arXiv subjects

Pei Wang

Publications and source records attributed to Pei Wang.

At least 19 recordsLinked to original sources

arXiv2026

Polarization and Burst Properties of the Repeating FRB~20240209A in a Quiescent Elliptical Host

We report polarimetric observations of the repeating fast radio burst (FRB) source FRB 20240209A with the Green Bank Telescope at L band and 800 MHz. This source has previously been associated with a quiescent elliptical galaxy with a mass-weighted stellar age of approximately 11 Gyr. We detected three bursts, and the brightest burst resolves into eight components that drift toward lower frequencies. The bursts have mutually consistent rotation measures (RMs), and the small extragalactic RM agrees with the low RM predicted from the small host-galaxy dispersion measure ($\mathrm{DM}_\mathrm{host}$). The linear polarization fraction decreases toward lower frequencies within the brightest burst, consistent with depolarization by an RM scatter of $σ_\mathrm{RM}\simeq3.5$ rad m$^{-2}$. This RM scatter is comparable to those of several repeaters in star-forming galaxies. Therefore, FRBs in both star-forming and old, quiescent hosts can reside in relatively complex magneto-ionic environments, which suggests that FRB sources may form through multiple channels.

astro-ph.HE↗

arXiv2026

IQuest-Q1: Advancing Foundation Capabilities for Agentic CLI Systems

We present IQuest-Q1, a sparse Mixture-of-Experts (MoE) agentic foundation model with 320B total parameters (15B activated). IQuest-Q1 is developed through a three-stage process: (1) stable and efficient pre-training, combining Stable SplitNorm, Muon, and muP with a two-stage curriculum whose second stage increases the coding-data mixture, preserving a wide stability margin between safe and optimal learning rates; (2) fine-grained mid-training, expanding the data mixture across domains and formats and progressively extending the context; and (3) multi-scaffold with multi-version post-training, combining SFT, domain-specialized RL, and MOPD across various scaffolds to integrate complementary training signals and strengthen agentic coding capabilities. Furthermore, model merging combines different experts and checkpoints from different stages while mitigating conflicts among their parameter updates. The resulting IQuest-Q1 demonstrates capabilities in both coding and general tasks. We release IQuest-Q1 as an open-source model to support community research and development.

cs.CL↗

arXiv2026

A Unified Multi-Band Secular Framework for the Hierarchical Triple PSR J0435+3233 System

The millisecond pulsar PSR J0435+3233, discovered by FAST, exhibits an unusually large apparent spin-down rate, potentially reflecting line-of-sight acceleration from a distant companion. Recent investigations reveal a hierarchical triple consisting of an inner pulsar-helium white dwarf binary with an orbital period of ~8 days, orbited by a G-type subgiant tertiary on an eccentric ~74 yr orbit. The current timing description admits two distinct geometrical solutions. We develop a singly averaged (SA) formulation that averages over the inner orbit while retaining the instantaneous phase of the eccentric outer orbit, thereby capturing the time-dependent quadrupolar perturbations induced by the tertiary. We show how this physical three-body description can be incorporated naturally into pulsar-timing analyses and use it to forward-propagate the two currently viable configurations, Tv1 and Tv2, to forecast when their timing signatures become distinguishable with FAST. The SA evolution predicts that the two configurations will become distinguishable in the coming years, with the effective projected semi-major axis, $x_{1,\mathrm{eff}}$, providing the earliest discriminating observable. We further extend the framework to predict the tertiary's spectroscopic radial-velocity and astrometric evolution, linking FAST timing with LAMOST time-domain spectroscopy and Gaia astrometry. This multi-band framework has the potential to establish PSR J0435+3233 as an exceptionally well-constrained laboratory for three-body dynamics, compact-binary and stellar evolution, and gravitational physics.

astro-ph.HE↗

arXiv2026

Polarization Transport and Polarized Images of a Rotating Braneworld Black Hole

We investigate polarized synchrotron emission from an optically thin accretion flow around a rotating braneworld black hole, focusing on how higher-dimensional gravity modifies photon propagation and polarization transport. We integrate the null-geodesic and parallel-transport equations for the polarization four-vector and, within a common emissivity and magnetic-field prescription, examine the effects of tidal charge, black-hole spin, observer inclination, and prograde or retrograde orbital motion. We analyze the linearly polarized intensity, electric-vector position angle (EVPA), Stokes parameters, the joint $Q$--$U$ polarization-state space, and the cumulative net linear polarization degree, and compare representative braneworld configurations with Kerr and Kerr--Newman spacetimes. Within the parameter range considered, tidal charge and spin do not generate a qualitatively new large-scale polarization morphology. Instead, they primarily reorganize the polarized intensity, orientation, and polarization-state distribution in the strong-field region near the shadow boundary and photon ring. Viewing inclination controls the dominant projected morphology, while high inclinations enhance projection and Doppler asymmetries and make the distinction between prograde and retrograde flows more pronounced. The Stokes components and their joint state space expose local bending, stretching, and trajectory concentration that are less apparent in intensity images. The cumulative net linear polarization degree decreases with increasing cumulative radius, and larger tidal charge or spin lowers its amplitude for the representative models considered. Together, these complementary polarization observables provide a potential route for separating spacetime, viewing-geometry, and accretion-flow effects and for constraining braneworld tidal charge with horizon-scale polarimetry.

astro-ph.HE↗

arXiv2026

A dynamic magneto-ionic environment around a long-period radio transient

Faraday rotation provides one of the most direct probes of magnetized plasma along the line of sight, from the large-scale Galactic magnetic field to compact plasma environments local to radio sources. In particular, temporal variations in rotation measure can trace changes in the density, magnetic-field strength, or geometry of the Faraday-rotating medium, as observed in radio pulsars and fast radio bursts. Long-period radio transients (LPTs), a recently identified class of coherent radio sources with periods of hundreds to thousands of seconds, remain poorly explored because their physical nature is still uncertain. Here we present full-Stokes observations of the 421-s LPT source CHIME J0630+25 and report, for the first time, extreme rotation measure (RM) variability from approximately -400 to -1630 rad m^-2. The rapid RM changes occur on short timescales, including a difference of approximately 50 rad m^-2 between two consecutive rotation periods and a sudden change of approximately 536 rad m^-2 within approximately 2500 s. Compared with other radio-emitting sources, CHIME J0630+25 occupies an unusual region of the DM-RM plane: despite its very small DM of approximately 22 pc cm^-3, |RM| reaches over 1.6 x 10^3 rad m^-2, far exceeding that expected from the interstellar medium and pulsars. The fast and large-amplitude RM variations should arise from a compact, structured, and rapidly evolving magneto-ionic environment local to the source, possibly associated with a compact binary system.

astro-ph.HE↗

arXiv2026

Understanding Reliability in LLM-based Human Behavior Simulation

Large language models (LLMs) are increasingly used to simulate human survey responses and behavioral reactions, yet unreliable simulations can mislead social science conclusions. However, existing evaluations focus on end-to-end scores, leaving it unclear how different aspects of the simulation process interact to determine reliability. We propose ReliMap, which decomposes LLM-based human behavior simulation into three structured layers and evaluates reliability at both the individual level (R1) and population level (R2) across three configuration dimensions: model capacity, profile completeness, and population coverage. Through experiments across four simulation tasks and eleven LLMs, we find that all models exhibit substantial distributional bias without profile conditioning. Profile conditioning reduces this bias with diminishing returns. Larger models benefit more, and attribute informativeness matters more than quantity. Critically, R1 gains do not reliably transfer to R2--individual and population-level reliability can move in opposite directions. At the population layer, increasing coverage reduces variance but not systematic bias, with R2 stabilizing at around 50-100 individuals. These findings highlight that reliable simulation cannot be achieved by optimizing any single layer in isolation, but requires coordinated improvement across all three.

cs.CL↗

arXiv2026

Optical Images of the Braneworld Black Hole Surrounded by an Optically Thin Accretion Disk

This work examines the observational signatures of rotating black holes with tidal charge in the Randall--Sundrum braneworld scenario. Combining an elliptic-integral-based analytical treatment with numerical ray tracing, we characterize photon motion around braneworld black holes in detail. For small observer inclinations, secondary images remain embedded inside the primary emission rings. As the inclination becomes larger, the primary and secondary images gradually separate and produce a strongly asymmetric image morphology. We show that the image asymmetry and the deformation of the inner shadow are jointly controlled by the black hole spin, the tidal charge, and the observer inclination. To analyze the frequency shifts across the accretion disk, we extend the emitting region from the ISCO down to the event horizon by including the plunging flow. The results indicate that the observer inclination is the dominant factor governing the frequency-shift distribution. In addition, we reconstruct braneworld black hole images using a fisheye-lens ray-tracing model. The optical morphology and brightness distribution show a clear dependence on the observing frequency, especially when comparing 230~GHz with 86~GHz. We also contrast the brightness distributions of prograde and retrograde disks, finding that both the total intensity and the peak intensity at 86~GHz are higher than those at 230~GHz. For negative values of the tidal charge, we further investigate the corresponding frequency-shift behavior and 230~GHz intensity profiles, which may provide useful theoretical guidance for future studies of extra-dimensional gravity models.

astro-ph.HE↗

arXiv2026

A Missing Tool for Calculating Auto/Cross-correlation Function under Nonuniform Sampling Observations

Nonuniform sampling presents a long-standing challenge in astrophysical time-domain analysis, invalidating the standard autocorrelation and cross-correlation functions and forcing researchers to adopt ad-hoc methods like interpolation or binning, which introduce unquantified biases and lack rigorous error estimation. Here we introduce a new method for calculating the nonuniform autocorrelation function (NUACF) and nonuniform cross-correlation function (NUCCF) for irregularly sampled time series. Instead of relying on interpolation, it naturally evaluates the correlation function by incorporating time-interval weights and misalignment penalties. Monte Carlo simulations provide confidence bands for significance assessment and a complete error budget for the time delays that accounts for both flux uncertainties and sampling irregularity (essential but generally absent from existing methods). Through extensive simulations, we demonstrate our method outperforms traditional methods across various conditions, from strictly periodic to complex repeating variability patterns (e.g., intermittent but aperiodic). Its effectiveness is demonstrated via various real astrophysical data sets, revealing repetitive variability in stellar light curves, measuring time delays for multi-band disc reverberation in the AGN Fairall 9, and providing model-independent validation of time delays for the gravitationally lensed quasar HE 0435-1223. The method provides a rigorous and general solution to the ubiquitous problem of nonuniform sampling, positioning it as a useful tool for large-scale time-domain survey data analysis. The framework is also directly applicable to emerging time-domain phenomena such as fast radio bursts (FRBs), enabling, e.g., the study of correlations between persistent radio source luminosity and repeating FRB activity, or among the multi-parameter variability curves of FRB emission itself.

astro-ph.IM↗

arXiv2026

Constraints on the Low-frequency Radio Emission of the Galactic FRB Source SGR 1935+2154

We present a search for radio pulses from the Galactic magnetar SGR 1935+2154, a well-known source of fast radio bursts (FRBs), at $\sim$110 MHz using the Large Phased Array (LPA) of the Pushchino Radio Astronomy Observatory. Data from two active periods in 2020 (March -- May and September -- November, with $\sim 3.5$ minutes of daily coverage) were analyzed with new methods tailored to both FRB-like single pulses and pulsar-like periodic signals. No significant FRB-like pulses were found. Using Monte Carlo simulations, $3σ$ upper limits were derived for the burst rate: for a log-normal energy distribution the limit is $\sim$${10}^{1.5}~{\rm{d}}^{-1}$ for a mean of average monochromatic isotropic luminosity $L_{ν{\rm ,mean}}\sim1.3\times{10}^{29}~{\rm{erg~s^{-1}~ {Hz}^{-1}}}$ and a natural log-space scatter of $σ\sim0.85$; while for a power-law distribution it is $\sim$${10}^{1.8}~{\rm{d}}^{-1}$ for an index $β\lesssim3.0$ and a minimum average monochromatic isotropic luminosity $L_{ν{\rm{,min}}}\lesssim0.7\times{10}^{25}~{\rm{erg~s^{-1}~{Hz}^{-1}}}$. When folded at the known 3.24781628 s period of SGR 1935+2154, a weak pulse was noted (S/N $<$ 3.16), but the significance is insufficient for a secure detection of the pulsar-like emission signal. A conservative upper limit on the average monochromatic isotropic luminosity of any possible periodic emission is $2.08\times{10}^{19}~{\rm{erg~s^{-1}~{Hz}^{-1}}}$. Our results offer meaningful low-frequency upper limits on the burst rate of SGR 1935+2154, and hint for very faint pulsar-like radiation at meter wavelengths.

astro-ph.HE↗

arXiv2026

Decoding FRB Energetics and Frequency Features Hidden by Observational Incompleteness

Fast radio bursts (FRBs) are millisecond-duration extragalactic radio flashes likely powered by magnetars, yet their radiation mechanism remains unknown. Limited sensitivity and finite observing bandwidth inevitably lead to observational truncation, biasing our understanding of intrinsic burst properties. Assuming Gaussian-like spectra, we present a general inverse-modeling framework that reconstructs the intrinsic frequency and energy characteristics of repeating FRBs directly from truncated data, without spectral profile fitting. In our approach, detected bursts are classified as in-band (affected only by the sensitivity cutoff) or band-chipped (affected by both sensitivity and operating-band cutoffs) events. For in-band events, observed and intrinsic quantities are linked through a set of equations. For band-chipped bursts, with spectral peaks possibly outside the telescope's operating band, a population-based method is used to infer individual burst properties from the statistical properties of the entire sample. Applied to 2,223 bursts from FRB 20121102A, it is found that intrinsically energetic bursts tend to have narrower spectra than weak ones. We further quantify, for the first time, the number of out-of-band bursts, and reveal distinct frequency-evolution behaviors across active periods and frequency bands. Comparisons between reconstructed and original samples show that the sensitivity cutoff barely affects burst energy but biases the observed bandwidth, whereas the operating-band cutoff may cause severe energy leakage and bandwidth underestimation, suggesting that the energy release of some repeaters may be underestimated, with potential implications for the energy supply beyond the magnetar magnetosphere. Our methodology transforms incomplete archival observations into physically meaningful probes, bridging instrumental readouts and intrinsic FRB physics.

astro-ph.HE↗

arXiv2026

A Thermodynamically Consistent High-Order Framework for Staggered Lagrangian Hydrodynamics

We present a consistent high-order staggered Lagrangian hydrodynamics framework designed to reconcile an underlying disparity in existing curvilinear formulations: the mismatch between quadrature-based "strong" mass conservation and the discrete degrees of freedom (DOFs) of thermodynamic variables. By mathematically coupling the numerical quadrature rule with the density representation, our approach ensures rigorous point-wise consistency between density, internal energy, and pressure. This synchronization eliminates the ambiguity of equation-of-state (EOS) updates inherent in previous high-order staggered methods. To stabilize the discretization, we develop a high-order generalization of the subzonal pressure method by conceptually enriching the pressure field from the $Q^{m-1}$ to the $Q^m$ finite element space. We prove that evaluating this enriched field using a high-order quadrature rule naturally generates a restorative anti-hourglass force, which exactly recovers the classical $Q^1-P^0$ compatible hydrodynamics algorithm as a limiting case for $m=1$. Furthermore, we introduce a concise, algorithmic formulation of tensor artificial viscosity that streamlines implementation and significantly reduces computational overhead in high-order settings. The resulting framework yields strictly diagonal mass matrices for both momentum and energy equations, enabling highly efficient, fully explicit time integration without global linear solves. Extensive numerical benchmarks, including smooth convergence tests and complex shock-dominated flows, demonstrate that the proposed method achieves optimal high-order accuracy while maintaining superior geometric robustness.

math.NA↗

arXiv2026

Statistical Symmetry Breaking and Emergent Colored Noise in a Stochastic Scalar-Doublet Field Theory

We investigate a relativistic stochastic field theory in which a complex scalar doublet is coupled to a complex white-noise source. The action preserves Lorentz and $\mathrm{U(1)}\times\mathrm{SU(2)}$ symmetries at the statistical level, whereas the corresponding Euler-Lagrange equations exhibit symmetry breaking along individual stochastic realizations. Within a gauge-field-free sector introduced to obtain analytical solutions, we show that the scalar doublet undergoes a noise-driven random walk in field space, leading to a finite, time-dependent ensemble average of its magnitude. As an illustrative application, we further investigate the coupling of the stochastic scalar field to fermions through a Yukawa interaction. The scalar-field solution naturally separates into a tail component, which contributes as an effective mass-like term, and a light-cone component, which acts as a colored-noise source that induces a spatially correlated stochastic phase in the fermion wave function. The statistical properties and correlation length of this emergent colored noise are derived analytically within the adopted approximations. The present work provides an exploratory study of statistical symmetry breaking and emergent colored-noise dynamics in a relativistic stochastic scalar-doublet field theory.

hep-ph↗

arXiv2026

Fast radio bursts, magnetars and earthquakes: their "family feud"?

Fast radio bursts (FRBs) are millisecond-duration cosmic transients whose origin remains elusive. Competing models invoke either earthquake-like processes or flare-like mechanisms. To discriminate between these scenarios, we develop a novel diagnostic, the Pincus-Lyapunov diagram (PLD), to characterize the energetic transients in the stochasticity-chaos phase space. We compile burst sequences from five representative FRBs (FRB 20121102A, FRB 20190520B, FRB 20201124A, FRB 20220912A, and FRB 20240114A), together with those from magnetar flares (SGR J1550$-$5418, SGR J0501+4516, SGR 1806$-$20, SGR 1900+14, and SGR J1935+2154), pulsar glitches, solar flares, and earthquakes, and map them onto the PLD for comparative analysis. The resulting diagram shows that FRBs occupy a distinct region of the phase space. Specifically, a permutation test reveals a statistically significant difference in the distributions of magnetar flares and pulsar glitches compared to those of repeating FRBs ($p$-value $\simeq 0.05$). To examine whether temporal variations in source activity can shift a repeater's position in this phase space, we analyze the time evolution of the most prolific repeater, FRB~20240114A. For this repeating FRB, both Pincus Index and Lyapunov Exponent demonstrate statistically stable behaviour over the eight-month observation session, with Augmented Dickey--Fuller tests yielding $p \simeq 1.78\times10^{-3}$ and $9.91\times10^{-3}$, respectively. By assembling the most comprehensive dataset to date, our work indicates that the trigger mechanisms of repeating FRBs are likely to be distinct from those driving magnetar flares, pulsar glitches, solar flares, and earthquakes.

astro-ph.HE↗

arXiv2026

Timing, Polarization, and Single-Pulse Properties of Long-Period FAST Pulsars

We present phase-connected timing and polarization measurements for two long-period FAST-CRAFTS pulsars, PSRs J0000+6252 and J2131+3642, and extend single-pulse emission-state analysis to a five-source sample including PSRs J1903+1407, J1502+4653, and J2112+4058. FAST L-band timing baselines span 414 to 511 days; the two pulsars have spin periods of 1.11 to 1.55 s, period derivatives of $(3.99$ to $4.06)\times10^{-15}~{\rm s~s^{-1}}$, characteristic ages of 4.34 to 6.16 Myr, surface magnetic fields of $(2.15$ to $2.52)\times10^{12}$ G, and spin-down luminosities of $(4.21\times10^{31}$ to $1.16\times10^{32})~{\rm erg~s^{-1}}$. Their rotation measures are $70.2\pm26.7$ and $-44.6\pm7.6~{\rm rad~m^{-2}}$, with linear polarization fractions of 17.6\% to 27.2\%. PSR J2131+3642 shows a short monotonic PA segment permitting a formal rotating-vector-model fit, though limited longitude coverage leaves the geometric parameters poorly constrained; PSR J0000+6252 has too few PA points for such a fit. Gaussian mixture modeling (GMM) of single-pulse energy distributions identifies null, weak, and burst components in PSRs J0000+6252, J1903+1407, J1502+4653, and J2112+4058, while J2131+3642 shows only weak and burst states. We also identify bright single pulses (peak intensity $\geq10\times$ the integrated average profile): 42 in J0000+6252, four each in J1903+1407 and J2112+4058, and none in J2131+3642 or J1502+4653. These bright pulses occur within the main emission window with no evidence of periodic recurrence, consistent with sporadic enhancements of the normal radio-emission beam. For J2112+4058, we measure a scattering timescale $τ_{\rm sc}=5.84\pm0.18$ ms at $ν_{\rm ref}=1.25$ GHz. Together, these results highlight the diversity of magnetospheric variability among slowly rotating neutron stars.

astro-ph.HE↗

arXiv2026

Rethinking the Evaluation and Optimization of LLM-Based Social Simulation

LLM-based social simulation is a promising complement to traditional methods such as surveys and behavioral experiments. A core question is how to evaluate the fidelity of LLM-simulated human behavior and optimize LLMs toward it. Prevailing practice evaluates by accuracy, checking whether the model selects the single response observed from a human, and trains the LLM to reproduce this hard label. However, human behavior is inherently subjective: the same person in the same situation may reasonably act differently, so an observed response is only one draw from an underlying response distribution, rendering accuracy-based evaluation unreliable and hard-label training misleading. To address these problems, we first introduce the subjectivity coefficient, an entropy-based quantity distinguishing objective tasks such as coding from subjective ones such as social simulation, and use it to systematically analyze how accuracy-based evaluation and hard-label training fail as subjectivity grows. Based on the subjectivity coefficient, we propose Subjectivity-Adaptive soft-Label Training (SALT): it pools observed outputs from semantically nearby inputs into soft distributional labels, with an aggregation radius adapted to the estimated subjectivity of each input; in the near-objective limit the neighborhood shrinks, so SALT naturally falls back to standard single-label training. Moreover, since existing datasets record only single observed responses and cannot support distributional evaluation, we construct SUBJSIM, a benchmark of 19,300 contexts covering 193 annotators and 100 subjective questions. Since real-world data typically provide only a single observation per input, our experiments train models from single observed outputs while evaluating them against the full response distributions, verifying feasibility in realistic settings. Results on SUBJSIM demonstrate the advantages of our method.

cs.AI↗

arXiv2026

Indication for Decreasing Dispersion Measure in the Population of Repeating Fast Radio Bursts and Connection to Young Supernova Remnant Expansion

Fast Radio Bursts (FRBs) are millisecond-duration, highly energetic radio transients of uncertain origin. Repeating FRBs provide an excellent population for investigating their nature, particularly through studies of parameter evolution. Out of the 63 repeaters monitored by CHIME, we select the 19 sources with more than 10 detected bursts, and examine their long-term dispersion measure (DM) evolution. Seven sources show statistically significant DM evolution and are classified as the golden sample. Of these, five exhibit a decreasing DM trend and two show an increasing trend. We then perform a binomial test under the null hypothesis that decreasing and increasing DM variation trends have equal probabilities. The current combined sample, including our golden sample and additional repeaters with reported DM change rate from the literature, gives a p-value of 0.033, supporting that decreasing DM trends are more common in the repeating FRB population. This statistical result is consistent with scenarios in that the local electron density around repeaters generally decreases with time, for example, due to expansion of a young supernova remnant (SNR). Finally, within the SNR expansion model, we provide an illustrative estimate of the SNR contributions to the DM for different ejecta masses.

astro-ph.HE↗

arXiv2026

A Helium-shell Burning Blue Horizontal Branch Star Produced from Common Envelope Evolution

Observationally, blue horizontal branch (BHB) stars are defined as hot stars occupying a characteristic region between the extreme blue horizontal branch and RR Lyrae variables in the Hertzsprung-Russell diagram. Most of them are interpreted as stripped core-helium-burning stars, but the role of binary interaction in their formation remains unclear. Here, we report the discovery of a metal-rich BHB star in a 0.82628-day binary system (Feige 64) comprising a $0.35\pm0.03\,M_{\odot}$ BHB star and a likely $1.26\pm0.17\,M_{\odot}$ white dwarf (WD). The BHB star has an effective temperature of $15{,}524\pm310$ K and a luminosity of $39.7\pm4.1\,L_{\odot}$. Stellar evolution modelling indicates that it is a helium-shell-burning star produced through the common-envelope channel, retaining a hydrogen-rich envelope that is more massive than previously thought for low-mass stars. This finding provides direct evidence for binary interaction in the formation of BHB stars, offering a fresh perspective on interpreting this emerging population.

astro-ph.SR↗

arXiv2026

Analyzing and Correcting Benevolence Bias in Large Language Models

Large language models (LLMs) are increasingly used as stand-ins for human respondents, from opinion polls and simulated survey participants to agent-based social simulations. These uses rest on one assumption: that conditioning a model on who a person is yields answers resembling those of real people from that group. Here we identify and measure benevolence bias, a small but consistent tendency for aligned LLMs to lean toward the kinder, safer, more socially approved answer on value-laden survey questions. Across 18 widely used models, four social-science datasets (ANES, GSS, WVS, and a cross-cultural prospect-theory replication) and six psychological categories, we find that the bias is a stable model property, not a quirk of any one system: it points the same way across models, grows with model size, and traces to the post-training stage. Prompt language and framing change its size but never its direction, and a "malicious persona" stress test shows a one-sided limit: aligned models struggle to play people who are less kind, less prosocial or more harm-tolerant than average. The issue is thus not only a shifted average, but a narrowed range of people the model can imitate. The bias sits in the middle of the answer distribution rather than its tails, and survives changes in sampling temperature and simple prompted reflection. The encouraging news is that it is easy to diagnose and straightforward to fix: a light-touch contrastive calibration, which needs no retraining and works on black-box APIs, brings all six categories back to the human baseline. Our results give researchers a clear map of where aligned LLMs can already be trusted as human stand-ins, where they need care, and a ready-to-use method for closing the gap.

cs.HC↗