How does the conversational past of large language models (LLMs) influence their future performance? Recent work suggests that LLMs are affected by their conversational history in unexpected ways. For instance, hallucinations in prior interactions may influence subsequent model responses. In this work, we introduce History-Echoes, a framework that investigates how conversational history biases subsequent generations. The framework explores this bias from two perspectives: probabilistically, we model conversations as Markov chains to quantify state consistency; geometrically, we measure the consistency of consecutive hidden representations. Across three model families and six datasets spanning diverse phenomena, our analysis reveals a strong correlation between the two perspectives. By bridging these perspectives, we demonstrate that behavioral persistence manifests as a geometric trap, where gaps in the latent space confine the model's trajectory. Code available at https://github.com/technion-cs-nlp/OldHabitsDieHard.
Recent advances in large language models (LLMs) enable agentic systems trained with reinforcement learning (RL) over multi-turn interaction trajectories, but practical deployment is bottlenecked by rapidly growing textual histories that inflate token budgets and memory usage. We introduce AgentOCR, a framework that exploits the superior information density of visual tokens by representing the accumulated observation-action history as a compact rendered image. To make multi-turn rollouts scalable, AgentOCR proposes segment optical caching. By decomposing history into hashable segments and maintaining a visual cache, this mechanism eliminates redundant re-rendering. Beyond fixed rendering, AgentOCR introduces agentic self-compression, where the agent actively emits a compression rate and is trained with compression-aware reward to adaptively balance task success and token efficiency. We conduct extensive experiments on challenging agentic benchmarks, ALFWorld and search-based QA. Remarkably, results demonstrate that AgentOCR preserves over 95\% of text-based agent performance while substantially reducing token consumption (>50\%), yielding consistent token and memory efficiency. Our further analysis validates a 20x rendering speedup from segment optical caching and the effective strategic balancing of self-compression.
Wikipedia's founders could not have dreamed they were creating the most important laboratory for social scientific and computing research in history but that is exactly what happened. Hill and Shaw take account of Wikipedia's enormous effect on academic scholarship
Sibylle Kautz-Freimuth, Arim Shukri, Claudia Stracke
et al.
Abstract Background Patients who actively engage in their medical decision-making processes can experience better health outcomes. This exploratory study aimed to identify predictors of preferred and actual roles in decision-making in healthy women with BRCA1/2 pathogenic variants (PVs). Methods Women with BRCA1/2 PVs without a history of breast and/or ovarian cancer were recruited in six centres across Germany. Those returning the baseline questionnaires (T1) were randomly assigned to the intervention or control group (IG, CG). The IG completed a decision-coaching (DC) programme, the CG received standard care. A second survey (T2) followed after 12 weeks. Ordinal regression analyses were performed. Sociodemographic and outcome-related baseline variables were used to identify predictors of (i) desired role at T1 in the total group and (ii) actual role at T2 in the CG and the IG. Role preferences were measured with the Control Preferences Scale. Results 389 women completed the baseline questionnaires, 191 were randomised to the CG and 198 to the IG. At T1, high decisional conflict (OR 1.016, 95% CI 1.001–1.023, p = 0.038) and a negative self-concept (OR 1.030, 95% CI 1.008–1.054, p = 0.009) were significant predictors for preferring a more passive role. At T2, high baseline decisional conflict significantly predicted taking a more passive role in the CG, whereas in the IG, baseline decisional conflict showed no influence. Furthermore, in the IG, younger age (OR 1.049, 95% CI 1.001–1.098, p = 0.044) and a non-academic education (OR 0.46, 95% CI 0.213–0.775, p = 0.006) were identified as significant predictors for taking a more active role. Conclusions High initial decisional conflict was identified as an important predictor for preferring and taking a passive role in decision-making among women with BRCA1/2 PVs. Participating in the DC programme can counteract passivating effects of an initially high decisional conflict and particularly support younger PV carriers and those with lower educational status to take an active role. With this profile, the DC programme expands the existing counselling and care concept to include a measure that can also specifically cover the support needs of younger women and those with a lower education level. Trial registration DRKS-ID: DRKS00015527. Registered 30/10/2019.
Neoplasms. Tumors. Oncology. Including cancer and carcinogens
The aim of this study is to present the history of the establishment and growth of some tree species, which are nonnative to the Polish flora, in north-eastern Poland up to the present day. The introduction of that tree species to Europe or its parts (Weymouth pine, Douglas fir, western red cedar, and other species) was initiated by Schwappach at the end of the 19th century in Germany and Prussia on areas between 0.05 and 0.15 ha. The tree growth and natural regeneration were observed until 1945 and continued today. The present study updates the information on the location of some of these stands in north-eastern Poland (Warmia and Mazury), taking into account their dendrometric parameters for future studies.
This study investigates precedent names as nationally marked linguistic units and cultural symbols. It presents data from a free associative experiment conducted in 2022 with two groups: Germans from the Federal Republic of Germany and Russian Germans. The stimuli for the experiment were the precedent names Stalin and Catherine II. Based on the collected data, an associative field was constructed. Utilizing semantic interpretation methodology, the psychologically salient meanings of these proper names were identified. The inclusion of Russian Germans as participants is justified by their unique dual identity, situated at the intersection of German and Russian mentalities. The experiment revealed significant differences in perceptions of Stalin between the ethnic groups studied. In the linguistic consciousness of Germans from the Federal Republic, the name Stalin is primarily associated with geographical location, national affiliation, or a precedent event. In contrast, the psycholinguistic significance of the name Stalin among Russian Germans carries a distinctly negative connotation and is linked to dramatic historical processes. The precedent name Catherine II exhibits a similar connotative significance across both groups but reveals differences in perceptions regarding the ruler's national identity and her role in the country's history.
Slavic languages. Baltic languages. Albanian languages
While large language models (LLMs) have proven effective in leveraging textual data for recommendations, their application to multimodal recommendation tasks remains relatively underexplored. Although LLMs can process multimodal information through projection functions that map visual features into their semantic space, recommendation tasks often require representing users' history interactions through lengthy prompts combining text and visual elements, which not only hampers training and inference efficiency but also makes it difficult for the model to accurately capture user preferences from complex and extended prompts, leading to reduced recommendation performance. To address this challenge, we introduce HistLLM, an innovative multimodal recommendation framework that integrates textual and visual features through a User History Encoding Module (UHEM), compressing multimodal user history interactions into a single token representation, effectively facilitating LLMs in processing user preferences. Extensive experiments demonstrate the effectiveness and efficiency of our proposed mechanism.
Pascal Querner, Nikola Szucsich, Bill Landsberger
et al.
This paper introduces a new method for identifying museum pests through the analysis of DNA present in settled dust. Traditionally, the identification of pests in cultural institutions such as museums and depositories has relied on insect trapping (monitoring). They give good results but need time (minimum spring until summer of one year for a complete survey) and face challenges related to the identification of small, rare, or damaged species. Our study presents a non-invasive approach that utilizes metabarcoding analysis of dust samples to identify pests and other arthropods at the species level. We collected dust samples with a handheld vacuum cleaner in summer 2023 from the six different floors of the Natural History Museum in Vienna and compared the results with the insect monitoring. We found over 359 different species of arthropods in the museum and could show how the diversity increases with the elevation of the building floor. This method could be used for rapid and cost-effective screening of pests before monitoring. But the interpretation of results is sometimes difficult (for Lepismatidae, for example), and it cannot replace a continuous monitoring of pests with traps. This investigation might present the highest indoor animal biodiversity ever found in a single building.
The question of scientific understanding of the features of the Republic of Crimea and the city of Sevastopol reintegration into the sociocultural and identification space of the Russian Federation is relevant from the point of view of using this experience in the integration processes taking place today in new Russian territories. The theoretical approaches of modern Russian and Western authors devoted to the identification side of integration using the example of the formation of the European Union, the reintegration of East and West Germany, as well as the entry of Crimea into the Russian Federation were taken as the conceptual basis for the study. The study relied on a combination of quantitative and qualitative data collection and analysis methods. The applied part was carried out at Sevastopol Federal University. Analysis of the data obtained allowed us to draw several conclusions. There are some multidirectional trends in the formation of the Russian national-state identity of residents of the regions. The majority of residents of the Republic of Crimea and the city of Sevastopol generally identify themselves with Russia and see their regions as part of it. They consider common space, history, culture, and traditions to be the basis for this. They consider these same factors, to a much greater extent than the political or civil components, to be the basis of patriotism. Among these factors, a significant role is played by the memory of historical events of different periods, perceived as part of a single Russian history. At the same time, in the system of spatial and historical images there is a tendency towards a region-centric perception of the country, a desire to highlight and emphasize the special significance of its territory both in history and in the modern space of the Russian Federation. The desire to preserve the unity and cohesion of the peoples of the country, expressed in various forms, became significant. There is some difference here from the results of all-Russian studies - representatives of other nationalities are perceived much more neutrally and calmly than in Russia as a whole. Foreign countries act as “others”, enemies or competitors. At the same time, the images of these countries are poorly specified and reflect the superficial reproduction of media information rather than the result of one’s own understanding of events.
Abinash Kumar Shaw, Raghunath Ghara, Paz Beniamini
et al.
We propose different estimators to probe the intergalactic medium (IGM) during epoch of reionization (EoR) using the dispersion measure (${\rm DM}$) of the fast radio bursts. We consider three different reionization histories, which we can distinguish with a total of $\lesssim 1000\,{\rm DM}$ measurements during EoR if their redshifts are known. We note that the redshift derivatives of ${\rm DM}$ are also directly sensitive to the reionization history. The major point of this work is to explore the variance in the ${\rm DM}$ measurements and the information encoded in them. We find that the all-sky average $\overline{\rm DM}(z)$ gets biased from the line-of-sight (LoS) fluctuations in the ${\rm DM}$ measurements introduced by the ionization of IGM during EoR. We find that the ratio $σ_{\rm DM}/\overline{\rm DM}$ depends directly on the ionization bubble sizes as well as the reionization history. On the other hand, we also find that angular variance (coined as $\textit{structure function}$) of ${\rm DM}$ encodes the information about the duration of reionization and the typical bubble sizes as well. We establish the usefulness of variances in ${\rm DM}$ using toy models of reionization and later verify it with the realistic reionization simulations.
Lekshmi Thulasidharan, Elena D'Onghia, Robert Benjamin
et al.
The prevailing model of galaxy formation proposes that galaxies like the Milky Way are built through a series of mergers with smaller galaxies over time. However, the exact details of the Milky Way's assembly history remain uncertain. In this study, we show that the Milky Way's merger history is uniquely encoded in the vertical thickness of its stellar disk. By leveraging age estimates from the value-added LAMOST DR8 catalog and the StarHorse ages from SDSS-IV DR12 data, we investigate the relationship between disk thickness and stellar ages in the Milky Way using a sample comprising Red Giants (RG), Red Clump Giants (RCG), and metal-poor stars (MPS). Guided by the IllustrisTNG50 simulations, we show that an increase in the dispersion of the vertical displacement of stars in the disk traces its merger history. This analysis reveals the epoch of a major merger event that assembled the Milky Way approximately 11.13 billion years ago, as indicated by the abrupt increase in disk thickness among stars of that age, likely corresponding to the Gaia-Sausage Enceladus (GSE) event. The data do not exclude an earlier major merger, which may have occurred about 1.3 billion years after the Big Bang. Furthermore, the analysis suggests that the geometric thick disk of the Milky Way was formed around 11.13 billion years ago, followed by a transition period of approximately 2.6 billion years leading to the formation of the geometric thin disk, illustrating the galaxy's structural evolution. Additionally, we identified three more recent events -- 5.20 billion, 2.02 billion, and 0.22 billion years ago -- potentially linked to multiple passages of the Sagittarius dwarf galaxy. Our study not only elucidates the complex mass assembly history of the Milky Way and highlights its past interactions but also introduces a refined method for examining the merger histories of external galaxies.
When Germany annexed colonies in Africa and the Pacific beginning in the 1880s, many German women were enthusiastic. At the same time, however, they found themselves excluded from what they saw as a great nationalistic endeavor. In German Women for Empire, 1884–1945 Lora Wildenthal untangles the varied strands of racism, feminism, and nationalism that thread through German women’s efforts to participate in this episode of overseas colonization. In confrontation and sometimes cooperation with men over their place in the colonial project, German women launched nationalist and colonialist campaigns for increased settlement and new state policies. Wildenthal analyzes recently accessible Colonial Office archives as well as mission society records, periodicals, women’s memoirs, and fiction to show how these women created niches for themselves in the colonies. They emphasized their unique importance for white racial “purity” and the inculcation of German culture in the family. While pressing for career opportunities for themselves, these women also campaigned against interracial marriage and circulated an image of African and Pacific women as sexually promiscuous and inferior. As Wildenthal discusses, the German colonial imaginary persisted even after the German colonial empire was no longer a reality. The women’s colonial movement continued into the Nazi era, combining with other movements to help turn the racialist thought of the late-nineteenth and early-twentieth centuries into the hierarchical evaluation of German citizens as well as colonial subjects. Students and scholars of women’s history, modern German history, colonial politics and culture, postcolonial theory, race/ethnicity, and gender will welcome this groundbreaking study.
The aim of this study is to unearth an unknown text by Aka Gündüz published in Germany in the in the second morning edition of 1 January 1915 of the ‘Frankfurter Zeitung und Handelsblatt’. This text titled “O Young German!” is an example of German–Turkish joint war propaganda during the First World War. The text was handwritten and signed by Aka Gündüz. In the introductory part of the newspaper text, “The most important Turkish poet of our age, Ata Gündüz wrote a poem to the Germans at our request. [...] We have added a photograph of the original text in Ottoman (i. e. Turkish) language next to the German translation.” note is included.
According to archive records, Aka Gündüz did not publish it in Turkey. The text published in the ‘Frankfurter Zeitung und Handelsblatt’ was translated from Ottoman into German by Dr. Fr. Schrader. When the handwritten version is compared with the one published in the newspaper, it is revealed that there are some (intentional or unintentional) translation errors. Although the name of the poet was written as ‘Ata Gündüz’ instead of ‘Aka Gündüz’ at the beginning of the text, when the signature in the text is compared with the signatures in other texts by Aka Gündüz, it is seen that it belongs to him. It was written in a bespoke manner to show the German people the friendship and unity of the Turkish allies. The text was created with idyllic landscapes and an emotional atmosphere. Some sentences such as “You are nice and humane…”, “As long as you and my Hungarian brother live, we will make this world strong”, “Let's move on to war. Come on, so that there will be no savagery, no hatred, no enmity in the world” reveal the propaganda purpose of the text. This text is important in terms of both war propaganda studies and the identification of an unknown text in terms of Turkish literary history.
We use the Millennium Simulation (MS) to study the statistics of Λ cold dark matter (ΛCDM) halo concentrations at z= 0. Our results confirm that the average halo concentration declines monotonically with mass; the concentration–mass relation is well fitted by a power law over three decades in mass, up to the most massive objects that form in a ΛCDM universe (∼ 1015 h−1 M⊙). This is in clear disagreement with the predictions of the model proposed by Bullock et al. for these rare objects, and agrees better with the original predictions of Navarro, Frenk & White. The large volume surveyed, together with the unprecedented numerical resolution of the MS, allows us to estimate with confidence the distribution of concentrations and, consequently, the abundance of systems with unusual properties. About one in a hundred cluster haloes (M200≳ 3 × 1014 h−1 M⊙) have concentrations exceeding c200= 7.5, a result that may be useful in interpreting the likelihood of unusually strong massive gravitational lenses, such as Abell 1689, in the ΛCDM cosmogony. A similar fraction of about 1 per cent of galaxy-sized haloes (M200∼ 1012 h−1 M⊙) have c200 < 4.5 and this could be relevant to models that attempt to reconcile the ΛCDM cosmology with rotation curves of low surface brightness galaxies by appealing to haloes of unexpectedly low concentration. We find that halo concentrations are independent of spin once haloes manifestly out of equilibrium have been removed from the sample. Compared to their relaxed brethren, the concentrations of out-of-equilibrium haloes tend to be lower and have more scatter, while their spins tend to be higher. A number of previously noted trends within the halo population are induced primarily by these properties of unrelaxed systems. Finally, we compare the result of predicting halo concentrations using the mass assembly history of the main progenitor with predictions based on simple arguments regarding the assembly time of all progenitors. The latter are typically as good or better than the former, suggesting that halo concentration depends not only on the evolutionary path of a halo's main progenitor, but on how and when all of its constituents collapsed to form non-linear objects.
Résumé en français : Les écrits de Theodor W. Adorno (1903-1969) présentent deux types de réflexions sur les femmes. Le premier « dénaturalise » la féminité en montrant ses ressorts historiques. Le second contribue à la « renaturaliser » symboliquement, notamment en faisant de ses caractéristiques dites naturelles autant de leviers pour la résistance au patriarcat. L’article propose d’interroger ces deux gestes tout en s’attardant sur le second, qui semble plus difficilement intégrable à la dialectique adornienne. Le geste de renaturalisation symbolique est-il réellement compatible avec la puissante démystification de la féminité qu’on trouve chez Adorno ? Comment s’intègre-t-il dans la dialectique adornienne de la nature ?