Search

The development and evaluation of Large Language Models (LLMs) have largely focused on individual capabilities. However, this overlooks the intersection of multiple abilities across different types of expertise that are often required for real-world tasks, which we term cross capabilities. To systematically explore this concept, we first define seven core individual capabilities and then pair them to form seven common cross capabilities, each supported by a manually constructed taxonomy. Building on these definitions, we introduce CrossEval, a benchmark comprising 1,400 human-annotated prompts, with 100 prompts for each individual and cross capability. To ensure reliable evaluation, we involve expert annotators to assess 4,200 model responses, gathering 8,400 human ratings with detailed explanations to serve as reference examples. Our findings reveal that, in both static evaluations and attempts to enhance specific abilities, current LLMs consistently exhibit the "Law of the Weakest Link," where cross-capability performance is significantly constrained by the weakest component. Specifically, across 58 cross-capability scores from 17 models, 38 scores are lower than all individual capabilities, while 20 fall between strong and weak, but closer to the weaker ability. These results highlight the under-performance of LLMs in cross-capability tasks, making the identification and improvement of the weakest capabilities a critical priority for future research to optimize performance in complex, multi-dimensional scenarios., Comment: Data, Code, & Benchmark: www.llm-cross-capabilities.org

Since multiple MRI contrasts of the same anatomy contain redundant information, one contrast can be used as a prior for guiding the reconstruction of an undersampled subsequent contrast. To this end, several learning-based guided reconstruction methods have been proposed. However, two key challenges remain - (a) the requirement of large paired training datasets and (b) the lack of intuitive understanding of the model's internal representation and utilization of the shared information. We propose a modular two-stage approach for guided reconstruction, addressing these challenges. A content/style model of two-contrast image data is learned in a largely unpaired manner and is subsequently applied as a plug-and-play operator in iterative reconstruction. The disentanglement of content and style allows explicit representation of contrast-independent and contrast-specific factors. Based on this, incorporating prior information into the reconstruction reduces to simply replacing the aliased reconstruction content with clean content derived from the reference scan. We name this novel approach PnP-MUNIT. Various aspects like interpretability and convergence are explored via simulations. Furthermore, its practicality is demonstrated on the NYU fastMRI DICOM dataset and two in-house raw datasets, obtaining up to 32.6% more acceleration over learning-based non-guided reconstruction for a given SSIM. In a radiological task, PnP-MUNIT allowed 33.3% more acceleration over clinical reconstruction at diagnostic quality., Comment: This work has been submitted to the IEEE for possible publication

The Local Friendliness argument is an extended Wigner's friend no-go theorem that provides strong constraints on the nature of reality -- stronger even than those imposed by Bell's theorem or by noncontextuality arguments. In this work, we prove a variety of connections between Local Friendliness scenarios and Kochen-Specker noncontextuality. Specifically, we first show how one can derive new Local Friendliness inequalities using known tools and results from the literature on Kochen-Specker noncontextuality. In doing so, we provide a new derivation for some of the facets of the Local Friendliness polytope, and we prove that this polytope is equal to the Bell polytope in a wide range of extended Wigner's friend scenarios with multipartite agents and sequential measurements. We then show how any possibilistic Kochen-Specker argument can be mathematically translated into a related proof of the Local Friendliness no-go theorem. In particular, we construct a novel kind of Local Friendliness scenario where a friend implements several compatible measurements (or joint measurements of these) in between the superobserver's operations on them. We illustrate this with the well-known 5-cycle and Peres-Mermin contextuality arguments., Comment: 22 pages + appendix, 16 figures, all comments welcome! The second half of this work is a follow-up work to arXiv:2310.06976v2

We present DemoStart, a novel auto-curriculum reinforcement learning method capable of learning complex manipulation behaviors on an arm equipped with a three-fingered robotic hand, from only a sparse reward and a handful of demonstrations in simulation. Learning from simulation drastically reduces the development cycle of behavior generation, and domain randomization techniques are leveraged to achieve successful zero-shot sim-to-real transfer. Transferred policies are learned directly from raw pixels from multiple cameras and robot proprioception. Our approach outperforms policies learned from demonstrations on the real robot and requires 100 times fewer demonstrations, collected in simulation. More details and videos in https://sites.google.com/view/demostart., Comment: 15 pages total with 7 pages of appendix. 9 Figures, 4 in the main text and 5 in the appendix

The Frauchiger--Renner paradox derives an inconsistency when quantum theory is used to describe the use of itself, by means of a scenario where agents model other agents quantumly and reason about each other's knowledge. We observe that logical contextuality (\`a la Hardy) is the key ingredient of the FR paradox, and we provide a stronger paradox based on the strongly contextual GHZ--Mermin scenario. In contrast to the FR paradox, this GHZ--FR paradox neither requires post-selection nor any reasoning by observers who are modelled quantumly. If one accepts the universality of quantum theory including superobservers, we propose a natural extension of Peres's dictum to resolve these extended Wigner's friend paradoxes., Comment: All comments welcome!

Open microcavities offer great potential for the exploration and utilization of efficient spin-photon interfaces with Purcell-enhanced quantum emitters thanks to their large spectral and spatial tunability combined with high versatility of sample integration. However, a major challenge for this platform is the sensitivity to cavity length fluctuations in the cryogenic environment, which leads to cavity resonance frequency variations and thereby a lowered averaged Purcell enhancement. This work presents a closed-cycle cryogenic fiber-based microcavity setup, which is in particular designed for a low passive vibration level, while still providing large tunability and flexibility in fiber and sample integration, and high photon collection efficiency from the cavity mode. At temperatures below 10 Kelvin, a stability level of around 25 picometer is reproducibly achieved in different setup configurations, including the extension with microwave control for manipulating the spin of cavity-coupled quantum emitters, enabling a bright photonic interface with optically active qubits., Comment: Authors Yanik Herrmann and Julius Fischer contributed equally, 14 pages, 9 figures

We investigate the nature of the quantum phase transition in modulated SU(N) Heisenberg spin chains. In the odd-N case, the transition separates a trivial non-degenerate phase to a doubly-degenerate gapped chiral PSU(N) symmetry-protected topological (SPT) phase which breaks spontaneously the inversion symmetry. The transition is not an Ising transition associated to the breaking of the $\mathbb{Z}_2$ inversion symmetry, but is governed by the delocalization of the edge states of the SPT phase. In this respect, a modulated SU(N) Heisenberg spin chain provides a simple example in one dimension of a non-Landau phase transition which is described by the SU(N)$_1$ conformal field theory. We show that the chiral SPT phase exhibits fractionalized spinon excitations, which can be confined by changing the model parameters slightly., Comment: 13 pages, 12 figures

The Sleeping Beauty problem is a puzzle in probability theory that has gained much attention since Elga's discussion of it [Elga, Adam, Analysis 60 (2), p.143-147 (2000)]. Sleeping Beauty is put asleep, and a coin is tossed. If the outcome of the coin toss is Tails, Sleeping Beauty is woken up on Monday, put asleep again and woken up again on Tuesday (with no recollection of having woken up on Monday). If the outcome is Heads, Sleeping Beauty is woken up on Monday only. Each time Sleeping Beauty is woken up, she is asked what her belief is that the outcome was Heads. What should Sleeping Beauty reply? In literature arguments have been given for both 1/3 and 1/2 as the correct answer. In this short note we argue using simple Bayesian probability theory why 1/3 is the right answer, and not 1/2. Briefly, when Sleeping Beauty awakens, her being awake is nontrivial extra information that leads her to update her beliefs about Heads to 1/3. We strengthen our claim by considering an additional observer, Prince Probability, who may or may not meet Sleeping Beauty. If he meets Sleeping Beauty while she is awake, he lowers his credence in Heads to 1/3. We also briefly consider the credence in Heads of a Sleeping Beauty who knows that she is dreaming (and thus asleep)., Comment: 12 pages, 1 figure, all comments welcome!

We leverage the interplay between gapped phases and dualities of symmetric one-dimensional quantum lattice models to demonstrate that every phase is efficiently characterised by the maximal breaking of the dual (genereralised) symmetry whose structure encodes the quasiparticle excitations. This result has strong implications for the complexity of simulating many-body systems using variational tensor network methods. For every phase in the phase diagram, the dual representation of the ground state that breaks all symmetries minimises both the entanglement entropy and the required number of variational parameters. We demonstrate the applicability of this idea by developing a generalised density matrix renormalisation group algorithm that works on (dual) constrained Hilbert spaces, and quantify the computational gains obtained over traditional DMRG methods in a perturbed Heisenberg model. Our work testifies to the usefulness of generalised non-invertible symmetries and their formal category theoretic description for the nuts and bolts simulation of strongly correlated systems., Comment: 7+6 pages

The fractional AC Josephson effect is a discerning property of topological superconductivity in hybrid Josephson junctions. Recent experimental observations of missing odd Shapiro steps and half Josephson frequency emission in various materials have sparked significant debate regarding their potential origin in the effect. In this study, we present microwave emission measurements on a resistively shunted Josephson junction based on a HgTe quantum well. We demonstrate that, with significant spurious inductance in the shunt wiring, the experiment operates in a nonlinear dynamic regime characterized by period-doubling. This leads to additional microwave emission peaks at half of the Josephson frequency, $f_J/2$, which can mimic the $4\pi$-periodicity of topological Andreev states. The observed current-voltage characteristics and emission spectra are well-described by a simple RCLSJ model. Furthermore, we show that the nonlinear dynamics of the junction can be controlled using gate voltage, magnetic field, and temperature, with our model accurately reproducing these effects without incorporating any topological attributes. Our observations urge caution in interpreting emission at $f_J/2$ as evidence for gapless Andreev bound states in topological junctions and suggest the appropriate parameter range for future experiments., Comment: 13 pages, 8 figures

Modern artificial intelligence (AI) systems are powered by foundation models. This paper presents a new set of foundation models, called Llama 3. It is a herd of language models that natively support multilinguality, coding, reasoning, and tool usage. Our largest model is a dense Transformer with 405B parameters and a context window of up to 128K tokens. This paper presents an extensive empirical evaluation of Llama 3. We find that Llama 3 delivers comparable quality to leading language models such as GPT-4 on a plethora of tasks. We publicly release Llama 3, including pre-trained and post-trained versions of the 405B parameter language model and our Llama Guard 3 model for input and output safety. The paper also presents the results of experiments in which we integrate image, video, and speech capabilities into Llama 3 via a compositional approach. We observe this approach performs competitively with the state-of-the-art on image, video, and speech recognition tasks. The resulting models are not yet being broadly released as they are still under development.

Immersing a mobile impurity in a quantum many-body environment can reveal fundamental properties of the background medium, hence providing a powerful probe of quantum matter. This approach is particularly intriguing when considering media with exotic properties, such as strongly-correlated phases and topological states of matter. In this work, we study the dressing of a mobile impurity interacting with a chiral mode, as provided by the edge of topological quantum matter. The resulting ''chiral polaron'' is characterized by an asymmetric spectral function, which reflects the chirality and group velocity of the topological edge mode and the drag experienced by the mobile impurity. We first build our theoretical understanding from an effective one-dimensional chiral model, which captures the hallmark signatures of the chiral polaron. We then demonstrate how this simple picture extends to realistic models of integer and fractional Chern insulator states, by adapting tensor-network methods to polaron spectroscopy. Injecting mobile impurities on the edge of topological quantum matter is shown to be a powerful tool to probe exotic edge properties, particularly suitable for cold-atom experiments., Comment: 28 pages, 7 Sections and 5 Appendices

Non-adiabatic coupling matrix elements (NACMEs) are important in quantum chemistry, particularly for molecular dynamics methods such as surface hopping. However, NACMEs are gauge dependent. This presents a difficulty for their calculation in general, where there are no restrictions on the gauge function except that it be differentiable. These cases are relevant for complex-valued electronic wave functions, such as those that arise in the presence of a magnetic field or spin-orbit coupling. Additionally, the Berry curvature and Berry force play an important role in molecular dynamics in a magnetic field, and are also relevant in the context of spin-orbit coupling. For methods such as surface hopping, excited-state Berry curvatures will also be of interest. With this in mind, we have developed a scheme for the calculation of continuous, differentiable NACMEs as a function of the molecular geometry for complex-valued wave functions. We demonstrate the efficacy of the method by using the H$_2$ molecule at the full configuration-interaction (FCI) level of theory. Additionally, ground- and excited- state Berry curvatures are computed for the first time using FCI theory. Finally, Berry phases are computed directly in terms of diagonal NACMEs.

The software project kdotpy provides a Python application for simulating electronic band structures of semiconductor devices with $\mathbf{k}\cdot\mathbf{p}$ theory on a lattice. The application implements the widely used Kane model, capable of reliable predictions of transport and optical properties for a large variety of topological and non-topological materials with a zincblende crystal structure. The application automates the tedious steps of simulating band structures. The user inputs the relevant physical parameters on the command line, for example materials and dimensions of the device, magnetic field, and temperature. The program constructs the appropriate matrix Hamiltonian on a discretized lattice of spatial coordinates and diagonalizes it. The physical observables are extracted from the eigenvalues and eigenvectors and saved as output. The program is highly customizable with a large set of configuration options and material parameters., Comment: 143 pages, submission to SciPost Physics Codebases

2D topological insulators promise novel approaches towards electronic, spintronic, and quantum device applications. This is owing to unique features of their electronic band structure, in which bulk-boundary correspondences enforces the existence of 1D spin-momentum locked metallic edge states - both helical and chiral - surrounding an electrically insulating bulk. Forty years since the first discoveries of topological phases in condensed matter, the abstract concept of band topology has sprung into realization with several materials now available in which sizable bulk energy gaps - up to a few hundred meV - promise to enable topology for applications even at room-temperature. Further, the possibility of combining 2D TIs in heterostructures with functional materials such as multiferroics, ferromagnets, and superconductors, vastly extends the range of applicability beyond their intrinsic properties. While 2D TIs remain a unique testbed for questions of fundamental condensed matter physics, proposals seek to control the topologically protected bulk or boundary states electrically, or even induce topological phase transitions to engender switching functionality. Induction of superconducting pairing in 2D TIs strives to realize non-Abelian quasiparticles, promising avenues towards fault-tolerant topological quantum computing. This roadmap aims to present a status update of the field, reviewing recent advances and remaining challenges in theoretical understanding, materials synthesis, physical characterization and, ultimately, device perspectives.

Novel imaging and neurostimulation techniques open doors for advancements in closed-loop control of activity in biological neural networks. This would allow for applications in the investigation of activity propagation, and for diagnosis and treatment of pathological behaviour. Due to the partially observable characteristics of activity propagation, through networks in which edges can not be observed, and the dynamic nature of neuronal systems, there is a need for adaptive, generalisable control. In this paper, we introduce an environment that procedurally generates neuronal networks with different topologies to investigate this generalisation problem. Additionally, an existing transformer-based architecture is adjusted to evaluate the generalisation performance of a deep RL agent in the presented partially observable environment. The agent demonstrates the capability to generalise control from a limited number of training networks to unseen test networks.

We construct a convergent family of outer approximations for the problem of optimizing polynomial functions over convex bodies subject to polynomial constraints. This is achieved by generalizing the polarization hierarchy, which has previously been introduced for the study of polynomial optimization problems over state spaces of $C^*$-algebras, to convex cones in finite dimensions. If the convex bodies can be characterized by linear or semidefinite programs, then the same is true for our hierarchy. Convergence is proven by relating the problem to a certain de Finetti theorem for general probabilistic theories, which are studied as possible generalizations of quantum mechanics. We apply the method to the problem of nonnegative matrix factorization, and in particular to the nested rectangles problem. A numerical implementation of the third level of the hierarchy is shown to give rise to a very tight approximation for this problem., Comment: 11 pages, 1 figure

This paper presents the development of a lexicon centered on emerging concepts, focusing on non-technological innovation. It introduces a four-step methodology that combines human expertise, statistical analysis, and machine learning techniques to establish a model that can be generalized across multiple domains. This process includes the creation of a thematic corpus, the development of a Gold Standard Lexicon, annotation and preparation of a training corpus, and finally, the implementation of learning models to identify new terms. The results demonstrate the robustness and relevance of our approach, highlighting its adaptability to various contexts and its contribution to lexical research. The developed methodology promises applicability in conceptual fields., Comment: in French language. JADT 2024

This paper explores the use of XGBoost for composite quantile regression. XGBoost is a highly popular model renowned for its flexibility, efficiency, and capability to deal with missing data. The optimization uses a second order approximation of the loss function, complicating the use of loss functions with a zero or vanishing second derivative. Quantile regression -- a popular approach to obtain conditional quantiles when point estimates alone are insufficient -- unfortunately uses such a loss function, the pinball loss. Existing workarounds are typically inefficient and can result in severe quantile crossings. In this paper, we present a smooth approximation of the pinball loss, the arctan pinball loss, that is tailored to the needs of XGBoost. Specifically, contrary to other smooth approximations, the arctan pinball loss has a relatively large second derivative, which makes it more suitable to use in the second order approximation. Using this loss function enables the simultaneous prediction of multiple quantiles, which is more efficient and results in far fewer quantile crossings., Comment: 24 pages, 9 figures

High-resolution transmission electron microscopy and superconducting quantum interference device magnetometry shows that epitaxial CuMnSb films exhibit a coexistence of two magnetic phases, coherently intertwined in nanometric scales. The dominant $\alpha$~phase is half-Heusler cubic antiferromagnet with the N\'{e}el temperature of 62~K, the equilibrium structure of bulk CuMnSb. The secondary phase is its ferromagnetic tetragonal $\beta$ polymorph with the Curie temperature of about 100~K. First principles calculations provide a consistent interpretation of experiment, since (i) total energy of $\beta$--CuMnSb is higher than that of $\alpha$--CuMnSb only by 0.12~eV per formula unit, which allows for epitaxial stabilization of this phase, (ii) the metallic character of $\beta$--CuMnSb favors the Ruderman-Kittel-Kasuya-Yoshida ferromagnetic coupling, and (iii) the calculated effective Curie-Weiss magnetic moment of Mn ions in both phases is about $5.5~\mu_\mathrm{B}$, favorably close to the measured value. Calculated properties of all point native defects indicate that the most likely to occur are $\mathrm{Mn}_\mathrm{Cu}$ antisites. They affect magnetic properties of epilayers, but they cannot induce the ferromagnetic order in CuMnSb. Combined, the findings highlight a practical route towards fabrication of functional materials in which coexisting polymorphs provide complementing functionalities in one host., Comment: 14 pages, 8 figures, 5 tables

This paper aims to advance our understanding of how Visual Language Models (VLMs) handle privacy-sensitive information, a crucial concern as these technologies become integral to everyday life. To this end, we introduce a new benchmark PrivBench, which contains images from 8 sensitive categories such as passports, or fingerprints. We evaluate 10 state-of-the-art VLMs on this benchmark and observe a generally limited understanding of privacy, highlighting a significant area for model improvement. Based on this we introduce PrivTune, a new instruction-tuning dataset aimed at equipping VLMs with knowledge about visual privacy. By tuning two pretrained VLMs, TinyLLaVa and MiniGPT-v2, on this small dataset, we achieve strong gains in their ability to recognize sensitive content, outperforming even GPT4-V. At the same time, we show that privacy-tuning only minimally affects the VLMs performance on standard benchmarks such as VQA. Overall, this paper lays out a crucial challenge for making VLMs effective in handling real-world data safely and provides a simple recipe that takes the first step towards building privacy-aware VLMs., Comment: preprint

While there exist theories that have states "more strongly entangled" than quantum theory, in the sense that they show CHSH values above Tsirelson's bound, all known examples of such theories have a strictly smaller set of measurements. Therefore, in tasks which require both bipartite states and measurements, they do not perform better than QM. One of the simplest information processing tasks involving both bipartite states and measurements is that of entanglement swapping. In this paper, we study entanglement swapping in generalised probabilistic theories (GPTs). In particular, we introduce the iterated CHSH game, which measures the power of a GPT to preserve non-classical correlations, in terms of the largest CHSH value obtainable after $n$ rounds of entanglement swapping. Our main result is the construction of a GPT that achieves a CHSH value of $4$ after an arbitrary number of rounds. This addresses a question about the optimality of quantum theory for such games recently raised by Weilenmann and Colbeck. One challenge faced when treating this problem is that there seems to be no general framework for constructing GPTs in which entanglement swapping is a well-defined operation. Therefore, we introduce an algorithmic construction that turns a bipartite GPT into a multipartite GPT that supports entanglement swapping, if consistently possible., Comment: 11 pages, 12 figures, v2: added references and contact information, v3: spurious 'test' removed from title (oof), v4: published version

We provide an evaluation of an analytical workload in a confidential computing environment, combining DuckDB with two technologies: modular columnar encryption in Parquet files (data at rest) and the newest version of the Intel SGX Trusted Execution Environment (TEE), providing a hardware enclave where data in flight can be (more) securely decrypted and processed. One finding is that the "performance tax" for such confidential analytical processing is acceptable compared to not using these technologies. We eventually manage to run TPC-H SF30 with under 2x overhead compared to non-encrypted, non-enclave execution; we show that, specifically, columnar compression and encryption are a good combination. Our second finding consists of dos and don'ts to tune DuckDB to work effectively in this environment. There are various performance hazards: potentially 5x higher cache miss costs due to memory encryption inside the enclave, NUMA penalties, and highly elevated cost of swapping pages in and out of the enclave -- which is also triggered indirectly by using a non-SGX-aware malloc library.

We develop a Hamiltonian formalism for simulating interacting chiral fermions on the lattice while preserving unitarity and locality and without breaking the chiral symmetry. The fermion doubling problem is circumvented by constructing a Fock space endowed with a semi-definite norm. When projecting our theory on the the single-particle sector, we recover the framework of Stacey fermions, and we demonstrate that the scaling limit of the free model recovers the chiral fermion field. Technically, we make use of a matrix product operator norm to mimick the boundary of a higher dimensional topological theory. As a proof of principle, we consider a single Weyl fermion on a periodic ring with Hubbard-type nearest-neighbor interactions and construct a variational generalized DMRG code to demonstrate that the ground state for large system sizes can be determined efficiently. As our tensor network approach does not exhibit any sign problem, we can add a chemical potential and study real-time evolution.

We show how fermionic statistics can be naturally incorporated in tensor networks on arbitrary graphs through the use of graded Hilbert spaces. This formalism allows to use tensor network methods for fermionic lattice systems in a local way, avoiding the need of a Jordan-Wigner transformation or the explicit tracking of leg crossings by swap gates in 2D tensor networks. The graded Hilbert spaces can be readily integrated with other internal and lattice symmetries in tensor networks, and only require minor extensions to an existing tensor network software package. We review and benchmark the fermionic versions of common algorithms for matrix product states and projected entangled-pair states., Comment: 73 pages, 6 figures

We present a locality preserving unitary mapping from fermions to qubits on a 2D torus whilst accounting for the mapping of topological sectors. Extending the work of Shukla et al. [Phys. Rev. B 101, 155105], an explicit intertwiner is constructed in the form of a projected entangled pair operator. By encoding the information about the charge sectors (and if applicable the twisted boundary conditions) in ancillary qubit(s), the intertwiner becomes a unitary operator which exchanges boundary conditions and charge sectors., Comment: 5 pages, many figures, comments welcome

We present overlapping top gate electrodes for the formation of gate defined lateral junctions in semiconducting layers as an alternative to the back gate/top gate combination and to the split gate configuration. The optical lithography microfabrication of the overlapping top gates is based on multiple layers of low-temperature atomic layer deposited hafnium oxide, which acts as a gate dielectric and as a robust insulating layer between two overlapping gate electrodes exhibiting a large dielectric breakdown field of > 1E9 V/m. The advantage of overlapping gates over the split gate approach is confirmed in model calculations of the electrostatics of the gate stack. The overlapping gate process is applied to Hall bar devices of mercury telluride in order to study the interaction of different quantum Hall states in the nn', np, pn and pp' regime., Comment: 7 pages, 3 figures

Protecting privacy during inference with deep neural networks is possible by adding noise to the activations in the last layers prior to the final classifiers or other task-specific layers. The activations in such layers are known as "features" (or, less commonly, as "embeddings" or "feature embeddings"). The added noise helps prevent reconstruction of the inputs from the noisy features. Lower bounding the variance of every possible unbiased estimator of the inputs quantifies the confidentiality arising from such added noise. Convenient, computationally tractable bounds are available from classic inequalities of Hammersley and of Chapman and Robbins -- the HCR bounds. Numerical experiments indicate that the HCR bounds are on the precipice of being effectual for small neural nets with the data sets, "MNIST" and "CIFAR-10," which contain 10 classes each for image classification. The HCR bounds appear to be insufficient on their own to guarantee confidentiality of the inputs to inference with standard deep neural nets, "ResNet-18" and "Swin-T," pre-trained on the data set, "ImageNet-1000," which contains 1000 classes. Supplementing the addition of noise to features with other methods for providing confidentiality may be warranted in the case of ImageNet. In all cases, the results reported here limit consideration to amounts of added noise that incur little degradation in the accuracy of classification from the noisy features. Thus, the added noise enhances confidentiality without much reduction in the accuracy on the task of image classification., Comment: 18 pages, 6 figures

Advances in Generative Artificial Intelligence (AI) are resulting in AI-generated media output that is (nearly) indistinguishable from human-created content. This can drastically impact users and the media sector, especially given global risks of misinformation. While the currently discussed European AI Act aims at addressing these risks through Article 52's AI transparency obligations, its interpretation and implications remain unclear. In this early work, we adopt a participatory AI approach to derive key questions based on Article 52's disclosure obligations. We ran two workshops with researchers, designers, and engineers across disciplines (N=16), where participants deconstructed Article 52's relevant clauses using the 5W1H framework. We contribute a set of 149 questions clustered into five themes and 18 sub-themes. We believe these can not only help inform future legal developments and interpretations of Article 52, but also provide a starting point for Human-Computer Interaction research to (re-)examine disclosure transparency from a human-centered AI lens., Comment: Accepted to CHI 2024 Late-Breaking Work