Search

We investigate the logical reasoning capabilities of large language models (LLMs) and their scalability in complex non-monotonic reasoning. To this end, we introduce ZebraLogic, a comprehensive evaluation framework for assessing LLM reasoning performance on logic grid puzzles derived from constraint satisfaction problems (CSPs). ZebraLogic enables the generation of puzzles with controllable and quantifiable complexity, facilitating a systematic study of the scaling limits of models such as Llama, o1 models, and DeepSeek-R1. By encompassing a broad range of search space complexities and diverse logical constraints, ZebraLogic provides a structured environment to evaluate reasoning under increasing difficulty. Our results reveal a significant decline in accuracy as problem complexity grows -- a phenomenon we term the curse of complexity. This limitation persists even with larger models and increased inference-time computation, suggesting inherent constraints in current LLM reasoning capabilities. Additionally, we explore strategies to enhance logical reasoning, including Best-of-N sampling, backtracking mechanisms, and self-verification prompts. Our findings offer critical insights into the scalability of LLM reasoning, highlight fundamental limitations, and outline potential directions for improvement., Comment: Website: https://huggingface.co/spaces/WildEval/ZebraLogic

Electroencephalography (EEG) is an invaluable tool in neuroscience, offering insights into brain activity with high temporal resolution. Recent advancements in machine learning and generative modeling have catalyzed the application of EEG in reconstructing perceptual experiences, including images, videos, and audio. This paper systematically reviews EEG-to-output research, focusing on state-of-the-art generative methods, evaluation metrics, and data challenges. Using PRISMA guidelines, we analyze 1800 studies and identify key trends, challenges, and opportunities in the field. The findings emphasize the potential of advanced models such as Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), and Transformers, while highlighting the pressing need for standardized datasets and cross-subject generalization. A roadmap for future research is proposed that aims to improve decoding accuracy and broadening real-world applications., Comment: 15 pages. Submitted as a conference paper to IntelliSys 2025

Recent direct preference alignment algorithms (DPA), such as DPO, have shown great promise in aligning large language models to human preferences. While this has motivated the development of many new variants of the original DPO loss, understanding the differences between these recent proposals, as well as developing new DPA loss functions, remains difficult given the lack of a technical and conceptual framework for reasoning about the underlying semantics of these algorithms. In this paper, we attempt to remedy this by formalizing DPA losses in terms of discrete reasoning problems. Specifically, we ask: Given an existing DPA loss, can we systematically derive a symbolic expression that characterizes its semantics? How do the semantics of two losses relate to each other? We propose a novel formalism for characterizing preference losses for single model and reference model based approaches, and identify symbolic forms for a number of commonly used DPA variants. Further, we show how this formal view of preference learning sheds new light on both the size and structure of the DPA loss landscape, making it possible to not only rigorously characterize the relationships between recent loss proposals but also to systematically explore the landscape and derive new loss functions from first principles. We hope our framework and findings will help provide useful guidance to those working on human AI alignment.

Channel and spatial attention mechanisms introduced by earlier works enhance the representation abilities of deep convolutional neural networks (CNNs) but often lead to increased parameter and computation costs. While recent approaches focus solely on efficient feature context modeling for channel attention, we aim to model both channel and spatial attention comprehensively with minimal parameters and reduced computation. Leveraging the principles of relational modeling in graphs, we introduce a constant-parameter module, STEAM: Squeeze and Transform Enhanced Attention Module, which integrates channel and spatial attention to enhance the representation power of CNNs. To our knowledge, we are the first to propose a graph-based approach for modeling both channel and spatial attention, utilizing concepts from multi-head graph transformers. Additionally, we introduce Output Guided Pooling (OGP), which efficiently captures spatial context to further enhance spatial attention. We extensively evaluate STEAM for large-scale image classification, object detection and instance segmentation on standard benchmark datasets. STEAM achieves a 2% increase in accuracy over the standard ResNet-50 model with only a meager increase in GFLOPs. Furthermore, STEAM outperforms leading modules ECA and GCT in terms of accuracy while achieving a three-fold reduction in GFLOPs.

Sensing with RF signals such as mmWave radar has gained considerable interest in recent years. This is particularly relevant to 6G networks, which aim to integrate sensing and communication (ISAC) capabilities for enhanced functionality. The contextual information provided by such sensing, whether collected by standalone non-ISAC units or integrated within ISAC, can not only optimize cellular network assets but can also serve as a valuable tool for a wide range of applications beyond network optimization. In this context, we present a novel methodology for crowd size estimation using monostatic mmWave radar, which is capable of accurately counting large crowds that are unevenly distributed across space. Our estimation approach relies on the rigorous derivation of occlusion probabilities, which are then used to mathematically characterize the probability distributions that describe the number of agents visible to the radar as a function of the crowd size. We then estimate the true crowd size by comparing these derived mathematical models to the empirical distribution of the number of visible agents detected by the radar. This method requires minimal sensing capabilities (e.g., angle-of-arrival information is not needed), thus being well suited for either a dedicated mmWave radar or an ISAC system. Extensive numerical simulations validate our methodology, demonstrating strong performance across diverse spatial distributions and for crowd sizes of up to (and including) 30 agents. We achieve a mean absolute error (MAE) of 0.48 agents, significantly outperforming a baseline which assumes that the agents are uniformly distributed in the area. Overall, our approach holds significant promise for a variety of applications including network resource allocation, crowd management, and urban planning.

Given that Large Language Models (LLMs) have made significant progress in writing code, can they now be used to autonomously reproduce results from research repositories? Such a capability would be a boon to the research community, helping researchers validate, understand, and extend prior work. To advance towards this goal, we introduce SUPER, the first benchmark designed to evaluate the capability of LLMs in setting up and executing tasks from research repositories. SUPERaims to capture the realistic challenges faced by researchers working with Machine Learning (ML) and Natural Language Processing (NLP) research repositories. Our benchmark comprises three distinct problem sets: 45 end-to-end problems with annotated expert solutions, 152 sub problems derived from the expert set that focus on specific challenges (e.g., configuring a trainer), and 602 automatically generated problems for larger-scale development. We introduce various evaluation measures to assess both task success and progress, utilizing gold solutions when available or approximations otherwise. We show that state-of-the-art approaches struggle to solve these problems with the best model (GPT-4o) solving only 16.3% of the end-to-end set, and 46.1% of the scenarios. This illustrates the challenge of this task, and suggests that SUPER can serve as a valuable resource for the community to make and measure progress.

Cargo distribution within eukaryotic cells relies on the active transport mechanisms driven by molecular motors. Despite their critical role, the intricate relationship between motor transport properties and cargo binding - and its impact on motor distribution - remains inadequately understood. Additionally, improper regulation of ubiquitination, a pivotal post-translational modification that affects protein degradation, activation, and localization, is associated with several neurodegenerative diseases. Recent data showed that ubiquitination can alter motor-cargo binding of the Kinesin-3 motor UNC-104 / KIF1A that transports synaptic vesicles. To investigate how ubiquitin-like modifications affect motor protein function, particularly cargo binding, transport properties, and distribution, we utilize the PLM neuron of C. elegans as a model system. Using fluorescent microscopy, we assess the distribution of cargo-bound UNC-104 motors along the axon and probe their dynamics using FRAP experiments. We model cargo binding kinetics with a Master equation and motor density dynamics using a Fokker-Planck approach. Our combined experimental and theoretical analysis reveals that ubiquitin-like knockdowns enhance UNC-104's cooperative binding to its cargo. However, these modifications do not affect UNC-104's transport properties, such as processivity and diffusivity. Thus, while ubiquitin-like modifications significantly impact the cargo-binding of UNC-104, they do not alter its transport dynamics, keeping the homeostatic distribution of UNC-104 unchanged., Comment: 13 pages, 8 figures

Accurate orientation estimation of an object in a scene is critical in robotics, aerospace, augmented reality, and medicine, as it supports scene understanding. This paper introduces a novel orientation estimation approach leveraging radio frequency (RF) sensing technology and leaky-wave antennas (LWAs). Specifically, we propose a framework for a radar system to estimate the orientation of a \textit{dumb} LWA-equipped backscattering tag, marking the first exploration of this method in the literature. Our contributions include a comprehensive framework for signal modeling and orientation estimation with multi-subcarrier transmissions, and the formulation of a maximum likelihood estimator (MLE). Moreover, we analyze the impact of imperfect tag location information, revealing that it minimally affects estimation accuracy. Exploiting related results, we propose an approximate MLE and introduce a low-complexity radiation-pointing angle-based estimator with near-optimal performance. We derive the feasible orientation estimation region of the latter and show that it depends mainly on the system bandwidth. Our analytical results are validated through Monte Carlo simulations and reveal that the low-complexity estimator achieves near-optimal accuracy and that its feasible orientation estimation region is also approximately shared by the other estimators. Finally, we show that the optimal number of subcarriers increases with sensing time under a power budget constraint., Comment: 14 pages, 2 tables, 8 figs. Submitted to IEEE TWC

There are no clean samples in nature. Therefore, when we come to discuss the entanglement properties of quantum materials, the effects of disorder must be taken into account. This question is of particular interest for high-entangled states, such as quantum spin liquids, which lie outside the Landau paradigm for classifying phases of matter. In this work, we explore what experimentally-accessible measures, in the form of concurrence, residual tangle and quantum Fisher information, can teach us about the entanglement in the presence of disorder. As a representative example, we consider the Tomonaga-Luttinger liquids (TLL) and disorder-driven random singlet (RS) states found in antiferromagnetic quantum spin chains. Using quantum Fisher information and residual tangle, we demonstrate that both TLL and RS states exhibit multi-partite entanglement. In the case of the RS state, we attribute this to entanglement localized below a crossover length scale. We further show that the order of disorder average matters in calculating measures like concurrence, and that this can lead to false inferences when interpreting experiment. None the less, correctly interpreted, these witnesses provide useful information about the effects of disorder. We explore how information about the central charge of the TLL can be extracted from the low-temperature behavior of concurrence, and conjecture that this analysis can be extended to the effective central charge of the RS state. Finally, we establish how RS and TLL states can be distinguished through the growth of multi-partite entanglement, as witnessed by the equal-time structure factor. These results establish that, used carefully, experiments based on entanglement witnesses can provide important information about quantum spin systems in the presence of disorder., Comment: More references

Autonomous agents that address day-to-day digital tasks (e.g., ordering groceries for a household), must not only operate multiple apps (e.g., notes, messaging, shopping app) via APIs, but also generate rich code with complex control flow in an iterative manner based on their interaction with the environment. However, existing benchmarks for tool use are inadequate, as they only cover tasks that require a simple sequence of API calls. To remedy this gap, we built $\textbf{AppWorld Engine}$, a high-quality execution environment (60K lines of code) of 9 day-to-day apps operable via 457 APIs and populated with realistic digital activities simulating the lives of ~100 fictitious users. We then created $\textbf{AppWorld Benchmark}$ (40K lines of code), a suite of 750 natural, diverse, and challenging autonomous agent tasks requiring rich and interactive code generation. It supports robust programmatic evaluation with state-based unit tests, allowing for different ways of completing a task while also checking for unexpected changes, i.e., collateral damage. The state-of-the-art LLM, GPT-4o, solves only ~49% of our 'normal' tasks and ~30% of 'challenge' tasks, while other models solve at least 16% fewer. This highlights the benchmark's difficulty and AppWorld's potential to push the frontiers of interactive coding agents. The project website is available at https://appworld.dev/., Comment: ACL'24 Camera Ready

Multiple-choice question answering (MCQA) is a key competence of performant transformer language models that is tested by mainstream benchmarks. However, recent evidence shows that models can have quite a range of performance, particularly when the task format is diversified slightly (such as by shuffling answer choice order). In this work we ask: how do successful models perform formatted MCQA? We employ vocabulary projection and activation patching methods to localize key hidden states that encode relevant information for predicting the correct answer. We find that prediction of a specific answer symbol is causally attributed to a single middle layer, and specifically its multi-head self-attention mechanism. We show that subsequent layers increase the probability of the predicted answer symbol in vocabulary space, and that this probability increase is associated with a sparse set of attention heads with unique roles. We additionally uncover differences in how different models adjust to alternative symbols. Finally, we demonstrate that a synthetic task can disentangle sources of model error to pinpoint when a model has learned formatted MCQA, and show that an inability to separate answer symbol tokens in vocabulary space is a property of models unable to perform formatted MCQA tasks., Comment: Preprint. Code will be available at https://github.com/allenai/understanding_mcqa

An important aspect of 5G networks is the development of Radio Access Network (RAN) slicing, a concept wherein the virtualized infrastructure of wireless networks is subdivided into slices (or enterprises), tailored to fulfill specific use-cases. A key focus in this context is the efficient radio resource allocation to meet various enterprises' service-level agreements (SLAs). In this work, we introduce a channel-aware and SLA-aware RAN slicing framework for massive multiple input multiple output (MIMO) networks where resource allocation extends to incorporate the spatial dimension available through beamforming. Essentially, the same time-frequency resource block (RB) can be shared across multiple users through multiple antennas. Notably, certain enterprises, particularly those operating critical infrastructure, necessitate dedicated RB allocation, denoted as private networks, to ensure security. Conversely, some enterprises would allow resource sharing with others in the public network to maintain network performance while minimizing capital expenditure. Building upon this understanding, the proposed scheduler comprises scheduling schemes under both scenarios: where different slices share the same set of RBs, and where they require exclusivity of allocated RBs. We validate the efficacy of our proposed schedulers through simulation by utilizing a channel data set collected from a real-world massive MIMO testbed. Our assessments demonstrate that resource sharing across slices using our approach can lead up to 60.9% reduction in RB usage compared to other approaches. Moreover, our proposed schedulers exhibit significantly enhanced operational efficiency, with significantly faster running time compared to exhaustive greedy approaches while meeting the stringent 5G sub-millisecond-level latency requirement.

Various generalizations of the scalar, axisymmetric Aretakis "horizon hair" for extremal black holes have recently appeared in the literature. In this paper, we present an expression for a non-axisymmetric Aretakis gravitational charge and its potentially observable imprint at a finite distance from the horizon (Ori-coefficient) in extremal Kerr black hole backgrounds. We offer strong and robust numerical evidence for the existence of this hair and its properties. Specifically, we consider the time evolution of horizon penetrating, quadrupolar and (subdominant) octupolar gravitational perturbations with compact support on extremal Kerr (EK) spacetime. We do this by numerically solving the Teukolsky equation and determining the Aretakis charge values on the horizon and at a finite distance from the black hole., Comment: 7 pages, 9 figures

Using matter-wave interference, we have investigated thermal phase fluctuations in narrow coplanar, concentric rings of ultracold fermionic superfluids. We found that the correlation length decreases with number density, consistent with theoretical expectations. We also observed that increasing the coupling between the rings leads to greater overall coherence in the system. The phase fluctuations increased with a change from periodic to closed boundary conditions as we applied a potential barrier at one point in a ring. These results are relevant for the implementation of proposals to utilize ultracold quantum gases in large and elongated circuit-like geometries, especially those that require deterministic preparation and control of quantized circulation states., Comment: 10 pages, 8 figures

Can the rapid advances in code generation, function calling, and data analysis using large language models (LLMs) help automate the search and verification of hypotheses purely from a set of provided datasets? To evaluate this question, we present DiscoveryBench, the first comprehensive benchmark that formalizes the multi-step process of data-driven discovery. The benchmark is designed to systematically assess current model capabilities in discovery tasks and provide a useful resource for improving them. Our benchmark contains 264 tasks collected across 6 diverse domains, such as sociology and engineering, by manually deriving discovery workflows from published papers to approximate the real-world challenges faced by researchers, where each task is defined by a dataset, its metadata, and a discovery goal in natural language. We additionally provide 903 synthetic tasks to conduct controlled evaluations across task complexity. Furthermore, our structured formalism of data-driven discovery enables a facet-based evaluation that provides useful insights into different failure modes. We evaluate several popular LLM-based reasoning frameworks using both open and closed LLMs as baselines on DiscoveryBench and find that even the best system scores only 25%. Our benchmark, thus, illustrates the challenges in autonomous data-driven discovery and serves as a valuable resource for the community to make progress., Comment: Website: https://github.com/allenai/discoverybench

State-space models (SSMs) have emerged as a potential alternative architecture for building large language models (LLMs) compared to the previously ubiquitous transformer architecture. One theoretical weakness of transformers is that they cannot express certain kinds of sequential computation and state tracking (Merrill & Sabharwal, 2023), which SSMs are explicitly designed to address via their close architectural similarity to recurrent neural networks (RNNs). But do SSMs truly have an advantage (over transformers) in expressive power for state tracking? Surprisingly, the answer is no. Our analysis reveals that the expressive power of SSMs is limited very similarly to transformers: SSMs cannot express computation outside the complexity class $\mathsf{TC}^0$. In particular, this means they cannot solve simple state-tracking problems like permutation composition. It follows that SSMs are provably unable to accurately track chess moves with certain notation, evaluate code, or track entities in a long narrative. To supplement our formal analysis, we report experiments showing that Mamba-style SSMs indeed struggle with state tracking. Thus, despite its recurrent formulation, the "state" in an SSM is an illusion: SSMs have similar expressiveness limitations to non-recurrent models like transformers, which may fundamentally limit their ability to solve real-world state-tracking problems., Comment: To appear at ICML 2024. 9 pages + appendices

We present Alljoined1, a dataset built specifically for EEG-to-Image decoding. Recognizing that an extensive and unbiased sampling of neural responses to visual stimuli is crucial for image reconstruction efforts, we collected data from 8 participants looking at 10,000 natural images each. We have currently gathered 46,080 epochs of brain responses recorded with a 64-channel EEG headset. The dataset combines response-based stimulus timing, repetition between blocks and sessions, and diverse image classes with the goal of improving signal quality. For transparency, we also provide data quality scores. We publicly release the dataset and all code at https://linktr.ee/alljoined1., Comment: 8 Pages, 6 Figures

We study the sequence-to-sequence mapping capacity of transformers by relating them to finite transducers, and find that they can express surprisingly large classes of transductions. We do so using variants of RASP, a programming language designed to help people "think like transformers," as an intermediate representation. We extend the existing Boolean variant B-RASP to sequence-to-sequence functions and show that it computes exactly the first-order rational functions (such as string rotation). Then, we introduce two new extensions. B-RASP[pos] enables calculations on positions (such as copying the first half of a string) and contains all first-order regular functions. S-RASP adds prefix sum, which enables additional arithmetic operations (such as squaring a string) and contains all first-order polyregular functions. Finally, we show that masked average-hard attention transformers can simulate S-RASP., Comment: To appear in Transactions of the Association for Computational Linguistics

With the accumulation of data at an unprecedented rate, its potential to fuel scientific discovery is growing exponentially. This position paper urges the Machine Learning (ML) community to exploit the capabilities of large generative models (LGMs) to develop automated systems for end-to-end data-driven discovery -- a paradigm encompassing the search and verification of hypotheses purely from a set of provided datasets, without the need for additional data collection or physical experiments. We first outline several desiderata for an ideal data-driven discovery system. Then, through DATAVOYAGER, a proof-of-concept utilizing GPT-4, we demonstrate how LGMs fulfill several of these desiderata -- a feat previously unattainable -- while also highlighting important limitations in the current system that open up opportunities for novel ML research. We contend that achieving accurate, reliable, and robust end-to-end discovery systems solely through the current capabilities of LGMs is challenging. We instead advocate for fail-proof tool integration, along with active user moderation through feedback mechanisms, to foster data-driven scientific discoveries with efficiency and reproducibility.

Semi-structured interviews (SSIs) are a commonly employed data-collection method in healthcare research, offering in-depth qualitative insights into subject experiences. Despite their value, the manual analysis of SSIs is notoriously time-consuming and labor-intensive, in part due to the difficulty of extracting and categorizing emotional responses, and challenges in scaling human evaluation for large populations. In this study, we develop RACER, a Large Language Model (LLM) based expert-guided automated pipeline that efficiently converts raw interview transcripts into insightful domain-relevant themes and sub-themes. We used RACER to analyze SSIs conducted with 93 healthcare professionals and trainees to assess the broad personal and professional mental health impacts of the COVID-19 crisis. RACER achieves moderately high agreement with two human evaluators (72%), which approaches the human inter-rater agreement (77%). Interestingly, LLMs and humans struggle with similar content involving nuanced emotional, ambivalent/dialectical, and psychological statements. Our study highlights the opportunities and challenges in using LLMs to improve research efficiency and opens new avenues for scalable analysis of SSIs in healthcare research.

The number of Hindi speakers on social media has increased dramatically in recent years. Regret is a common emotional experience in our everyday life. Many speakers on social media, share their regretful experiences and opinions regularly. It might cause a re-evaluation of one's choices and a desire to make a different option if given the chance. As a result, knowing the source of regret is critical for investigating its impact on behavior and decision-making. This study focuses on regret and how it is expressed, specifically in Hindi, on various social media platforms. In our study, we present a novel dataset from three different sources, where each sentence has been manually classified into one of three classes "Regret by action", "Regret by inaction", and "No regret". Next, we use this dataset to investigate the linguistic expressions of regret in Hindi text and also identify the textual domains that are most frequently associated with regret. Our findings indicate that individuals on social media platforms frequently express regret for both past inactions and actions, particularly within the domain of interpersonal relationships. We use a pre-trained BERT model to generate word embeddings for the Hindi dataset and also compare deep learning models with conventional machine learning models in order to demonstrate accuracy. Our results show that BERT embedding with CNN consistently surpassed other models. This described the effectiveness of BERT for conveying the context and meaning of words in the regret domain.

In-context learning (ICL) is an appealing approach for semantic parsing due to its few-shot nature and improved generalization. However, learning to parse to rare domain-specific languages (DSLs) from just a few demonstrations is challenging, limiting the performance of even the most capable LLMs. In this work, we improve the effectiveness of ICL for semantic parsing by (1) using general-purpose programming languages such as Python instead of DSLs, and (2) augmenting prompts with a structured domain description that includes, e.g., the available classes and functions. We show that both these changes significantly improve accuracy across three popular datasets. Combined, they lead to dramatic improvements (e.g. 7.9% to 66.5% on SMCalFlow compositional split), nearly closing the performance gap between easier i.i.d.\ and harder compositional splits when used with a strong model, and reducing the need for a large number of demonstrations. We find that the resemblance of the target parse language to general-purpose code is a more important factor than the language's popularity in pre-training corpora. Our findings provide an improved methodology for building semantic parsers in the modern context of ICL with LLMs., Comment: Accepted to NAACL 2024

Large Language Models (LLMs) are increasingly being used for interactive decision-making tasks requiring planning and adapting to the environment. Recent works employ LLMs-as-agents in broadly two ways: iteratively determining the next action (iterative executors) or generating plans and executing sub-tasks using LLMs (plan-and-execute). However, these methods struggle with task complexity, as the inability to execute any sub-task may lead to task failure. To address these shortcomings, we introduce As-Needed Decomposition and Planning for complex Tasks (ADaPT), an approach that explicitly plans and decomposes complex sub-tasks as-needed, i.e., when the LLM is unable to execute them. ADaPT recursively decomposes sub-tasks to adapt to both task complexity and LLM capability. Our results demonstrate that ADaPT substantially outperforms established strong baselines, achieving success rates up to 28.3% higher in ALFWorld, 27% in WebShop, and 33% in TextCraft -- a novel compositional dataset that we introduce. Through extensive analysis, we illustrate the importance of multilevel decomposition and establish that ADaPT dynamically adjusts to the capabilities of the executor LLM as well as to task complexity., Comment: NAACL 2024 (findings) camera-ready. Project Page: https://allenai.github.io/adaptllm

Recent works have showcased the ability of LLMs to embody diverse personas in their responses, exemplified by prompts like 'You are Yoda. Explain the Theory of Relativity.' While this ability allows personalization of LLMs and enables human behavior simulation, its effect on LLMs' capabilities remains unclear. To fill this gap, we present the first extensive study of the unintended side-effects of persona assignment on the ability of LLMs to perform basic reasoning tasks. Our study covers 24 reasoning datasets, 4 LLMs, and 19 diverse personas (e.g. an Asian person) spanning 5 socio-demographic groups. Our experiments unveil that LLMs harbor deep rooted bias against various socio-demographics underneath a veneer of fairness. While they overtly reject stereotypes when explicitly asked ('Are Black people less skilled at mathematics?'), they manifest stereotypical and erroneous presumptions when asked to answer questions while adopting a persona. These can be observed as abstentions in responses, e.g., 'As a Black person, I can't answer this question as it requires math knowledge', and generally result in a substantial performance drop. Our experiments with ChatGPT-3.5 show that this bias is ubiquitous - 80% of our personas demonstrate bias; it is significant - some datasets show performance drops of 70%+; and can be especially harmful for certain groups - some personas suffer statistically significant drops on 80%+ of the datasets. Overall, all 4 LLMs exhibit this bias to varying extents, with GPT-4-Turbo showing the least but still a problematic amount of bias (evident in 42% of the personas). Further analysis shows that these persona-induced errors can be hard-to-discern and hard-to-avoid. Our findings serve as a cautionary tale that the practice of assigning personas to LLMs - a trend on the rise - can surface their deep-rooted biases and have unforeseeable and detrimental side-effects., Comment: Project page: https://allenai.github.io/persona-bias. Paper to appear at ICLR 2024. Added results for other LLMs in v2 (similar findings)

Quantitative evaluation metrics have traditionally been pivotal in gauging the advancements of artificial intelligence systems, including large language models (LLMs). However, these metrics have inherent limitations. Given the intricate nature of real-world tasks, a single scalar to quantify and compare is insufficient to capture the fine-grained nuances of model behavior. Metrics serve only as a way to compare and benchmark models, and do not yield actionable diagnostics, thus making the model improvement process challenging. Model developers find themselves amid extensive manual efforts involving sifting through vast datasets and attempting hit-or-miss adjustments to training data or setups. In this work, we address the shortcomings of quantitative metrics by proposing QualEval, which augments quantitative scalar metrics with automated qualitative evaluation as a vehicle for model improvement. QualEval uses a powerful LLM reasoner and our novel flexible linear programming solver to generate human-readable insights that when applied, accelerate model improvement. The insights are backed by a comprehensive dashboard with fine-grained visualizations and human-interpretable analyses. We corroborate the faithfulness of QualEval by demonstrating that leveraging its insights, for example, improves the absolute performance of the Llama 2 model by up to 15% points relative on a challenging dialogue task (DialogSum) when compared to baselines. QualEval successfully increases the pace of model development, thus in essence serving as a data-scientist-in-a-box. Given the focus on critiquing and improving current evaluation metrics, our method serves as a refreshingly new technique for both model evaluation and improvement., Comment: NAACL 2024

We have observed the spontaneous appearance of currents in a ring of ultracold fermionic atoms (6Li) with attractive interactions, following a quench to a BCS-like pair superfluid. We have measured the winding number probability distribution for a range of quench rates, with a quench protocol using simultaneous forced evaporation and interaction ramps to achieve faster effective quench rates with less atom loss than a purely evaporative quench. We find that for the fastest quenches the mean square winding number of the current follows a scaling law in the quench rate with exponent {\sigma} = 0.24(2), which is somewhat lower than that predicted by the Kibble-Zurek mechanism (KZM) for the three-dimensional XY model (1/3), and unexpectedly closer to the value obtained from mean-field theory (1/4). For slower quenches non-universal effects become significant, and we observe a lower rate of spontaneous current formation that does not follow a simple scaling law., Comment: 10 pages, 4 figures

As wireless communication systems strive to improve spectral efficiency, there has been a growing interest in employing machine learning (ML)-based approaches for adaptive modulation and coding scheme (MCS) selection. In this paper, we introduce a new adaptive MCS selection framework for massive MIMO systems that operates without any feedback from users by solely relying on instantaneous uplink channel estimates. Our proposed method can effectively operate in multi-user scenarios where user feedback imposes excessive delay and bandwidth overhead. To learn the mapping between the user channel matrices and the optimal MCS level of each user, we develop a Convolutional Neural Network (CNN)-Long Short-Term Memory Network (LSTM)-based model and compare the performance with the state-of-the-art methods. Finally, we validate the effectiveness of our algorithm by evaluating it experimentally using real-world datasets collected from the RENEW massive MIMO platform.

Recent theoretical work has identified surprisingly simple reasoning problems, such as checking if two nodes in a graph are connected or simulating finite-state machines, that are provably unsolvable by standard transformers that answer immediately after reading their input. However, in practice, transformers' reasoning can be improved by allowing them to use a "chain of thought" or "scratchpad", i.e., generate and condition on a sequence of intermediate tokens before answering. Motivated by this, we ask: Does such intermediate generation fundamentally extend the computational power of a decoder-only transformer? We show that the answer is yes, but the amount of increase depends crucially on the amount of intermediate generation. For instance, we find that transformer decoders with a logarithmic number of decoding steps (w.r.t. the input length) push the limits of standard transformers only slightly, while a linear number of decoding steps, assuming projected pre-norm (a slight generalization of standard pre-norm), adds a clear new ability (under standard complexity conjectures): recognizing all regular languages. Our results also imply that linear steps keep transformer decoders within context-sensitive languages, and polynomial steps with generalized pre-norm make them recognize exactly the class of polynomial-time solvable problems -- the first exact characterization of a type of transformers in terms of standard complexity classes. Together, this provides a nuanced framework for understanding how the length of a transformer's chain of thought or scratchpad impacts its reasoning power., Comment: 9-page preprint. ICLR camera ready posted April 11

Despite their ubiquity in language generation, it remains unknown why truncation sampling heuristics like nucleus sampling are so effective. We provide a theoretical explanation for the effectiveness of the truncation sampling by proving that truncation methods that discard tokens below some probability threshold (the most common type of truncation) can guarantee that all sampled tokens have nonzero true probability. However, thresholds are a coarse heuristic, and necessarily discard some tokens with nonzero true probability as well. In pursuit of a more precise sampling strategy, we show that we can leverage a known source of model errors, the softmax bottleneck, to prove that certain tokens have nonzero true probability, without relying on a threshold. Based on our findings, we develop an experimental truncation strategy and the present pilot studies demonstrating the promise of this type of algorithm. Our evaluations show that our method outperforms its threshold-based counterparts under automatic and human evaluation metrics for low-entropy (i.e., close to greedy) open-ended text generation. Our theoretical findings and pilot experiments provide both insight into why truncation sampling works, and make progress toward more expressive sampling algorithms that better surface the generative capabilities of large language models.