Search

Large Language Models (LLMs) can generate content that is as persuasive as human-written text and appear capable of selectively producing deceptive outputs. These capabilities raise concerns about potential misuse and unintended consequences as these systems become more widely deployed. This review synthesizes recent empirical work examining LLMs' capacity and proclivity for persuasion and deception, analyzes theoretical risks that could arise from these capabilities, and evaluates proposed mitigations. While current persuasive effects are relatively small, various mechanisms could increase their impact, including fine-tuning, multimodality, and social factors. We outline key open questions for future research, including how persuasive AI systems might become, whether truth enjoys an inherent advantage over falsehoods, and how effective different mitigation strategies may be in practice., Comment: 37 pages, 1 figure

Language models perform differently across languages. It has been previously suggested that morphological typology may explain some of this variability (Cotterell et al., 2018). We replicate previous analyses and find additional new evidence for a performance gap between agglutinative and fusional languages, where fusional languages, such as English, tend to have better language modeling performance than morphologically more complex languages like Turkish. We then propose and test three possible causes for this performance gap: morphological alignment of tokenizers, tokenization quality, and disparities in dataset sizes and measurement. To test the morphological alignment hypothesis, we present MorphScore, a tokenizer evaluation metric, and supporting datasets for 22 languages. We find some evidence that tokenization quality explains the performance gap, but none for the role of morphological alignment. Instead we find that the performance gap is most reduced when training datasets are of equivalent size across language types, but only when scaled according to the so-called "byte-premium" -- the different encoding efficiencies of different languages and orthographies. These results suggest that no language is harder or easier for a language model to learn on the basis of its morphological typology. Differences in performance can be attributed to disparities in dataset size. These results bear on ongoing efforts to improve performance for low-performing and under-resourced languages., Comment: 9 pages

Non-small cell lung cancer (NSCLC) is often intrinsically resistant to several first- and second-line therapeutics and can rapidly acquire further resistance after a patient begins receiving treatment. Treatment outcomes are therefore significantly impacted by the optimization of therapeutic scheduling. Previous preclinical research has suggested scheduling bevacizumab in sequence with combination antiproliferatives could significantly improve clinical outcomes. Mathematical modeling is a well-suited tool for investigating this proposed scheduling modification. To address this critical need, individual patient tumor data from 11 clinical trials in NSCLC has been collated and used to develop a semi-mechanistic model of NSCLC growth and response to the various therapeutics represented in those trials. Precise estimates of clinical parameters fundamental to cancer modeling have been produced - such as the rate of acquired resistance to various pharmaceuticals, the relationship between drug concentration and cancer cell death, as well as the fine temporal dynamics of vascular remodeling in response to bevacizumab. In a reserved portion of the dataset, this model was used to predict the efficacy of individual treatment time courses with a mean error rate of 59.7% after a single tumor measurement and 11.7% after three successive tumor measurements. A delay of 9.6 hours between pemetrexed-cisplatin and bevacizumab administration is predicted to optimize the benefit of sequential administration. At this gap, approximately 93.5% of simulated patients benefited from a gap in sequential administration compared with concomitant administration. Of those simulated patients, the mean improvement in tumor reduction was 20.7%. This result suggests that scheduling a modest gap between the administration of bevacizumab and its partner antiproliferatives could meaningfully improve patient outcomes in NSCLC., Comment: arXiv admin note: text overlap with arXiv:2401.07719

In recent years, the rapid evolution of satellite communications play a pivotal role in addressing the ever-increasing demand for global connectivity, among which the Low Earth Orbit (LEO) satellites attract a great amount of attention due to their low latency and high data throughput capabilities. Based on this, we explore spatial modulation (SM) and space shift keying (SSK) designs as pivotal techniques to enhance spectral efficiency (SE) and bit-error rate (BER) performance in the LEO satellite-assisted multiple-input multiple-output (MIMO) systems. The various performance analysis of these designs are presented in this paper, revealing insightful findings and conclusions through analytical methods and Monte Carlo simulations with perfect and imperfect channel state information (CSI) estimation. The results provide a comprehensive analysis of the merits and trade-offs associated with the investigated schemes, particularly in terms of BER, computational complexity, and SE. This analysis underscores the potential of both schemes as viable candidates for future 6G LEO satellite-assisted wireless communication systems.

For many low-resource languages, the only available language models are large multilingual models trained on many languages simultaneously. However, using FLORES perplexity as a metric, we find that these models perform worse than bigrams for many languages (e.g. 24% of languages in XGLM 4.5B; 43% in BLOOM 7.1B). To facilitate research that focuses on low-resource languages, we pre-train and release Goldfish, a suite of monolingual autoregressive Transformer language models up to 125M parameters for 350 languages. The Goldfish reach lower FLORES perplexities than BLOOM, XGLM, and MaLA-500 on 98 of 204 FLORES languages, despite each Goldfish model being over 10x smaller. However, the Goldfish significantly underperform larger multilingual models on reasoning benchmarks, suggesting that for low-resource languages, multilinguality primarily improves general reasoning abilities rather than basic text generation. We release models trained on 5MB (350 languages), 10MB (288 languages), 100MB (166 languages), and 1GB (83 languages) of text data where available. The Goldfish models are available as baselines, fine-tuning sources, or augmentations to existing models in low-resource NLP research, and they are further useful for crosslinguistic studies requiring maximally comparable models across languages.

We present and analyze a framework for designing symplectic neural networks (SympNets) based on geometric integrators for Hamiltonian differential equations. The SympNets are universal approximators in the space of Hamiltonian diffeomorphisms, interpretable and have a non-vanishing gradient property. We also give a representation theory for linear systems, meaning the proposed P-SympNets can exactly parameterize any symplectic map corresponding to quadratic Hamiltonians. Extensive numerical tests demonstrate increased expressiveness and accuracy -- often several orders of magnitude better -- for lower training cost over existing architectures. Lastly, we show how to perform symbolic Hamiltonian regression with SympNets for polynomial systems using backward error analysis., Comment: 33 pages including appendices but not references, 7 figures

The dispersive motion of surface waves is now routinely used to remotely measure the currents close beneath the surface of oceans and other natural flows. The current manifests as wavelength-dependent Doppler shifts in the spatiotemporal wave spectrum, which is obtained by performing a Fourier transform of an observed wave field, a procedure which assumes that the current is horizontally uniform within the field of view. This assumption is frequently not satisfied. We here analyze the effects of the presence of horizontal shear in the velocity field, with special emphasis on biases: ``measured'' velocities which differ systematically in magnitude and/or direction compared to the average current within the field of view. We generate random synthetic wave data according to a spectrum prescribed at the domain center while varying the current and its gradient, and the mean direction $\theta_0$ and angular spread of the waves' propagation direction, from which we extract Doppler shift velocities (DSVs) using one of the most common numerical schemes. Assuming currents vary by up to 10\% of a characteristic wave phase velocity across the field of view, we find that strong biases can occur for highly directional wave spectra, strongly dependent on $\theta_0$ and the current direction relative to that of its magnitude gradient. We discuss how biases arise in the DSV extraction procedure and suggest practical steps to discover the presence of biases., Comment: 9 pages, 5 figures, submitted to IEEE Transactions on Geoscience and Remote Sensing

Everyday AI detection requires differentiating between people and AI in informal, online conversations. In many cases, people will not interact directly with AI systems but instead read conversations between AI systems and other people. We measured how well people and large language models can discriminate using two modified versions of the Turing test: inverted and displaced. GPT-3.5, GPT-4, and displaced human adjudicators judged whether an agent was human or AI on the basis of a Turing test transcript. We found that both AI and displaced human judges were less accurate than interactive interrogators, with below chance accuracy overall. Moreover, all three judged the best-performing GPT-4 witness to be human more often than human witnesses. This suggests that both humans and current LLMs struggle to distinguish between the two when they are not actively interrogating the person, underscoring an urgent need for more accurate tools to detect AI in conversations.

A novel wireless transmission scheme, as named the reconfigurable intelligent surface (RIS)-assisted received adaptive spatial modulation (RASM) scheme, is proposed in this paper. In this scheme, the adaptive spatial modulation (ASM)-based antennas selection works at the receiver by employing the characteristics of the RIS in each time slot, where the signal-to-noise ratio at specific selected antennas can be further enhanced with near few powers. Besides for the bits from constellation symbols, the extra bits can be mapped into the indices of receive antenna combinations and conveyed to the receiver through the ASM-based antenna-combination selection, thus providing higher spectral efficiency. To explicitly present the RASM scheme, the analytical performance of bit error rate of it is discussed in this paper. As a trade-off selection, the proposed scheme shows higher spectral efficiency and remains the satisfactory error performance. Simulation and analytical results demonstrate the better performance and exhibit more potential to apply in practical wireless communication.

Recent empirical results have sparked a debate about whether or not Large Language Models (LLMs) are capable of Theory of Mind (ToM). While some have found LLMs to be successful on ToM evaluations such as the False Belief task (Kosinski, 2023), others have argued that LLMs solve these tasks by exploiting spurious correlations -- not representing beliefs -- since they fail on trivial alterations to these tasks (Ullman, 2023). In this paper, we introduce SCALPEL: a technique to generate targeted modifications for False Belief tasks to test different specific hypotheses about why LLMs fail. We find that modifications which make explicit common inferences -- such as that looking at a transparent object implies recognizing its contents -- preserve LLMs' performance. This suggests that LLMs' failures on modified ToM tasks could result from a lack of more general commonsense reasoning, rather than a failure to represent mental states. We argue that SCALPEL could be helpful for explaining LLM successes and failures in other cases.

We evaluated 3 systems (ELIZA, GPT-3.5 and GPT-4) in a randomized, controlled, and preregistered Turing test. Human participants had a 5 minute conversation with either a human or an AI, and judged whether or not they thought their interlocutor was human. GPT-4 was judged to be a human 54% of the time, outperforming ELIZA (22%) but lagging behind actual humans (67%). The results provide the first robust empirical demonstration that any artificial system passes an interactive 2-player Turing test. The results have implications for debates around machine intelligence and, more urgently, suggest that deception by current AI systems may go undetected. Analysis of participants' strategies and reasoning suggests that stylistic and socio-emotional factors play a larger role in passing the Turing test than traditional notions of intelligence., Comment: 23 pages, 13 figures

Transformers have generally supplanted recurrent neural networks as the dominant architecture for both natural language processing tasks and for modelling the effect of predictability on online human language comprehension. However, two recently developed recurrent model architectures, RWKV and Mamba, appear to perform natural language tasks comparably to or better than transformers of equivalent scale. In this paper, we show that contemporary recurrent models are now also able to match - and in some cases, exceed - the performance of comparably sized transformers at modeling online human language comprehension. This suggests that transformer language models are not uniquely suited to this task, and opens up new directions for debates about the extent to which architectural features of language models make them better or worse models of human language comprehension., Comment: Accepted at COLM 2024

The pursuit of higher data rates and efficient spectrum utilization in modern communication technologies necessitates novel solutions. In order to provide insights into improving spectral efficiency and reducing latency, this study investigates the maximum channel coding rate (MCCR) of finite block length (FBL) multiple-input multiple-output (MIMO) faster-than-Nyquist (FTN) channels. By optimizing power allocation, we derive the system's MCCR expression. Simulation results are compared with the existing literature to reveal the benefits of FTN in FBL transmission.

How should text dataset sizes be compared across languages? Even for content-matched (parallel) corpora, UTF-8 encoded text can require a dramatically different number of bytes for different languages. In our work, we define the byte premium between two languages as the ratio of bytes used to encode content-matched text in those languages. We compute byte premiums for 1155 languages, and we use linear regressions to estimate byte premiums for other languages. We release a tool to obtain byte premiums for any two languages, enabling comparisons of dataset sizes across languages for more equitable multilingual model development and data practices.

Understanding electronic interactions in high-temperature superconductors is an outstanding challenge. In the widely studied cuprate materials, experimental evidence points to strong electron-phonon ($e$-ph) coupling and broad photoemission spectra. Yet, the microscopic origin of this behavior is not fully understood. Here we study $e$-ph interactions and polarons in a prototypical parent (undoped) cuprate, La$_2$CuO$_4$ (LCO), by means of first-principles calculations. Leveraging parameter-free Hubbard-corrected density functional theory, we obtain a ground state with band gap and Cu magnetic moment in nearly exact agreement with experiments. This enables a quantitative characterization of $e$-ph interactions. Our calculations reveal two classes of longitudinal optical (LO) phonons with strong $e$-ph coupling to hole states. These modes consist of Cu-O plane bond-stretching and bond-bending as well as vibrations of apical O atoms. The hole spectral functions, obtained with a cumulant method that can capture strong $e$-ph coupling, exhibit broad quasiparticle peaks with a small spectral weight ($Z\approx0.25$) and pronounced LO-phonon sidebands characteristic of polaron effects. Our calculations predict features observed in photoemission spectra, including a 40-meV peak in the $e$-ph coupling distribution function not explained by existing models. These results show that the universal strong $e$-ph coupling found experimentally in lanthanum cuprates is an intrinsic feature of the parent compound, and elucidates its microscopic origin., Comment: 6 pages, 4 figures

First-principles calculations of electron interactions in materials have seen rapid progress in recent years, with electron-phonon (e-ph) interactions being a prime example. However, these techniques use large matrices encoding the interactions on dense momentum grids, which reduces computational efficiency and obscures interpretability. For e-ph interactions, existing interpolation techniques leverage locality in real space, but the high dimensionality of the data remains a bottleneck to balance cost and accuracy. Here we show an efficient way to compress e-ph interactions based on singular value decomposition (SVD), a widely used matrix / image compression technique. Leveraging (un)constrained SVD methods, we accurately predict material properties related to e-ph interactions - including charge mobility, spin relaxation times, band renormalization, and superconducting critical temperature - while using only a small fraction (1-2%) of the interaction data. These findings unveil the hidden low-dimensional nature of e-ph interactions. Furthermore, they accelerate state-of-the-art first-principles e-ph calculations by about two orders of magnitudes without sacrificing accuracy. Our Pareto-optimal parametrization of e-ph interactions can be readily generalized to electron-electron and electron-defect interactions, as well as to other couplings, advancing quantitative studies of condensed matter., Comment: 12 pages, 3 figures

Multimode squeezed states of light have been proposed as a resource for achieving quantum advantage in computing and sensing. Recent experiments that demonstrate multimode Gaussian states to this end have most commonly opted for spatial or temporal modes, whereas a complete system based on frequency modes has yet to be realized. Instead, we show how to use the frequency modes simultaneously squeezed in a conventional, single-spatial-mode, optical parametric amplifier when pumped by ultrashort pulses. Specifically, we show how adiabatic frequency conversion can be used not only to convert the quantum state from infrared to visible wavelengths, but to concurrently manipulate the joint spectrum. This near unity-efficiency quantum frequency conversion, over a bandwidth >45 THz and, to our knowledge, the broadest to date, allows us to measure the state with an electron-multiplying CCD (EMCCD) camera-based spectrometer, at non-cryogenic temperatures. We demonstrate the squeezing of >400 frequency modes, with a mean of approximately 700 visible photons per shot. Our work shows how many-mode quantum states of light can be generated, manipulated, and measured with efficient use of hardware resources -- in our case, using one pulsed laser, two nonlinear crystals, and one camera. This ability to produce, with modest hardware resources, large multimode squeezed states with partial programmability motivates the use of frequency encoding for photonics-based quantum information processing.

Young terrestrial worlds are critical test beds to constrain prevailing theories of planetary formation and evolution. We present the discovery of HD 63433d - a nearby (22pc), Earth-sized planet transiting a young sunlike star (TOI-1726, HD 63433). HD 63433d is the third planet detected in this multiplanet system. The kinematic, rotational, and abundance properties of the host star indicate that it belongs to the young (414 $\pm$ 23 Myr) Ursa Major moving group, whose membership we update using new data from Gaia DR3 and TESS. Our transit analysis of the TESS light curves indicates that HD 63433 d has a radius of 1.1 $R_\oplus$ and closely orbits its host star with a period of 4.2 days. To date, HD 63433 d is the smallest confirmed exoplanet with an age less than 500 Myr, and the nearest young Earth-sized planet. Furthermore, the apparent brightness of the stellar host (V $\approx$ 6.9 mag) makes this transiting multiplanet system favorable to further investigations, including spectroscopic follow-up to probe atmospheric loss in a young Earth-sized world., Comment: 24 pages, 7 figures, 6 tables. Accepted for publication in AJ. The first two authors contributed equally to the manuscript

Scientists are increasingly leveraging advances in instruments, automation, and collaborative tools to scale up their experiments and research goals, leading to new bursts of discovery. Various scientific disciplines, including neuroscience, have adopted key technologies to enhance collaboration, reproducibility, and automation. Drawing inspiration from advancements in the software industry, we present a roadmap to enhance the reliability and scalability of scientific operations for diverse research teams tackling large and complex projects. We introduce a five-level Capability Maturity Model describing the principles of rigorous scientific operations in projects ranging from small-scale exploratory studies to large-scale, multi-disciplinary research endeavors. Achieving higher levels of operational maturity necessitates the adoption of new, technology-enabled methodologies, which we refer to as SciOps. This concept is derived from the DevOps methodologies that have revolutionized the software industry. SciOps involves digital research environments that seamlessly integrate computational, automation, and AI-driven efforts throughout the research cycle-from experimental design and data collection to analysis and dissemination, ultimately leading to closed-loop discovery. This maturity model offers a framework for assessing and improving operational practices in multidisciplinary research teams, guiding them towards greater efficiency and effectiveness in scientific inquiry., Comment: 16 pages, two figures

Multi-node optical clock networks will enable future studies of fundamental physics and enable applications in quantum and classical communications as well as navigation and geodesy. We implement the first ever multi-node optical clock network with real-time, relative synchronization over free-space communication channels and precision on the order of 10 fs, realized as a three-node system in a hub-and-spoke topology. In this paper we describe the system and its performance, including a new, independent, out-of-loop verification of two-way optical time synchronization.

Multilingual language models are widely used to extend NLP systems to low-resource languages. However, concrete evidence for the effects of multilinguality on language modeling performance in individual languages remains scarce. Here, we pre-train over 10,000 monolingual and multilingual language models for over 250 languages, including multiple language families that are under-studied in NLP. We assess how language modeling performance in each language varies as a function of (1) monolingual dataset size, (2) added multilingual dataset size, (3) linguistic similarity of the added languages, and (4) model size (up to 45M parameters). We find that in moderation, adding multilingual data improves low-resource language modeling performance, similar to increasing low-resource dataset sizes by up to 33%. Improvements depend on the syntactic similarity of the added multilingual data, with marginal additional effects of vocabulary overlap. However, high-resource languages consistently perform worse in multilingual pre-training scenarios. As dataset sizes increase, adding multilingual data begins to hurt performance for both low-resource and high-resource languages, likely due to limited model capacity (the "curse of multilinguality"). These results suggest that massively multilingual pre-training may not be optimal for any languages involved, but that more targeted models can significantly improve performance.

Abstract grammatical knowledge - of parts of speech and grammatical patterns - is key to the capacity for linguistic generalization in humans. But how abstract is grammatical knowledge in large language models? In the human literature, compelling evidence for grammatical abstraction comes from structural priming. A sentence that shares the same grammatical structure as a preceding sentence is processed and produced more readily. Because confounds exist when using stimuli in a single language, evidence of abstraction is even more compelling from crosslingual structural priming, where use of a syntactic structure in one language primes an analogous structure in another language. We measure crosslingual structural priming in large language models, comparing model behavior to human experimental results from eight crosslingual experiments covering six languages, and four monolingual structural priming experiments in three non-English languages. We find evidence for abstract monolingual and crosslingual grammatical representations in the models that function similarly to those found in humans. These results demonstrate that grammatical representations in multilingual language models are not only similar across languages, but they can causally influence text produced in different languages., Comment: Accepted at EMNLP 2023

We present direct observations of acoustic waves in warm dense matter. We analyze wave-number- and energy-resolved x-ray spectra taken from warm dense methane created by laser heating a cryogenic liquid jet. X-ray diffraction and inelastic free-electron scattering yield sample conditions of 0.3$\pm$0.1 eV and 0.8$\pm$0.1 g/cm$^3$, corresponding to a pressure of $\sim$13 GPa. Inelastic x-ray scattering was used to observe the collective oscillations of the ions. With a highly improved energy resolution of $\sim$50 meV, we could clearly distinguish the Brillouin peaks from the quasielastic Rayleigh feature. Data at different wave numbers were utilized to derive a sound speed of 5.9$\pm$0.5 km/s, marking a high-temperature data point for methane and demonstrating consistency with Birch's law in this parameter regime., Comment: 7 pages, 4 figures

Remotely measuring subsurface water currents from imagery of the wave field has become a much-used technique. We study the biases and errors in such measurements due to spectral leakage, and suggest mitigating procedures. Deviations between peak values in the three-dimensional wave spectrum and the known dispersion relation in quiescent water are extracted and interpreted as current-induced Doppler shifts, from which the sub-surface current is inferred. The use of discrete Fourier transforms, however, introduces spectral leakage between nearby frequency bins. Analysing synthetically generated wave data adhering to realistic input spectra we show that although no current is in fact present, spurious currents can be "measured" which can amount to a significant fraction of the phase speed at the spectral peak. We analyse the effects of data tapering, method of Doppler shift extraction, limited wavenumber and frequency resolution, peakedness and angular width of the input spectrum, and average misalignment between waves and Doppler shift velocity direction. The narrower the input wave spectrum in frequency and/or direction, the greater the biases become. The use of a window function reduces the severity in nearly all cases, yet mitigates the effects of limited resolution more effectively in space than in time. When a current is present the absolute biases remain essentially unchanged, when waves and currents are roughly aligned tapering much alleviates the concomitant biases, whereas in the case of a cross-current, biases remain significant even for tapered data.

We evaluated GPT-4 in a public online Turing test. The best-performing GPT-4 prompt passed in 49.7% of games, outperforming ELIZA (22%) and GPT-3.5 (20%), but falling short of the baseline set by human participants (66%). Participants' decisions were based mainly on linguistic style (35%) and socioemotional traits (27%), supporting the idea that intelligence, narrowly conceived, is not sufficient to pass the Turing test. Participant knowledge about LLMs and number of games played positively correlated with accuracy in detecting AI, suggesting learning and practice as possible strategies to mitigate deception. Despite known limitations as a test of intelligence, we argue that the Turing test continues to be relevant as an assessment of naturalistic communication and deception. AI models with the ability to masquerade as humans could have widespread societal consequences, and we analyse the effectiveness of different strategies and criteria for judging humanlikeness., Comment: 28 pages, 21 figures

Do multilingual language models share abstract grammatical representations across languages, and if so, when do these develop? Following Sinclair et al. (2022), we use structural priming to test for abstract grammatical representations with causal effects on model outputs. We extend the approach to a Dutch-English bilingual setting, and we evaluate a Dutch-English language model during pre-training. We find that crosslingual structural priming effects emerge early after exposure to the second language, with less than 1M tokens of data in that language. We discuss implications for data contamination, low-resource transfer, and how abstract grammatical representations emerge in multilingual models., Comment: Extended abstract accepted to the 3rd Multilingual Representation Learning workshop at EMNLP 2023