Search

Frontier artificial intelligence (AI) systems pose increasing risks to society, making it essential for developers to provide assurances about their safety. One approach to offering such assurances is through a safety case: a structured, evidence-based argument aimed at demonstrating why the risk associated with a safety-critical system is acceptable. In this article, we propose a safety case template for offensive cyber capabilities. We illustrate how developers could argue that a model does not have capabilities posing unacceptable cyber risks by breaking down the main claim into progressively specific sub-claims, each supported by evidence. In our template, we identify a number of risk models, derive proxy tasks from the risk models, define evaluation settings for the proxy tasks, and connect those with evaluation results. Elements of current frontier safety techniques - such as risk models, proxy tasks, and capability evaluations - use implicit arguments for overall system safety. This safety case template integrates these elements using the Claims Arguments Evidence (CAE) framework in order to make safety arguments coherent and explicit. While uncertainties around the specifics remain, this template serves as a proof of concept, aiming to foster discussion on AI safety cases and advance AI assurance., Comment: 21 pages, 6 figures, 1 table

We sketch how developers of frontier AI systems could construct a structured rationale -- a 'safety case' -- that an AI system is unlikely to cause catastrophic outcomes through scheming. Scheming is a potential threat model where AI systems could pursue misaligned goals covertly, hiding their true capabilities and objectives. In this report, we propose three arguments that safety cases could use in relation to scheming. For each argument we sketch how evidence could be gathered from empirical evaluations, and what assumptions would need to be met to provide strong assurance. First, developers of frontier AI systems could argue that AI systems are not capable of scheming (Scheming Inability). Second, one could argue that AI systems are not capable of posing harm through scheming (Harm Inability). Third, one could argue that control measures around the AI systems would prevent unacceptable outcomes even if the AI systems intentionally attempted to subvert them (Harm Control). Additionally, we discuss how safety cases might be supported by evidence that an AI system is reasonably aligned with its developers (Alignment). Finally, we point out that many of the assumptions required to make these safety arguments have not been confidently satisfied to date and require making progress on multiple open research problems.

Humans acquire knowledge by observing the external world, but also by introspection. Introspection gives a person privileged access to their current state of mind (e.g., thoughts and feelings) that is not accessible to external observers. Can LLMs introspect? We define introspection as acquiring knowledge that is not contained in or derived from training data but instead originates from internal states. Such a capability could enhance model interpretability. Instead of painstakingly analyzing a model's internal workings, we could simply ask the model about its beliefs, world models, and goals. More speculatively, an introspective model might self-report on whether it possesses certain internal states such as subjective feelings or desires and this could inform us about the moral status of these states. Such self-reports would not be entirely dictated by the model's training data. We study introspection by finetuning LLMs to predict properties of their own behavior in hypothetical scenarios. For example, "Given the input P, would your output favor the short- or long-term option?" If a model M1 can introspect, it should outperform a different model M2 in predicting M1's behavior even if M2 is trained on M1's ground-truth behavior. The idea is that M1 has privileged access to its own behavioral tendencies, and this enables it to predict itself better than M2 (even if M2 is generally stronger). In experiments with GPT-4, GPT-4o, and Llama-3 models (each finetuned to predict itself), we find that the model M1 outperforms M2 in predicting itself, providing evidence for introspection. Notably, M1 continues to predict its behavior accurately even after we intentionally modify its ground-truth behavior. However, while we successfully elicit introspection on simple tasks, we are unsuccessful on more complex tasks or those requiring out-of-distribution generalization., Comment: 15 pages, 9 figures

Our experiments on a sphere falling under gravity in Stokes flow show significant history effects. We observe an algebraic, not exponential, relaxation rate to the terminal velocity, validating the solution to the Basset-Boussinesq-Oseen equation. Unlike in steady Stokes theory, our experiments and theory reveal a vortex ring forming around the sphere and drifting away. As the Reynolds number nears unity, the vortex ring lags behind the sphere, departing from Stokesian theory, though the sphere's algebraic response persists. These findings are critical for interactions in the Stokes limit.

Large language models (LLMs) have demonstrated impressive performance in mathematical and commonsense reasoning tasks using chain-of-thought (CoT) prompting techniques. But can they perform emotional reasoning by concatenating `Let's think step-by-step' to the input prompt? In this paper we investigate this question along with introducing a novel approach to zero-shot emotion detection and emotional reasoning using LLMs. Existing state of the art zero-shot approaches rely on textual entailment models to choose the most appropriate emotion label for an input text. We argue that this strongly restricts the model to a fixed set of labels which may not be suitable or sufficient for many applications where emotion analysis is required. Instead, we propose framing the problem of emotion analysis as a generative question-answering (QA) task. Our approach uses a two step methodology of generating relevant context or background knowledge to answer the emotion detection question step-by-step. Our paper is the first work on using a generative approach to jointly address the tasks of emotion detection and emotional reasoning for texts. We evaluate our approach on two popular emotion detection datasets and also release the fine-grained emotion labels and explanations for further training and fine-tuning of emotional reasoning systems.

Submillimeter and millimeter wavelengths provide a unique view of the Universe, from the gas and dust that fills and surrounds galaxies to the chromosphere of our own Sun. Current single-dish facilities have presented a tantalising view of the brightest (sub-)mm sources, and interferometers have provided the exquisite resolution necessary to analyse the details in small fields, but there are still many open questions that cannot be answered with current facilities. In this report we summarise the science that is guiding the design of the Atacama Large Aperture Submillimeter Telescope (AtLAST). We demonstrate how tranformational advances in topics including star formation in high redshift galaxies, the diffuse circumgalactic medium, Galactic ecology, cometary compositions and solar flares motivate the need for a 50m, single-dish telescope with a 1-2 degree field of view and a new generation of highly multiplexed continuum and spectral cameras. AtLAST will have the resolution to drastically lower the confusion limit compared to current single-dish facilities, whilst also being able to rapidly map large areas of the sky and detect extended, diffuse structures. Its high sensitivity and large field of view will open up the field of submillimeter transient science by increasing the probability of serendipitous detections. Finally, the science cases listed here motivate the need for a highly flexible operations model capable of short observations of individual targets, large surveys, monitoring programmes, target of opportunity observations and coordinated observations with other observatories. AtLAST aims to be a sustainable, upgradeable, multipurpose facility that will deliver orders of magnitude increases in sensitivity and mapping speeds over current and planned submillimeter observatories., Comment: 47 pages, 12 figures. For further details on AtLAST see https://atlast.uio.no

The identification of Figurative Language (FL) features in text is crucial for various Natural Language Processing (NLP) tasks, where understanding of the author's intended meaning and its nuances is key for successful communication. At the same time, the use of a specific blend of various FL forms most accurately reflects a writer's style, rather than the use of any single construct, such as just metaphors or irony. Thus, we postulate that FL features could play an important role in Authorship Attribution (AA) tasks. We believe that our is the first computational study of AA based on FL use. Accordingly, we propose a Multi-task Figurative Language Model (MFLM) that learns to detect multiple FL features in text at once. We demonstrate, through detailed evaluation across multiple test sets, that the our model tends to perform equally or outperform specialized binary models in FL detection. Subsequently, we evaluate the predictive capability of joint FL features towards the AA task on three datasets, observing improved AA performance through the integration of MFLM embeddings.

The behavior and decision making of groups or communities can be dramatically influenced by individuals pushing particular agendas, e.g., to promote or disparage a person or an activity, to call for action, etc.. In the examination of online influence campaigns, particularly those related to important political and social events, scholars often concentrate on identifying the sources responsible for setting and controlling the agenda (e.g., public media). In this article we present a methodology for detecting specific instances of agenda control through social media where annotated data is limited or non-existent. By using a modest corpus of Twitter messages centered on the 2022 French Presidential Elections, we carry out a comprehensive evaluation of various approaches and techniques that can be applied to this problem. Our findings demonstrate that by treating the task as a textual entailment problem, it is possible to overcome the requirement for a large annotated training dataset.

Social media platforms are popular tools for disseminating targeted information during major public events like elections or pandemics. Systematic analysis of the message traffic can provide valuable insights into prevailing opinions and social dynamics among different segments of the population. We are specifically interested in influence spread, and in particular whether more deliberate influence operations can be detected. However, filtering out the essential messages with telltale influence indicators from the extensive and often chaotic social media traffic is a major challenge. In this paper we present a novel approach to extract influence indicators from messages circulating among groups of users discussing particular topics. We build upon the concept of a convo to identify influential authors who are actively promoting some particular agenda around that topic within the group. We focus on two influence indicators: the (control of) agenda and the use of emotional language., Comment: Accepted to LREC-COLING 2024

Red novae are transients powered by collisions of non-compact stars. Among their progenitors are systems of evolved subgiants and giants stars. Remnants of such red novae display bipolar structures which have remarkably close characteristics to many post-AGB or pre-PN systems. It is important to ask (and eventually verify) whether some of the less-known post-main-sequence objects (mis-)classified as pre-PNe can be merger remnants similar to the red nova remnants., Comment: proceedings of IAU Symposium on Planetary Nebulae: a Universal Toolbox in the Era of Precision Astrophysics

Nova 1670 is a historical transient bearing strong similarities to a recently-recognized type of stellar eruptions known as red novae, which are thought to be powered by stellar mergers. The remnant of the transient, CK Vul, is observable today mainly through cool circumstellar gas and dust, and recombining plasma, but we have no direct view on the stellar object. Within the merger hypothesis, we aim to infer the most likely makeup of the progenitor system that resulted in Nova 1670. We collect and summarize the literature data on the physical properties of the outburst and the remnant, and on the chemical composition of the circumstellar material which resulted from optical and submillimeter observations of the circumstellar gas of CK Vul. Simple simulations yield the form and level of mixing of material associated with the merger. Products of nuclear burning are identified, among them ashes of H burning in the CNO cycles and the MgAl chain, as well as of partial He burning. Based on the luminosity and chemical composition of the remnant, we find that the progenitor primary had to be an evolutionarily advanced red-giant branch star of a mass of 1-2 M$_{\odot}$. The secondary was either a very similar giant, or a He white dwarf. While the eruption event was mainly powered by accretion, we estimate that about 12% of total energy might have come from He burning activated during the merger. The coalescence of a first-ascent giant with a He white dwarf created a star with a rather unique internal structure and composition, which resemble these of early R-type carbon stars. Nova 1670 was the result of a merger between a He white dwarf and a first-ascent red giant and is likely now evolving to become an early R-type carbon star., Comment: accepted to A&A, comments welcome

Research into AI alignment has grown considerably since the recent introduction of increasingly capable Large Language Models (LLMs). Unfortunately, modern methods of alignment still fail to fully prevent harmful responses when models are deliberately attacked. Such vulnerabilities can lead to LLMs being manipulated into generating hazardous content: from instructions for creating dangerous materials to inciting violence or endorsing unethical behaviors. To help mitigate this issue, we introduce Bergeron: a framework designed to improve the robustness of LLMs against attacks without any additional parameter fine-tuning. Bergeron is organized into two tiers; with a secondary LLM acting as a guardian to the primary LLM. This framework better safeguards the primary model against incoming attacks while monitoring its output for any harmful content. Empirical analysis reviews that by using Bergeron to complement models with existing alignment training, we can significantly improve the robustness and safety of multiple, commonly used commercial and open-source LLMs. Specifically, we found that models integrated with Bergeron are, on average, nearly seven times more resistant to attacks compared to models without such support.

U Equ is an unusual maser-hosting IR source discovered in the 1990s. It was tentatively classified as a post-AGB star with a unique optical spectrum displaying rare emission and absorption features from molecular gas. In 2022, we discovered that its optical spectrum has drastically changed. Methods: Optical high-resolution spectra of U Equ from SALT are supplemented by archival data and NIR photometry from NOT. New spectral line observations with the Effelsberg telescope and ALMA are presented. Results: No circumstellar molecular features are present in the contemporary optical spectra of U Equ. Non-photospheric absorption and emission from neutral and ionized species dominate the current spectrum. Some of the observed features indicate an outflow with a terminal velocity of 215 km\s. The H\&K lines of [Ca II] indicate a photosphere of spectral type F. Photometric measurements show that the source has been monotonically increasing its optical and NIR fluxes since the beginning of this century. SEDs at different epochs show dusty circumstellar material arranged in a highly-inclined disk. At a distance of 4 kpc, the source's luminosity is 10$^4$ L$_{\odot}$. Conclusions: The object has changed considerably in the last three decades, either due to geometrical reconfiguration of the circumstellar medium, evolutionary changes in the central star, or owing to an accretion event that has started in the system very recently. Observationally, U Equ appears to resemble the Category 0 of disk-hosting post-AGB stars, especially the post-common envelope binary HD 101584. It is uncertain if the drastic spectral change and the associated optical/MIR rise in brightness are common in post-AGB stars but such a radical change may be related to the real-time onset of the evolution of the system into a planetary nebula. We find that the post-AGB star V576 Car has undergone a similar transformation as U Equ., Comment: Comments welcome!

Context. Galactic red novae are thought to be produced in stellar mergers between non-compact stars, such as main-sequence stars and cool giants. They are hoped to help in explaining physical processes involved in common envelope evolution and stellar binary collisions. Aims. We investigate the presence of lithium in three best-observed Galactic red nova remnants. Explaining the origin of lithium may point to mixing mechanism present before, during, or after the merger. Methods. The lithium line at 6707.81 A was compared to a feature of [Ca I at 6572.78 A to derive relative abundances in circumstellar gas. Absolute abundances were next calculated assuming the Solar calcium to lithium abundance ratio. Results. Lithium abundances were measured in the merger remnants of V838 Mon with A(Li)=2.3, CK Vul with A(Li)=2.5, and V1309 Sco with A(Li)=1.8. Conclusions. Lithium is overabundant in red novae, suggesting that at least some merger products activate the Cameron-Fowler mechanism whereby convective mixing can reach the deep stellar interior. Whether deep convection and associated diffusion alone or some other processes (e.g. spin down) can be responsible for driving the Cameron-Fowler mechanism in the remnants requires further studies. Early observations of lithium in V838 Mon hint that these mechanisms can activate early, perhaps already in the common envelope phase. These observations should be taken into account in modelling these complex systems., Comment: submitted to A&A; comments welcome

Recent advances in large-scale language modeling and generation have enabled the creation of dialogue agents that exhibit human-like responses in a wide range of conversational scenarios spanning a diverse set of tasks, from general chit-chat to focused goal-oriented discourse. While these agents excel at generating high-quality responses that are relevant to prior context, they suffer from a lack of awareness of the overall direction in which the conversation is headed, and the likelihood of task success inherent therein. Thus, we propose a framework in which dialogue agents can evaluate the progression of a conversation toward or away from desired outcomes, and use this signal to inform planning for subsequent responses. Our framework is composed of three key elements: (1) the notion of a "global" dialogue state (GDS) space, (2) a task-specific progression function (PF) computed in terms of a conversation's trajectory through this space, and (3) a planning mechanism based on dialogue rollouts by which an agent may use progression signals to select its next response., Comment: Accepted at NAACL 2022

Detecting molecular line emission from classical nova remnants has the potential of revealing information on the composition of the ejecta, in particular, it can deliver accurate isotopic ratios in the matter processed by a thermonuclear runaway. We conducted searches toward more than 100 classical novae for emission in lines of the CO or HCN molecules using single-dish telescopes and interferometric arrays at millimeter and submillimeter wavelengths. The survey demonstrates that classical novae, young or old, are not strong sources of molecular emission at submillimeter and millimeter wavelengths. Additionally, we mapped CO emission around Nova Persei 1901 (GK Per), earlier claimed to be circumstellar in origin. Our measurements indicate that the observed emission is from the interstellar medium. Although no molecular emission at millimeter and submillimeter wavelengths has been found in classical novae, it is still likely that some will be detected with high-sensitivity interferometers such as ALMA., Comment: comments welcome, submitted to A&A

V838 Mon erupted in 2002 quickly becoming the prototype of a new type of stellar eruptions known today as (luminous) red novae. The red nova outbursts are thought to be caused by stellar mergers. The merger in V838 Mon took place in a triple or higher system involving two B-type stars. We mapped the merger site with ALMA at a resolution of 25 mas in continuum dust emission and in rotational lines of simple molecules, including CO, SiO, SO, SO$_2$, AlOH, and H$_2$S. We use radiative transfer calculations to reproduce the remnant's architecture at the epoch of the ALMA observations. For the first time, we identify the position of the B-type companion relative to the outbursting component of V838 Mon. The stellar remnant is surrounded by a clumpy wind with characteristics similar to winds of red supergiants. The merger product is also associated with an elongated structure, $17.6 \times 7.6$ mas, seen in continuum emission, and which we interpret as a disk seen at a moderate inclination. Maps of continuum and molecular emission show also a complex region of interaction between the B-type star (its gravity, radiation, and wind) and the flow of matter ejected in 2002. The remnant's molecular mass is about 0.1 M$_{\odot}$ and the dust mass is 8.3$\cdot$10$^{-3}$ M$_{\odot}$. The mass of the atomic component remains unconstrained. The most interesting region for understanding the merger of V838 Mon remains unresolved but appears elongated. To study it further in more detail will require even higher angular resolutions. ALMA maps show us an extreme form of interaction between the merger ejecta with a distant (250 au) companion. This interaction is similar to that known from the Antares AB system but at a much higher mass loss rate. The B-type star not only deflects the merger ejecta but also changes its chemical composition with an involvement of circumstellar shocks., Comment: A&A submitted. Comments welcome