arXiv subjects

Jeet Sharma

Publications and source records attributed to Jeet Sharma.

3 recordsLinked to original sources

arXiv2026

UNSPECIFIC: General Constraint Synthesis for Breaking Copy-and-Paste Shortcut in LLM Instruction Following

Large language models (LLMs) are increasingly expected to follow long lists of constraints in complex instructions, and synthesizing instructions from a reference document (i.e., back-translation) is a widely used method to measure/enhance LLMs' ability to follow complex instructions. However, this method introduces a critical loophole: the constraint synthesis model copies text from the reference as a very specific constraint and the evaluated LLM trivially satisfies the constraint by copying its text in the response. To address these issues, we propose UNSPECIFIC, a novel framework that synthesizes constraints common to two similar reference articles to reduce copy-pasting, selectively hardens only trivially satisfied constraints to balance difficulty and naturalness, and evaluates satisfaction on both the generated article and its summary to penalize superficial instruction following. Consequently, we built the UNSPECIFIC benchmark on news, story, and blog domains to analyze the copy-pasting behavior of LLMs. Our results show that our synthesized constraints are not only more challenging (e.g., the satisfaction rate of GPT-5 Mini drops from 90% to 78%) and natural (LLM win-rate gap improves by 30%) from a human perspective but also mitigate the copy-pasting. We also find that a large portion of constraints are satisfied superficially (i.e., not satisfied in the core narrative of the article). The code and datasets are released at https://github.com/JeetDSharma/UNSPECIFIC.

cs.CL↗

arXiv2026

Physically motivated iso-orbital indicator for meta-GGA exchange functionals

The iso-orbital indicator $\alpha = (\tau - \tau^\mathrm{vW})/\tau^\mathrm{UEG}$ is a key ingredient of meta-generalized gradient approximation (meta-GGA) functionals, but diverges in low-density tails , causing unphysical exchange potentials and systematic band gap errors as noted in [J. Chem. Phys. 150, 161101 (2019)]. We replace the denominator of $\alpha$ with a physically motivated Pauli KED drawn from the orbital-free DFT literature, eliminating the divergence in the low density atomic tail without any empirical regularization parameter. Testing two such enhancement factors: LKT and PGS, within the r$^2$SCAN and MS2 exchange functionals, we find that the modified indicators suppress spurious oscillations in the semilocal exchange potential and restore correct electron localization in atomic tails. For a ten-member cubic semiconductor benchmark, the band gap mean absolute error is reduced by 41.1 % for r$^2$SCAN@PGS and 48.8 % for MS2@PGS, while cohesive energy accuracy is largely preserved. The consistent improvement across two functionals with distinct constructions confirms a physical rather than functional specific origin, and motivates further development of meta-GGA functionals with constraint satisfying iso-orbital indicators.

cond-mat.mtrl-sci↗

arXiv2025

Krylov complexity in ergodically constrained nonintegrable transverse-field Ising model

The nonintegrable transverse-field Ising model is a common platform for studying ergodic quantum dynamics. In this work, we introduce a simple variant of the model in which this ergodic behaviour is suppressed by introducing a spatial inhomogeneity in the interaction strengths. For this we partition the chain into two equal segments within which the spins interact with different coupling strengths. The ratio of these couplings defines an inhomogeneity parameter, whose variation away from unity leads to constrained dynamics. We characterize this crossover using multiple diagnostics, such as the long-time saturation of out-of-time-ordered correlators, level-spacing statistics, and the spectral form factor. We further examine the consequences for operator growth in Krylov space and for entanglement generation in the system's eigenstates. Together, these results demonstrate that introducing a macroscopic inhomogeneity in coupling strengths provides a minimal, disorder-free route to breaking ergodicity in this specific model of interacting spins.

quant-ph↗