Search

We show that the moduli space of degree $e$ maps from smooth genus $g \ge 1$ curves to an arbitrary low degree smooth hypersurface is singular when $e$ is large compared to $g$. We also give a lower bound for the dimension of the singular locus., Comment: 10 pages, comments welcome!

Large language models (LLMs) are susceptible to persuasion, which can pose risks when models are faced with an adversarial interlocutor. We take a first step towards defending models against persuasion while also arguing that defense against adversarial (i.e. negative) persuasion is only half of the equation: models should also be able to accept beneficial (i.e. positive) persuasion to improve their answers. We show that optimizing models for only one side results in poor performance on the other. In order to balance positive and negative persuasion, we introduce Persuasion-Balanced Training (or PBT), which leverages multi-agent recursive dialogue trees to create data and trains models via preference optimization to accept persuasion when appropriate. PBT consistently improves resistance to misinformation and resilience to being challenged while also resulting in the best overall performance on holistic data containing both positive and negative persuasion. Crucially, we show that PBT models are better teammates in multi-agent debates. We find that without PBT, pairs of stronger and weaker models have unstable performance, with the order in which the models present their answers determining whether the team obtains the stronger or weaker model's performance. PBT leads to better and more stable results and less order dependence, with the stronger model consistently pulling the weaker one up., Comment: Code: https://github.com/esteng/persuasion_balanced_training

Polarons can control carrier mobility and can also be used in the design of quantum devices. Although much effort has been directed into investigating the nature of polarons, observation of defect-related polarons is challenging due to electron-defect scattering. Here we explore the polaronic behavior of nitrogen-vacancy (NV) centers in a diamond crystal using an ultrafast pump-probe technique. A 10-fs optical pulse acts as a source of high electric field exceeding the dielectric breakdown threshold, in turn exerting a force on the NV charge distribution and polar optical phonons. The electronic and phononic responses are enhanced by an order of magnitude for a low density of NV centers, which we attribute to a combination of cooperative polaronic effects and scattering by defects. First-principles calculations support the presence of dipolar Fr\"ohlich interaction via non-zero Born effective charges. Our findings provide insights into the physics of color centers in diamonds., Comment: 20 pages, 4 figures

Language models can be used to solve long-horizon planning problems in two distinct modes: a fast 'System-1' mode, directly generating plans without any explicit search or backtracking, and a slow 'System-2' mode, planning step-by-step by explicitly searching over possible actions. While System-2 is typically more effective, it is also more computationally expensive, making it infeasible for long plans or large action spaces. Moreover, isolated System-1 or 2 ignores the user's end goals, failing to provide ways to control the model's behavior. To this end, we propose the System-1.x Planner, a controllable planning framework with LLMs that is capable of generating hybrid plans and balancing between the two planning modes based on the difficulty of the problem at hand. System-1.x consists of (i) a controller, (ii) a System-1 Planner, and (iii) a System-2 Planner. Based on a user-specified hybridization factor (x) governing the mixture between System-1 and 2, the controller decomposes a problem into sub-goals, and classifies them as easy or hard to be solved by either System-1 or 2, respectively. We fine-tune all three components on top of a single base LLM, requiring only search traces as supervision. Experiments with two diverse planning tasks -- Maze Navigation and Blocksworld -- show that our System-1.x Planner outperforms a System-1 Planner, a System-2 Planner trained to approximate A* search, and also a symbolic planner (A*). We demonstrate the following key properties of our planner: (1) controllability: increasing the hybridization factor (e.g., System-1.75 vs 1.5) performs more search, improving performance, (2) flexibility: by building a neuro-symbolic variant with a neural System-1 and a symbolic System-2, we can use existing symbolic methods, and (3) generalizability: by being able to learn from different search algorithms, our method is robust to the choice of search algorithm., Comment: 29 pages (10 tables)

Randomized smoothing is a defensive technique to achieve enhanced robustness against adversarial examples which are small input perturbations that degrade the performance of neural network models. Conventional randomized smoothing adds random noise with a fixed noise level for every input sample to smooth out adversarial perturbations. This paper proposes a new variational framework that uses a per-sample noise level suitable for each input by introducing a noise level selector. Our experimental results demonstrate enhancement of empirical robustness against adversarial attacks. We also provide and analyze the certified robustness for our sample-wise smoothing method., Comment: 20 pages, under preparation

This paper shows that retailers increase prices in response to organized retail crime. We match store-level crime data to scanner data from the universe of transactions for cannabis retailers in Washington state. Using quasi-experimental variation from robberies and burglaries, we find a 1.5-1.8% price increase at victimized stores and nearby competitors. This rise is not driven by short-to-medium-term demand changes but is consistent with an own-cost shock. Effects are larger for independent stores and less concentrated markets. We estimate that crime imposes a 1% "hidden" unit tax on affected stores, implying $33.9 million additional social costs, primarily borne by consumers.

The model editing problem concerns how language models should learn new facts about the world over time. While empirical research on model editing has drawn widespread attention, the conceptual foundations of model editing remain shaky -- perhaps unsurprisingly, since model editing is essentially belief revision, a storied problem in philosophy that has eluded succinct solutions for decades. Model editing nonetheless demands a solution, since we need to be able to control the knowledge within language models. With this goal in mind, this paper critiques the standard formulation of the model editing problem and proposes a formal testbed for model editing research. We first describe 12 open problems with model editing, based on challenges with (1) defining the problem, (2) developing benchmarks, and (3) assuming LLMs have editable beliefs in the first place. Many of these challenges are extremely difficult to address, e.g. determining far-reaching consequences of edits, labeling probabilistic entailments between facts, and updating beliefs of agent simulators. Next, we introduce a semi-synthetic dataset for model editing based on Wikidata, where we can evaluate edits against labels given by an idealized Bayesian agent. This enables us to say exactly how belief revision in language models falls short of a desirable epistemic standard. We encourage further research exploring settings where such a gold standard can be compared against. Our code is publicly available at: https://github.com/peterbhase/LLM-belief-revision, Comment: 23 pages, 4 figures

Do norms of rationality apply to machine learning models, in particular language models? In this paper we investigate this question by focusing on a special subset of rational norms: coherence norms. We consider both logical coherence norms as well as coherence norms tied to the strength of belief. To make sense of the latter, we introduce the Minimal Assent Connection (MAC) and propose a new account of credence, which captures the strength of belief in language models. This proposal uniformly assigns strength of belief simply on the basis of model internal next token probabilities. We argue that rational norms tied to coherence do apply to some language models, but not to others. This issue is significant since rationality is closely tied to predicting and explaining behavior, and thus it is connected to considerations about AI safety and alignment, as well as understanding model behavior more generally., Comment: added discussion and cross reference of new empirical work by the authors, updated references, fixed typos

When answering questions, LLMs can convey not only an answer, but a level of confidence about the answer being correct. This includes explicit confidence markers (e.g. giving a numeric score) as well as implicit markers, like an authoritative tone or elaborating with additional knowledge. For LLMs to be trustworthy knowledge sources, the confidence they convey should match their actual expertise; however, most current models tend towards overconfidence. To calibrate both implicit and explicit confidence markers, we introduce a pragmatic, listener-aware finetuning method (LACIE) that models the listener, considering not only whether an answer is right, but whether it will be accepted by a listener. We cast calibration as preference optimization, creating data via a two-agent game, where a speaker model's outputs are judged by a simulated listener. We then finetune three LLMs (Mistral-7B, Llama3-8B, Llama3-70B) with LACIE, and show that the resulting models are better calibrated w.r.t. a simulated listener. Crucially, these trends transfer to human listeners, helping them correctly predict model correctness: we conduct a human evaluation where annotators accept or reject an LLM's answers, finding that training with LACIE results in 47% fewer incorrect answers being accepted while maintaining the same level of acceptance for correct answers. Furthermore, LACIE generalizes to another dataset, resulting in a large increase in truthfulness on TruthfulQA when trained on TriviaQA. Our analysis indicates that LACIE leads to a better confidence separation between correct and incorrect examples. Qualitatively, we find that a LACIE-trained model hedges more and implicitly signals certainty when it is correct by using an authoritative tone or including details. Finally, LACIE finetuning leads to an emergent increase in model abstention (e.g. saying "I don't know") for answers that are likely wrong., Comment: 18 pages. Code: https://github.com/esteng/pragmatic_calibration

Sawin recently gave an axiomatic characterization of multiple Dirichlet series over the function field $\mathbb{F}_{q}(T)$ and proved their existence by exhibiting the coefficients as trace functions of specific perverse sheaves. However, he did not prove that these series actually converge anywhere, instead treating them as formal power series. In this paper, we prove that these series do converge in a certain region, and moreover that the functions obtained by analytically continuing them satisfy functional equations. For convergence, it suffices to obtain bounds on the coefficients, for which we use the decomposition theorem for perverse sheaves, in combination with the Kontsevich moduli space of stable maps to construct a suitable compactification. For the functional equations, the key identity is a multi-variable generalization of the relationship between a Dirichlet character and its Fourier transform; in the multiple Dirichlet series setting, this uses a density trick for simple perverse sheaves and an explicit formula for intermediate extensions from the complement of a normal crossings divisor., Comment: 43 pages; fixed a few typos

We introduce a novel laser-scanning optical microscopy technique that employs optical-frequency-comb (OFC) lasers. This method facilitates multimodal spectroscopic imaging by analyzing interferograms produced via a dual-comb spectroscopic approach. Such interferograms capture comprehensive light information, including amplitude, phase, polarization, frequency, and time of flight information, enabling multimodal imaging from a single measurement. We demonstrate the potential of this technique across several spectroscopic imaging applications., Comment: 11 pages, 8 figures

We investigate the ultrafast spin dynamics of the prototypical topological insulator $\mathrm{Bi_{2}Se_{3}}$ using time-resolved Kerr-rotation (polarization-change) measurements across near-infrared wavelengths. The Kerr-rotation angle $\Delta \theta_{K}$ of $\mathrm{Bi_{2}Se_{3}}$ was found to significantly depend on photon energy around a resonance transition ($\sim 1.0\ \mathrm{eV}$) of bulk states, as well as the ellipticity of the pump light, in the presence of spin excitation. The observed photon-energy dependence of $\Delta \theta_{K}$ can be well simulated by assuming spin-dependent refractive-index changes in the presence of band-gap renormalization and state-filling effect upon photoexcitation. Our study delivers comprehensive insights into the opto-spintronic properties of bulk $\mathrm{Bi_{2}Se_{3}}$ and the fundamental physical processes underlying polarization changes. These findings are expected to be crucial in developing ultrafast magneto-optical memory devices, which can perform read-and-write operations in the Terahertz regime., Comment: 15 pages, 3 figures, 1 table

Refractive-index (RI) sensing plays a pivotal role in various domains, encompassing applications like glucose sensing, biosensing, and gas detection. Despite the advantages of optical fiber sensors, such as their compact size, flexibility, and immunity to electromagnetic interference, they are often plagued by temperature-induced drift, which adversely impacts the accuracy of RI measurements. This study introduces an innovative approach to alleviate temperature-induced drift in RI-sensing optical frequency combs (OFCs) by employing active-dummy compensation. The central idea revolves around the utilization of a dual-comb setup, comprising an active-sensing OFC that monitors both sample RI and environmental temperature, and a dummy-sensing OFC that exclusively tracks environmental temperature. The disparity between these sensor signals, denoted as delta_frep, effectively nullifies the effects of temperature variations, yielding a temperature-independent sensor signal for precise RI measurements. This investigation delves into the relationship between active-dummy temperature compensation and delta_frep. It becomes evident that diminishing delta_frep values enhance temperature compensation, thereby diminishing fluctuations in delta_frep caused by environmental temperature shifts. This compensation technique establishes a direct link between delta_frep and sample RI, paving the way for absolute RI measurements based on delta_frep. The findings of this research are a valuable contribution to the advancement of accurate and temperature-compensated RI sensing methodologies using dual-comb setup. The insights gained regarding delta_frep dependency and the strategies proposed for enhancing measurement precision and stability hold significant promise for applications in fields of product quality control and biosensing., Comment: 25 pages, 9 figures

This work identifies 18 foundational challenges in assuring the alignment and safety of large language models (LLMs). These challenges are organized into three different categories: scientific understanding of LLMs, development and deployment methods, and sociotechnical challenges. Based on the identified challenges, we pose $200+$ concrete research questions.

Pyrrhotite, Fe$_7$S$_8$, provides an example of exceptionally strong magnetoelastic coupling through pinning of ferromagnetic domains by ferroelastic twins. Using direct imaging of both magnetic and ferroelastic domains by magnetic force microscopy (MFM), the mechanism by which this coupling controls local magnetic switching behaviour of regions on the pyrrhotite surface is revealed, and leads to quantitative fitting of field dependent MFM phase shifts with bulk magnetometry data. It is shown that characteristic inflection points in the magnetometry data along certain direction, in particular $[\overline 120]^*_h$ of the hexagonal parent structure, are in fact caused by ferroelastic pinning of the magnetic moments.

Browning and Vishe used the Hardy-Littlewood circle method to show the moduli space of rational curves on smooth hypersurfaces of low degree is irreducible and of the expected dimension. We reinterpret the circle method geometrically and prove a generalization for higher genus smooth projective curves. In particular, we explain how the geometry of numbers can be understood via the Beauville-Laszlo theorem in terms of vector bundles on curves and their slopes, allowing us to prove a higher genus variant of Davenport's shrinking lemma. As a corollary, we apply this result to show the Fujita invariant of any proper subvariety of a smooth hypersurface of low degree is less than 1., Comment: 36 pages

We explore machine unlearning (MU) in the domain of large language models (LLMs), referred to as LLM unlearning. This initiative aims to eliminate undesirable data influence (e.g., sensitive or illegal information) and the associated model capabilities, while maintaining the integrity of essential knowledge generation and not affecting causally unrelated information. We envision LLM unlearning becoming a pivotal element in the life-cycle management of LLMs, potentially standing as an essential foundation for developing generative AI that is not only safe, secure, and trustworthy, but also resource-efficient without the need of full retraining. We navigate the unlearning landscape in LLMs from conceptual formulation, methodologies, metrics, and applications. In particular, we highlight the often-overlooked aspects of existing LLM unlearning research, e.g., unlearning scope, data-model interaction, and multifaceted efficacy assessment. We also draw connections between LLM unlearning and related areas such as model editing, influence functions, model explanation, adversarial training, and reinforcement learning. Furthermore, we outline an effective assessment framework for LLM unlearning and explore its applications in copyright and privacy safeguards and sociotechnical harm reduction.

How can we train models to perform well on hard test data when hard training data is by definition difficult to label correctly? This question has been termed the scalable oversight problem and has drawn increasing attention as language models have continually improved. In this paper, we present the surprising conclusion that current pretrained language models often generalize relatively well from easy to hard data, even performing as well as oracle models finetuned on hard data. We demonstrate this kind of easy-to-hard generalization using simple finetuning methods like in-context learning, linear classifier heads, and QLoRA for seven different measures of datapoint hardness, including six empirically diverse human hardness measures (like grade level) and one model-based measure (loss-based). Furthermore, we show that even if one cares most about model performance on hard data, it can be better to collect easy data rather than hard data for finetuning, since hard data is generally noisier and costlier to collect. Our experiments use open models up to 70b in size and four publicly available question-answering datasets with questions ranging in difficulty from 3rd grade science questions to college level STEM questions and general-knowledge trivia. We conclude that easy-to-hard generalization in LMs is surprisingly strong for the tasks studied. Our code is available at: https://github.com/allenai/easy-to-hard-generalization, Comment: ACL 2024. 23 pages, 20 figures

Organic semiconductors (OSCs) are suitable materials for spintronics applications as they form a spinterface when placed next to a ferromagnet, which in turn leads to novel functionalities. The evolution of spinterface can tune the global magnetic anisotropy, magnetization reversal, magnetization dynamics, etc. Planar tris-(8-hydroxyquinoline)aluminum (Alq3) OSC has shown tremendous potential for spintronics applications, thanks to its efficient spin-polarized current transport ability. Here, we establish the spinterface when the Alq3 molecules are deposited on amorphous ferromagnet Co20Fe60B20(CFB). The $\pi$-d hybridization in CFB/Alq3 enhances the coercive field and significantly modifies the shape and size of the magnetic domains. A $\sim$100% increase in uniaxial anisotropic energies and a reduction in magnetic damping are also evident owing to the strong interfacial hybridization.

As artificial intelligence (AI) is integrated into various services and systems in society, many companies and organizations have proposed AI principles, policies, and made the related commitments. Conversely, some have proposed the need for independent audits, arguing that the voluntary principles adopted by the developers and providers of AI services and systems insufficiently address risk. This policy recommendation summarizes the issues related to the auditing of AI services and systems and presents three recommendations for promoting AI auditing that contribute to sound AI governance. Recommendation1.Development of institutional design for AI audits. Recommendation2.Training human resources for AI audits. Recommendation3. Updating AI audits in accordance with technological progress. In this policy recommendation, AI is assumed to be that which recognizes and predicts data with the last chapter outlining how generative AI should be audited.

The Next Generation Very Large Array (ngVLA) is a planned radio interferometer providing unprecedented sensitivity at wavelengths between 21 cm and 3 mm. Its 263 antenna element array will be spatially distributed across North America to enable both superb low surface brightness recovery and sub-milliarcsecond angular resolution imaging. The project was developed by the international astronomy community under the lead of the National Radio Astronomy Observatory (NRAO), and is anticipated to be built between 2027 and 2037. Two workshops have been held in 2022 and 2023 with the goal to discuss and consolidate the scientific interests in the ngVLA within the German astronomical community. This community paper constitutes a collection of 48 science ideas which the German community aims to pursue with the ngVLA in the 2030s. This is not a complete list and the ideas are not developed at the level of a "Science Book", such that the present document is mainly meant provide a basis for further discussion within the community. As such, additional contributions are welcome, and will be considered for inclusion in future revisions., Comment: Version 2.0 (status June 18, 2024): 169 pages, comments and future contributions welcome [v2.0: 7 new science cases added, some minor revisions to other chapters]

Magnetite has long been investigated across many disciplines due to the interplay between its ferroic order parameters, namely its ferrimagnetism, ferroelasticity and ferroelectricty. Despite this, the experimental difficulty in measuring low temperature real space images of the ferroelectric domains has meant that the local behaviour of ferroelectric domains emergent below the 38 K phase transition have yet to be realised. This work presents real space images of the ferroelectric domains, and uses piezo force microscopy to, as a function of temperature, probe the onset of piezoelectricty and ferroelectricity across the 38 K transition, Comment: 6 Pages, 3 figures

The Committee on Radio Astronomy Frequencies (CRAF) is an Expert Committee of the European Science Foundation. It aims to provide a cost-effective single voice on frequency protection issues for European radio astronomy observatories and research institutes, achieving a significantly greater impact than that achievable by individual national institutions. By working together, European observatories and institutes can profit from synergy effects, cover many more topics, and learn from each other. CRAF was founded in 1988 and has since then been engaged with the International Telecommunication Union (ITU), in particular its Radiocommunication Sector (ITU-R), and the European Conference of Postal and Telecommunications Administrations (CEPT) and its European Communications Committee (ECC). This is the self-evaluation report prepared by CRAF for its periodic review of the years 2011-2021., Comment: 75 pages