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.
Large Language Models (LLMs) have recently shown a promise and emergence of Theory of Mind (ToM) ability and even outperform humans in certain ToM tasks. To evaluate and extend the boundaries of the ToM reasoning ability of LLMs, we propose a novel concept, taxonomy, and framework, the ToM reasoning with Zero, Finite, and Infinite Belief History and develop a multi-round text-based game, called $\textit{Pick the Right Stuff}$, as a benchmark. We have evaluated six LLMs with this game and found their performance on Zero Belief History is consistently better than on Finite Belief History. In addition, we have found two of the models with small parameter sizes outperform all the evaluated models with large parameter sizes. We expect this work to pave the way for future ToM benchmark development and also for the promotion and development of more complex AI agents or systems which are required to be equipped with more complex ToM reasoning ability.
Jessica E. Thorne, Aaron S. G. Robotham, Sabine Bellstedt
et al.
Gas-phase metallicities of galaxies are typically measured through auroral or nebular emission lines, but metallicity also leaves an imprint on the overall spectral energy distribution (SED) of a galaxy and can be estimated through SED fitting. We use the ProSpect SED fitting code with a flexible parametric star formation history and an evolving metallicity history to self-consistently measure metallicities, stellar mass, and other galaxy properties for $\sim90\,000$ galaxies from the Deep Extragalactic VIsible Legacy Survey (DEVILS) and Galaxy and Mass Assembly (GAMA) survey. We use these to trace the evolution of the mass-metallicity relation (MZR) and show that the MZR only evolves in normalisation by $\sim0.1\,$dex at stellar mass $M_\star = 10^{10.5}\,M_\odot$. We find no difference in the MZR between galaxies with and without SED evidence of active galactic nuclei emission at low redshifts ($z<0.3$). Our results suggest an anti-correlation between metallicity and star formation activity at fixed stellar mass for galaxies with $M_\star > 10^{10.5}\,M_\odot$ for $z<0.3$. Using the star formation histories extracted using ProSpect we explore higher-order correlations of the MZR with properties of the star formation history including age, width, and shape. We find that at a given stellar mass, galaxies with higher metallicities formed most of their mass over shorter timescales, and before their peak star formation rate. This work highlights the value of exploring the connection of a galaxy's current gas-phase metallicity to its star formation history in order to understand the physical processes shaping the MZR.
We give a proof of the fact that a simply-connected symplectic homogeneous space $(M,ω)$ of a connected Lie group $G$ is the universal cover of a coadjoint orbit of a one-dimensional central extension of $G$. We emphasise the rôle of symplectic group cocycles and the relationship between such cocycles, left-invariant presymplectic structures on $G$ and central extensions of $G$; in particular, we show that integrability of a central extension of $\mathfrak{g}$ to a central extension of $G$ is equivalent to integrability of a representative Chevalley-Eilenberg 2-cocycle of $\mathfrak{g}$ to a symplectic cocycle of $G$.
A semiring is said to be centrally essential if for every non-zero element $x$, there exist two non-zero central elements $y, z$ with $xy = z$. We give some examples of non-commutative centrally essential semirings and describe some properties of additively cancellative centrally essential semirings.
Recently, context-dependent text-to-SQL semantic parsing which translates natural language into SQL in an interaction process has attracted a lot of attention. Previous works leverage context-dependence information either from interaction history utterances or the previous predicted SQL queries but fail in taking advantage of both since of the mismatch between natural language and logic-form SQL. In this work, we propose a History Information Enhanced text-to-SQL model (HIE-SQL) to exploit context-dependence information from both history utterances and the last predicted SQL query. In view of the mismatch, we treat natural language and SQL as two modalities and propose a bimodal pre-trained model to bridge the gap between them. Besides, we design a schema-linking graph to enhance connections from utterances and the SQL query to the database schema. We show our history information enhanced methods improve the performance of HIE-SQL by a significant margin, which achieves new state-of-the-art results on the two context-dependent text-to-SQL benchmarks, the SparC and CoSQL datasets, at the writing time.
Petra N. Tang, J. J. Eldridge, Elizabeth R. Stanway
et al.
We compare the impacts of uncertainties in both binary population synthesis models and the cosmic star formation history on the predicted rates of Gravitational Wave compact binary merger (GW) events. These uncertainties cause the predicted rates of GW events to vary by up to an order of magnitude. Varying the volume-averaged star formation rate density history of the Universe causes the weakest change to our predictions, while varying the metallicity evolution has the strongest effect. Double neutron-star merger rates are more sensitive to assumed neutron-star kick velocity than the cosmic star formation history. Varying certain parameters affects merger rates in different ways depending on the mass of the merging compact objects; thus some of the degeneracy may be broken by looking at all the event rates rather than restricting ourselves to one class of mergers.
The present notes refer to a lecture delivered on 27 September 2017 in Roscoff during the 2017 Evry Schatzman School. It concerns a general introduction to optical/IR interferometry, including a brief history, a presentation of the basic principles, some important theorems and relevant applications.The layout of these lecture notes is as follows. After a short introduction, we proceed with some reminders concerning the representation of a field of electromagnetic radiation. We then present a short history of interferometry, from the first experiment of Fizeau and Stefan to modern optical interferometers. We then discuss the notions of light coherence, including the van Cittert - Zernicke theorem and describe the principle of interferometry using two telescopes. We present some examples of modern interferometers and typical results obtained with these. Finally, we address three important theorems: the fundamental theorem, the convolution theorem and the Wiener-Khinchin theorem which enable to get a better insight into the field of optical/IR interferometry.
To understand a node's centrality in a multiplex network, its centrality values in all the layers of the network can be aggregated. This requires a normalization of the values, to allow their meaningful comparison and aggregation over networks with different sizes and orders. The concrete choices of such preprocessing steps like normalization and aggregation are almost never discussed in network analytic papers. In this paper, we show that even sticking to the most simple centrality index (the degree) but using different, classic choices of normalization and aggregation strategies, can turn a node from being among the most central to being among the least central. We present our results by using an aggregation operator which scales between different, classic aggregation strategies based on three multiplex networks. We also introduce a new visualization and characterization of a node's sensitivity to the choice of a normalization and aggregation strategy in multiplex networks. The observed high sensitivity of single nodes to the specific choice of aggregation and normalization strategies is of strong importance, especially for all kinds of intelligence-analytic software as it questions the interpretations of the findings.
Joshua P. Emery, Francesco Marzari, Alessandro Morbidelli
et al.
The Trojan asteroids provide a unique perspective on the history of Solar System. As a large population of small bodies, they record important gravitational interactions and dynamical evolution of the Solar System. In the past decade, significant advances have been made in understanding physical properties, and there has been a revolution in thinking about the origin of Trojans. The ice and organics generally presumed to be a significant part of Trojan compositions have yet to be detected directly, though low density of the binary system Patroclus (and possibly low density of the binary/moonlet system Hektor) is consistent with an interior ice component. By contrast, fine-grained silicates that appear to be similar to cometary silicates in composition have been detected, and a color bimodality may indicate distinct compositional groups among the Trojans. Whereas Trojans had traditionally been thought to have formed near 5 AU, a new paradigm has developed in which the Trojans formed in the proto-Kuiper Belt, and they were scattered inward and captured in the Trojan swarms as a result of resonant interactions of the giant planets. Whereas the orbital and population distributions of current Trojans are consistent with this origin scenario, there are significant differences between current physical properties of Trojans and those of Kuiper Belt objects. These differences may be indicative of surface modification due to the inward migration of objects that became the Trojans, but understanding of appropriate modification mechanisms is poor and would benefit from additional laboratory studies. Many open questions remain, and the future promises significant strides in our understanding of Trojans. The time is ripe for a spacecraft mission to the Trojans, to turn these objects into geologic worlds that can be studied in detail to unravel their complex history.
We present a new oblivious RAM that supports variable-sized storage blocks (vORAM), which is the first ORAM to allow varying block sizes without trivial padding. We also present a new history-independent data structure (a HIRB tree) that can be stored within a vORAM. Together, this construction provides an efficient and practical oblivious data structure (ODS) for a key/value map, and goes further to provide an additional privacy guarantee as compared to prior ODS maps: even upon client compromise, deleted data and the history of old operations remain hidden to the attacker. We implement and measure the performance of our system using Amazon Web Services, and the single-operation time for a realistic database (up to $2^{18}$ entries) is less than 1 second. This represents a 100x speed-up compared to the current best oblivious map data structure (which provides neither secure deletion nor history independence) by Wang et al. (CCS 14).
We derive the star formation history in four regions of the Small Magellanic Cloud (SMC) using the deepest VI color-magnitude diagrams (CMDs) ever obtained for this galaxy. The images were obtained with the Advanced Camera for Surveys onboard the Hubble Space Telescope and are located at projected distances of 0.5-2 degrees from the SMC center, probing the main body and the wing of the galaxy. We derived the star-formation histories (SFH) of the four fields using two independent procedures to fit synthetic CMDs to the data. We compare the SFHs derived here with our earlier results for the SMC bar to create a deep pencil-beam survey of the global history of the central SMC. We find in all the six fields observed with HST a slow star formation pace from 13 to 5-7 Gyr ago, followed by a ~ 2-3 times higher activity. This is remarkable because dynamical models do not predict a strong influence of either the LMC or the Milky Way (MW) at that time. The level of the intermediate-age SFR enhancement systematically increases towards the center, resulting in a gradient in the mean age of the population, with the bar fields being systematically younger than the outer ones. Star formation over the most recent 500 Myr is strongly concentrated in the bar, the only exception being the area of the SMC wing. The strong current activity of the latter is likely driven by interaction with the LMC. At a given age, there is no significant difference in metallicity between the inner and outer fields, implying that metals are well mixed throughout the SMC. The age-metallicity relations we infer from our best fitting models are monotonically increasing with time, with no evidence of dips. This may argue against the major merger scenario proposed by Tsujimoto and Bekki 2009, although a minor merger cannot be ruled out.
Recent X-ray observations of galaxy clusters suggest that cluster populations are bimodally distributed according to central gas entropy and are separated into two distinct classes: cool core (CC) and non-cool core (NCC) clusters. While it is widely accepted that AGN feedback plays a key role in offsetting radiative losses and maintaining many clusters in the CC state, the origin of NCC clusters is much less clear. At the same time, a handful of extremely powerful AGN outbursts have recently been detected in clusters, with a total energy ~10^{61}-10^{62} erg. Using two dimensional hydrodynamic simulations, we show that if a large fraction of this energy is deposited near the centers of CC clusters, which is likely common due to dense cores, these AGN outbursts can completely remove CCs, transforming them to NCC clusters. Our model also has interesting implications for cluster abundance profiles, which usually show a central peak in CC systems. Our calculations indicate that during the CC to NCC transformation, AGN outbursts efficiently mix metals in cluster central regions, and may even remove central abundance peaks if they are not broad enough. For CC clusters with broad central abundance peaks, AGN outbursts decrease peak abundances, but can not effectively destroy the peaks. Our model may simultaneously explain the contradictory (possibly bimodal) results of abundance profiles in NCC clusters, some of which are nearly flat, while others have strong central peaks similar to those in CC clusters. A statistical analysis of the sizes of central abundance peaks and their redshift evolution may shed interesting insights on the origin of both types of NCC clusters and the evolution history of thermodynamics and AGN activity in clusters.
We study merger histories of dark-matter haloes in a suite of N-body simulations that span different cosmological models. The simulated cases include the up-to-date WMAP5 cosmology and other test cases based on the Einstein-deSitter cosmology with different power spectra. We provide a robust fitting function for the conditional mass function (CMF) of progenitor haloes of a given halo. This fit is valid for the different cosmological models and for different halo masses and redshifts, and it is a significant improvement over earlier estimates. Based on this fit, we develop a simple and accurate technique for transforming the merger history of a given simulated halo into haloes of different mass, redshift and cosmology. Other statistics such as main-progenitor history and merger rates are accurately transformed as well. This method can serve as a useful tool for studying galaxy formation. It is less sensitive to the low accuracy of the fit at small time-steps, and it can thus replace the more elaborate task of construction Monte-Carlo realizations. As an alternative approach, we confirm the earlier finding by Neistein & Dekel that the main-progenitor follows a log-normal distribution. This property of merger trees allows us to better capture their behaviour as a function of time and descendant mass, but a broader suite of simulations is required for evaluating the dependence of the log-normal parameters on the cosmological model.
We introduce the notion of central extension of gerbes on a topological space. We then show that there are obstruction classes to lifting objects and isomorphisms in a central extension. We also discuss pronilpotent gerbes. These results are used in a subsequent paper to study twisted deformation quantization on algebraic varieties.
Over the past several years, the Hubble Space Telescope (HST) has acquired many broad-band images of various regions in the M31 disk. I have obtained 27 such fields from the HST data archive in order to produce color-magnitude diagrams (CMDs) of the stellar populations contained within these areas of the disk. I have attempted to reproduce these CMDs using theoretical stellar evolution models in conjunction with statistical tools for determining the star formation history which best fits the observations. The wide range of extinction values within any given field makes the data difficult to accurately reproduce; nevertheless, I have managed to find star formation histories which roughly reproduce the data. These statistically determined star formation histories reveal that, like the disk of the Galaxy, the disk of M31 contains very few old metal-poor stars. The histories also suggest that the star formation rate of the disk as a whole has been low over the past ~1 Gyr.