arXiv subjects

Gautam Gupta

Publications and source records attributed to Gautam Gupta.

9 recordsLinked to original sources

arXiv2026

Better and Worse with Scale: How Contextual Entrainment Diverges with Model Size

Larger language models become simultaneously better and worse at handling contextual information -- better at ignoring false claims, worse at ignoring irrelevant tokens. We formalize this apparent paradox through the first scaling laws for contextual entrainment, the tendency of models to favor tokens that appeared in context regardless of relevance. Analyzing the Cerebras-GPT (111M-13B) and Pythia (410M-12B) model families, we find entrainment follows predictable power-law scaling, but with opposite trends depending on context type: semantic contexts show decreasing entrainment with scale, while non-semantic contexts show increasing entrainment. Concretely, the largest models are four times more resistant to counterfactual misinformation than the smallest, yet simultaneously twice as prone to copying arbitrary tokens. These diverging trends, which replicate across model families, suggest that semantic filtering and mechanical copying are functionally distinct behaviors that scale in opposition -- scaling alone does not resolve context sensitivity, it reshapes it.

cs.CL↗

arXiv2026

MAC: Multi-Agent Constitution Learning

Constitutional AI is a method to oversee and control LLMs based on a set of rules written in natural language. These rules are typically written by human experts, but could in principle be learned automatically given sufficient training data for the desired behavior. Existing LLM-based prompt optimizers attempt this but are ineffective at learning constitutions since (i) they require many labeled examples and (ii) lack structure in the optimized prompts, leading to diminishing improvements as prompt size grows. To address these limitations, we propose Multi-Agent Constitutional Learning (MAC), which optimizes over structured prompts represented as sets of rules using a network of agents with specialized tasks to accept, edit, or reject rule updates. We also present MAC+, which improves performance by training agents on successful trajectories to reinforce updates leading to higher reward. We evaluate MAC on tagging Personally Identifiable Information (PII), a classification task with limited labels where interpretability is critical, and demonstrate that it generalizes to other agentic tasks such as tool calling. MAC outperforms recent prompt optimization methods by over 50%, produces human-readable and auditable rule sets, and achieves performance comparable to supervised fine-tuning and GRPO without requiring parameter updates.

cs.AI↗

arXiv2025

SemBench: A Benchmark for Semantic Query Processing Engines

We present a benchmark targeting a novel class of systems: semantic query processing engines. Those systems rely inherently on generative and reasoning capabilities of state-of-the-art large language models (LLMs). They extend SQL with semantic operators, configured by natural language instructions, that are evaluated via LLMs and enable users to perform various operations on multimodal data. Our benchmark introduces diversity across three key dimensions: scenarios, modalities, and operators. Included are scenarios ranging from movie review analysis to car damage detection. Within these scenarios, we cover different data modalities, including images, audio, and text. Finally, the queries involve a diverse set of operators, including semantic filters, joins, mappings, ranking, and classification operators. We evaluated our benchmark on three academic systems (LOTUS, Palimpzest, and ThalamusDB) and one industrial system, Google BigQuery. Although these results reflect a snapshot of systems under continuous development, our study offers crucial insights into their current strengths and weaknesses, illuminating promising directions for future research.

cs.DB↗

arXiv2017

Active bialkali photocathodes on free-standing graphene substrates

The hexagonal structure of graphene gives rise to the property of gas impermeability, motivating its investigation for a new application: protection of semiconductor photocathodes in electron accelerators. These materials are extremely susceptible to degradation in efficiency through multiple mechanisms related to contamination from the local imperfect vacuum environment of the host photoinjector. Few-layer graphene has been predicted to permit a modified photoemission response of protected photocathode surfaces, and recent experiments of single-layer graphene on copper have begun to confirm these predictions for single crystal metallic photocathodes. Unlike metallic photoemitters, the integration of an ultra-thin graphene barrier film with conventional semiconductor photocathode growth processes is not straightforward. A first step toward addressing this challenge is the growth and characterization of technologically relevant, high quantum efficiency bialkali photocathodes grown on ultra-thin free-standing graphene substrates. Photocathode growth on free-standing graphene provides the opportunity to integrate these two materials and study their interaction. Specifically, spectral response features and photoemission stability of cathodes grown on graphene substrates are compared to those deposited on established substrates. In addition we observed an increase of work function for the graphene encapsulated bialkali photocathode surfaces, which is predicted by our calculations. The results provide a unique demonstration of bialkali photocathodes on free-standing substrates, and indicate promise towards our goal of fabricating high-performance graphene encapsulated photocathodes with enhanced lifetime for accelerator applications.

cond-mat.mtrl-sci↗

arXiv2016

Valence band electronic structure evolution of graphene oxide upon thermal annealing for optoelectronics

We report valence band electronic structure evolution of graphene oxide (GO) upon its thermal reduction. Degree of oxygen functionalization was controlled by annealing temperatures, and an electronic structure evolution was monitored using real-time ultraviolet photoelectron spectroscopy. We observed a drastic increase in density of states around the Fermi level upon thermal annealing at ~600 oC. The result indicates that while there is an apparent band gap for GO prior to a thermal reduction, the gap closes after an annealing around that temperature. This trend of band gap closure was correlated with electrical, chemical, and structural properties to determine a set of GO material properties that is optimal for optoelectronics. The results revealed that annealing at a temperature of ~500 oC leads to the desired properties, demonstrated by a uniform and an order of magnitude enhanced photocurrent map of an individual GO sheet compared to as-synthesized counterpart.

cond-mat.mtrl-sci↗

arXiv2015

Chemical Vapor Deposition Synthesized Atomically Thin Molybdenum Disulfide with Optoelectronic-Grade Crystalline Quality

The ability to synthesize high-quality samples over large areas and at low cost is one of the biggest challenges during the developmental stage of any novel material. While chemical vapor deposition (CVD) methods provide a promising low-cost route for CMOS compatible, large-scale growth of materials, it often falls short of the high-quality demands in nanoelectronics and optoelectronics. We present large-scale CVD synthesis of single- and few- layered MoS2 using direct vapor-phase sulfurization of MoO2, which enables us to obtain extremely high-quality single-crystal monolayer MoS2 samples with field-effect mobility exceeding 30 cm2/Vs in monolayers. These samples can be readily synthesized on a variety of substrates, and demonstrate a high-degree of optoelectronic uniformity in Raman and photoluminescence mapping over entire crystals with areas exceeding hundreds of square micrometers. Owing to their high crystalline quality, Raman spectroscopy on these samples reveal a range of multi-phonon processes through peaks with equal or better clarity compared to past reports on mechanically exfoliated samples. This enables us to investigate the layer thickness- and substrate-dependence of the extremely weak phonon processes at 285 cm-1 and 487 cm-1 in 2D MoS2. The ultra-high, optoelectronic-grade crystalline quality of these samples could be further established through photocurrent spectroscopy, which clearly reveal excitonic states at room temperature, a feat that has been previously demonstrated only on device fabricated with mechanically exfoliated that were artificially suspended across trenches. Our method reflects a big step in the development of atomically thin, 2D MoS2 for scalable, high-quality optoelectronics.

cond-mat.mtrl-sci↗

arXiv2014

Spatially Resolved Photo-Excited Charge Carrier Dynamics in Phase-Engineered Monolayer MoS2

A fundamental understanding of the intrinsic optoelectronic properties of atomically thin transition metal dichalcogenides (TMDs) is crucial for its integration into high performance semiconductor devices. Here, we investigate the transport properties of chemical vapor deposition (CVD) grown monolayer molybdenum disulfide (MoS2) under photo-excitation using correlated scanning photocurrent microscopy and photoluminescence imaging. We examined the effect of local phase transformation underneath the metal electrodes on the generation of photocurrent across the channel length with diffraction-limited spatial resolution. While maximum photocurrent generation occurs at the Schottky contacts of semiconducting (2H-phase) MoS2, after the metallic phase transformation (1T-phase), the photocurrent peak is observed towards the center of the device channel, suggesting a strong reduction of native Schottky barriers. Analysis using the bias and position dependence of the photocurrent indicates that the Schottky barrier heights are few meV for 1T- and ~200 meV for 2H-contacted devices. We also demonstrate that a reduction of native Schottky barriers in a 1T device enhances the photo responsivity by more than one order of magnitude, a crucial parameter in achieving high performance optoelectronic devices. The obtained results pave a pathway for the fundamental understanding of intrinsic optoelectronic properties of atomically thin TMDs where Ohmic contacts are necessary for achieving high efficiency devices with low power consumption.

cond-mat.mtrl-sci↗

arXiv2013

Reduced graphene oxide thin films as ultrabarriers for organic electronics

Encapsulation of electronic devices based on organic materials that are prone to degradation even under normal atmospheric conditions with hermetic barriers is crucial for increasing their lifetime. A challenge is to develop 'ultrabarriers' that are impermeable, flexible, and preferably transparent. Another important requirement is that they must be compatible with organic electronics fabrication schemes (i.e. must be solution processable, deposited at room temperature and be chemically inert). Here, we report lifetime increase of 1,300 hours for poly(3-hexylthiophene) (P3HT) films encapsulated by uniform and continuous thin (~10 nm) films of reduced graphene oxide (rGO). This level of protection against oxygen and water vapor diffusion is substantially better than conventional polymeric barriers such as CytopTM, which degrades after only 350 hours despite being 400 nm thick. Analysis using atomic force microscopy, x-ray photoelectron spectroscopy and high resolution transmission electron microscopy suggest that the superior oxygen gas/moisture barrier property of rGO is due to the close interlayer distance packing and absence of pinholes within the impermeable sheets. These material properties can be correlated to the enhanced lag time of 500 hours. Our results provide new insight for the design of high performance and solution processable transparent 'ultrabarriers' for a wide range of encapsulation applications.

cond-mat.mtrl-sci↗

arXiv2011

A New Route to Fluorescent SWNT/Silica Nanocomposites: Balancing Fluorescence Intensity and Environmental Sensitivity

We investigate the relationship between photoluminescence (PL) intensity and environmental sensitivity of surfactant-wrapped single walled carbon nanotubes (SWNTs). SWNTs were studied under a variety of conditions in suspension as well as encapsulated in silica nanocomposites, which were prepared by an efficient chemical vapor into liquids (CViL) sol-gel process. The dramatically improved silica encapsulation process described here has several advantages, including fast preparation and high SWNT loading concentration, over other encapsulation methods used to prepare fluorescent SWNT/silica nanocomposites. Further, addition of glycerol to SWNT suspensions prior to performing the CViL sol-gel process allows for the preparation of freestanding fluorescent silica xerogels, which to the best of our knowledge is the first report of such nanocomposites. Our spectroscopic data on SWNTs suspended in aqueous surfactants or encapsulated in silica show that achieving maximum PL intensity results in decreased sensitivity of SWNT emission response to changes imparted by the local environment. In addition, silica encapsulation can be used to "lock-in" a surfactant micelle structure surrounding SWNTs to minimize interactions between SWNTs and ions/small molecules. Ultimately, our work demonstrates that one should consider a balance between maximum PL intensity and the ability to sense environmental changes when designing new SWNT systems for future sensing applications.

cond-mat.mes-hall↗