactive strategies to longer memories. A reactive-$n$ strategy takes into account the sequence of the last $n$ moves of the co-player. A reactive-$n$ counting strategy records how often the co-player has cooperated during the last $n$ rounds. We derive an algorithm to identify all partner strategies among reactive-$n$ strategies. We give explicit conditions for all partner strategies among reactive-2, reactive-3 strategies, and reactive-$n$ counting strategies. Partner strategies are those that ensure mutual cooperation without exploitation. We perform evolutionary simulations and find that longer memory increases the average cooperation rate for reactive-$n$ strategies but not for reactive counting strategies. Paying attention to the sequence of moves is necessary for reaping the advantages of longer memory.
In a world where deepfakes and cloned voices are emerging as sophisticated attack vectors, organizations require a new security mindset: Sensorial Zero Trust [9]. This article presents a scientific analysis of the need to systematically doubt information perceived through the senses, establishing rigorous verification protocols to mitigate the risks of fraud based on generative artificial intelligence. Key concepts, such as Out-of-Band verification, Vision-Language Models (VLMs) as forensic collaborators, cryptographic provenance, and human training, are integrated into a framework that extends Zero Trust principles to human sensory information. The approach is grounded in empirical findings and academic research, emphasizing that in an era of AI-generated realities, even our eyes and ears can no longer be implicitly trusted without verification. Leaders are called to foster a culture of methodological skepticism to protect organizational integrity in this new threat landscape.
The feeling of something belonging to someone is called "psychological ownership." A common assumption is that writing with generative AI lowers psychological ownership, but the extent to which this occurs and the role of prompt length are unclear. We report on two experiments to examine the relationship between psychological ownership and prompt length. Participants wrote short stories either completely by themselves or wrote prompts of varying lengths. Results show that when participants wrote longer prompts, they had higher levels of psychological ownership. Their comments suggest they thought more about their prompts, often adding more details about the plot. However, benefits plateaued when prompt length was 75-100% of the target story length. To encourage users to write longer prompts, we propose augmenting the prompt submission button so it must be held down a long time if the prompt is short. Results show that this technique is effective at increasing prompt length.
When traveling by car from one location to another, our route is constrained by the road network. The network distance between the two locations is generally longer than the geodetic distance as the crow flies. We report a systematic relation between the statistical properties of these two distances. Empirically, we find a robust scaling between network and geodetic distance distributions for a variety of large motorway networks. A simple consequence is that we typically have to drive $1.3\pm0.1$ times longer than the crow flies. This scaling is not present in standard random networks; rather, it requires non-random adjacency. We develop a set of rules to build a realistic motorway network, also consistent with the above scaling. We hypothesize that the scaling reflects a compromise between two societal needs: high efficiency and accessibility on the one hand, and limitation of costs and other burdens on the other.
Advancements in Large Language Models (LLMs) inspire various strategies for integrating video modalities. A key approach is Video-LLMs, which incorporate an optimizable interface linking sophisticated video encoders to LLMs. However, due to computation and data limitations, these Video-LLMs are typically pre-trained to process only short videos, limiting their broader application for understanding longer video content. Additionally, fine-tuning Video-LLMs to handle longer videos is cost-prohibitive. Consequently, it becomes essential to explore the interpolation of Video-LLMs under a completely training-free setting. In this paper, we first identify the primary challenges in interpolating Video-LLMs: (1) the video encoder and modality alignment projector are fixed, preventing the integration of additional frames into Video-LLMs, and (2) the LLM backbone is limited in its content length capabilities, which complicates the processing of an increased number of video tokens. To address these challenges, we propose a specific INTerPolation method for Video-LLMs (INTP-Video-LLMs). We introduce an alternative video token rearrangement technique that circumvents limitations imposed by the
Hydrodynamic modulation of short ocean surface waves by longer ambient waves significantly influences remote sensing, interpretation of in situ wave measurements, and numerical wave forecasting. This paper revisits the wave crest and action conservation laws and derives steady, nonlinear, analytical solutions for the change of short-wave wavenumber, action, and gravitational acceleration due to the presence of longer waves. We validate the analytical solutions with numerical simulations of the full crest and action conservation equations. The nonlinear analytical solutions of short-wave wavenumber, amplitude, and steepness modulation significantly deviate from the linear analytical solutions of Longuet-Higgins & Stewart (1960), and are similar to the nonlinear numerical solutions by Longuet-Higgins (1987) and Zhang & Melville (1990). The short-wave steepness modulation is attributed 5/8 to wavenumber, 1/4 due to wave action, and 1/8 due to effective gravity. Examining the homogeneity and stationarity requirements for the conservation of wave action reveals that stationarity is a stronger requirement and is generally not satisfied for very steep long waves. We examine the re