arXiv subjects

Rahul Gupta

Publications and source records attributed to Rahul Gupta.

At least 55 records · Page 3Linked to original sources

arXiv2025

Extremely luminous optical afterglow of an energetic gamma-ray burst GRB 230204B

Robotic telescope networks play an important role in capturing early and bright optical afterglows, providing critical insights into the energetics and emission mechanisms of GRBs. In this study, we analyze GRB 230204B, an exceptionally energetic and multi-pulsed long GRB, detected by the Fermi GBM and MAXI detectors, with an isotropic equivalent gamma-ray energy exceeding 10$^{54}$ erg. Time-resolved spectral analysis reveals a transition in the prompt emission from hard (sub-photospheric dominated) spectra during early pulses to softer (synchrotron radiation dominated) spectra in later pulses, indicative of a hybrid jet composition. We report the discovery and characterization of the optical afterglow using the MASTER and BOOTES robotic telescope networks, which enabled rapid follow-up observations starting at $\sim$1.3 ks post-burst. The optical luminosity at this time was exceptionally high, surpassing that of many other optically bright GRBs, such as GRB 990123, GRB 080319B, etc. This places the burst among the most luminous optical GRBs observed to date. Long-term radio observations extending to 335 days post-burst were conducted with the ATCA. Multi-wavelength modeling was conducted using an external ISM forward-shock top-hat jet model with \sw{afterglowpy}. The results reveal a narrow and highly collimated jet with a circumburst density of $n_{0} \sim$ 28.12 cm$^{-3}$, kinetic energy $E_{\rm K} \sim$ 4.18 $\times 10^{55}$ erg, and a relatively low value of $ε_{B}$ = 2.14 $\times 10^{-6}$, indicating shock-compression of magnetic field in the surrounding interstellar medium. We constrained a low radiative efficiency of $\sim$ 4.3 \%. This study highlights the indispensable contribution of robotic networks to early afterglow observations and advances our understanding of GRB 230204B unique characteristics and underlying jet physics.

astro-ph.HE↗

arXiv2025

Time-resolved spectro-polarimetric analysis of extremely bright GRB 230307A: Possible Evidence of evolution from photospheric to synchrotron dominated emission

The radiation mechanisms powering Gamma-ray bursts (GRBs) and their physical processes remain one of the unresolved questions in high-energy astrophysics. Spectro-polarimetric observations of exceptionally bright GRBs provide a powerful diagnostic tool to address these challenges. GRB 230307A, the second-brightest long-duration GRB ever detected, exhibits a rare association with a Kilonova, offering a unique and rare probe into the emission processes of GRBs originating from compact object mergers. We present a comprehensive time-averaged and time-resolved spectro-polarimetric analysis of GRB 230307A using joint observations from the $AstroSat$ Cadmium Zinc Telluride Imager (CZTI), the $Fermi$ Gamma-ray Burst Monitor (GBM) and $Konus$-Wind. Spectral analysis reveals a temporal evolution in the low-energy photon index, $α$, transitioning from a hard to a softer state over the burst duration. Time-averaged polarimetric measurements yield a low polarization fraction ($<$ 12.7 %), whereas time-resolved polarization analysis unveils a marked increase in polarization fractions ($>$ 49 %) in the later stages of the emission episode. This spectro-polarimetric evolution suggests a transition in the dominant radiative mechanism: the initial phase, possibly characterized by thermal-dominated photospheric emission (unpolarized or weakly polarized), gives way to a regime dominated by non-thermal synchrotron emission (highly polarized). This transition provides possible evidence for the evolving influence of magnetic fields in shaping the GRB emission process and jet dynamics.

astro-ph.HE↗

arXiv2025

From Narrow Unlearning to Emergent Misalignment: Causes, Consequences, and Containment in LLMs

Recent work has shown that fine-tuning on insecure code data can trigger an emergent misalignment (EMA) phenomenon, where models generate malicious responses even to prompts unrelated to the original insecure code-writing task. Such cross-domain generalization of harmful behavior underscores the need for a deeper understanding of the algorithms, tasks, and datasets that induce emergent misalignment. In this work, we extend this study by demonstrating that emergent misalignment can also arise from narrow refusal unlearning in specific domains. We perform refusal unlearning on Cybersecurity and Safety concept, and evaluate EMA by monitoring refusal scores across seven responsible AI (RAI) domains, Cybersecurity, Safety, Toxicity, Bias, Sensitive Content, Medical/Legal, and Privacy. Our work shows that narrow domain unlearning can yield compliance responses for the targeted concept, however, it may also propagate EMA to unrelated domains. Among the two intervened concepts, Cybersecurity and Safety, we find that the safety concept can have larger EMA impact, i.e, causing lower refusal scores, across other unrelated domains such as bias. We observe this effect consistently across two model families, Mistral-7b-0.3v, and Qwen-7b-2.5. Further, we show that refusal unlearning augmented with cross-entropy loss function on a small set of retain data from the affected domains can largely, if not fully, restore alignment across the impacted domains while having lower refusal rate on the concept we perform unlearning on. To investigate the underlying causes of EMA, we analyze concept entanglements at the representation level via concept vectors. Our analysis reveals that concepts with higher representation similarity in earlier layers are more susceptible to EMA after intervention when the refusal stream is altered through targeted refusal unlearning.

cs.LG↗

arXiv2025

VMDT: Decoding the Trustworthiness of Video Foundation Models

As foundation models become more sophisticated, ensuring their trustworthiness becomes increasingly critical; yet, unlike text and image, the video modality still lacks comprehensive trustworthiness benchmarks. We introduce VMDT (Video-Modal DecodingTrust), the first unified platform for evaluating text-to-video (T2V) and video-to-text (V2T) models across five key trustworthiness dimensions: safety, hallucination, fairness, privacy, and adversarial robustness. Through our extensive evaluation of 7 T2V models and 19 V2T models using VMDT, we uncover several significant insights. For instance, all open-source T2V models evaluated fail to recognize harmful queries and often generate harmful videos, while exhibiting higher levels of unfairness compared to image modality models. In V2T models, unfairness and privacy risks rise with scale, whereas hallucination and adversarial robustness improve -- though overall performance remains low. Uniquely, safety shows no correlation with model size, implying that factors other than scale govern current safety levels. Our findings highlight the urgent need for developing more robust and trustworthy video foundation models, and VMDT provides a systematic framework for measuring and tracking progress toward this goal. The code is available at https://sunblaze-ucb.github.io/VMDT-page/.

cs.CV↗

arXiv2025

Can LLMs Grasp Implicit Cultural Values? Benchmarking LLMs' Cultural Intelligence with CQ-Bench

Cultural Intelligence (CQ) refers to the ability to understand unfamiliar cultural contexts, a crucial skill for large language models (LLMs) to effectively engage with globally diverse users. Existing studies often focus on explicitly stated cultural norms, but fail to capture the subtle, implicit values that are common in daily conversation. To address this gap, we introduce CQBench, a benchmark specifically designed to assess LLMs' capability to infer implicit cultural values from natural conversational contexts. CQBench consists of multi character conversation based stories using values from the World Value Survey and the GlobalOpinions, with topics including ethical, religious, social, etc. Our automatic dataset construction pipeline integrates rigorous validation procedures (incorporation, consistency, and implicitness checks), achieving a 94.5% human model agreement in the final validation. To leverage CQBench data, we design three tasks of increasing complexity: attitude detection, value selection, and value extraction. These tasks evaluate whether models can detect attitude and recognize values embedded within natural dialogues rather than relying on explicit cultural knowledge. We find that while frontier models like o1 reach human level performance in value selection (0.809 F1), they still fall short in nuanced attitude detection (0.622 F1). Notably, finetuning a smaller LLaMA-3.2-3B on only 500 culturally rich examples improves performance by over 10%, even outperforming o3-mini in some cases. Using CQ-Bench, we provide insights into the current challenges in LLMs' CQ research and suggest practical pathways for enhancing LLMs' cross-cultural reasoning abilities.

cs.CL↗

arXiv2025

Assessing Visual Privacy Risks in Multimodal AI: A Novel Taxonomy-Grounded Evaluation of Vision-Language Models

Artificial Intelligence have profoundly transformed the technological landscape in recent years. Large Language Models (LLMs) have demonstrated impressive abilities in reasoning, text comprehension, contextual pattern recognition, and integrating language with visual understanding. While these advances offer significant benefits, they also reveal critical limitations in the models' ability to grasp the notion of privacy. There is hence substantial interest in determining if and how these models can understand and enforce privacy principles, particularly given the lack of supporting resources to test such a task. In this work, we address these challenges by examining how legal frameworks can inform the capabilities of these emerging technologies. To this end, we introduce a comprehensive, multi-level Visual Privacy Taxonomy that captures a wide range of privacy issues, designed to be scalable and adaptable to existing and future research needs. Furthermore, we evaluate the capabilities of several state-of-the-art Vision-Language Models (VLMs), revealing significant inconsistencies in their understanding of contextual privacy. Our work contributes both a foundational taxonomy for future research and a critical benchmark of current model limitations, demonstrating the urgent need for more robust, privacy-aware AI systems.

cs.CV↗

arXiv2025

D-REX: A Benchmark for Detecting Deceptive Reasoning in Large Language Models

The safety and alignment of Large Language Models (LLMs) are critical for their responsible deployment. Current evaluation methods predominantly focus on identifying and preventing overtly harmful outputs. However, they often fail to address a more insidious failure mode: models that produce benign-appearing outputs while operating on malicious or deceptive internal reasoning. This vulnerability, often triggered by sophisticated system prompt injections, allows models to bypass conventional safety filters, posing a significant, underexplored risk. To address this gap, we introduce the Deceptive Reasoning Exposure Suite (D-REX), a novel dataset designed to evaluate the discrepancy between a model's internal reasoning process and its final output. D-REX was constructed through a competitive red-teaming exercise where participants crafted adversarial system prompts to induce such deceptive behaviors. Each sample in D-REX contains the adversarial system prompt, an end-user's test query, the model's seemingly innocuous response, and, crucially, the model's internal chain-of-thought, which reveals the underlying malicious intent. Our benchmark facilitates a new, essential evaluation task: the detection of deceptive alignment. We demonstrate that D-REX presents a significant challenge for existing models and safety mechanisms, highlighting the urgent need for new techniques that scrutinize the internal processes of LLMs, not just their final outputs.

cs.CL↗

arXiv2025

Investigating Temporal Features in Swift GRB Afterglows: A Comparative Study of UVOT and XRT Data

This study presents a statistical analysis of optical light curves (LCs) of 200 UVOT-detected GRBs from 2005 to 2018. We have categorised these LCs based on their distinct morphological features, including early flares, bumps, breaks, plateaus, etc. Additionally, to compare features across different wavelengths, we have also included XRT LCs in our sample. The early observation capability of UVOT has allowed us to identify very early flares in 21 GRBs preceding the normal decay or bump, consistent with predictions of external reverse or internal shock. The decay indices of optical LCs following a simple power-law (PL) are shallower than corresponding X-ray LCs, indicative of a spectral break between two wavelengths. Not all LCs with PL decay align with the forward shock model and require additional components such as energy injection or a structured jet. Further, plateaus in the optical LCs are primarily consistent with energy injection from the central engine to the external medium. However, in four cases, plateaus followed by steep decay may have an internal origin. The optical luminosity observed during the plateau is tightly correlated with the break time, indicative of a magnetar as their possible central engine. For LCs with early bumps, the peak position, correlations between the parameters, and observed achromaticity allowed us to constrain their origin as the onset of afterglow, off-axis jet, late re-brightening, etc. In conclusion, the ensemble of observed features is explained through diverse physical mechanisms or emissions observed from different outflow locations and, in turn, diversity among possible progenitors.

astro-ph.HE↗

arXiv2025

Diagnosing Memorization in Chain-of-Thought Reasoning, One Token at a Time

Large Language Models (LLMs) perform well on reasoning benchmarks but often fail when inputs alter slightly, raising concerns about the extent to which their success relies on memorization. This issue is especially acute in Chain-of-Thought (CoT) reasoning, where spurious memorized patterns can trigger intermediate errors that cascade into incorrect final answers. We introduce STIM, a novel framework for Source-aware Token-level Identification of Memorization, which attributes each token in a reasoning chain to one of multiple memorization sources - local, mid-range, or long-range - based on their statistical co-occurrence with the token in the pretraining corpus. Our token-level analysis across tasks and distributional settings reveals that models rely more on memorization in complex or long-tail cases, and that local memorization is often the dominant driver of errors, leading to up to 67% of wrong tokens. We also show that memorization scores from STIM can be effective in predicting the wrong tokens in the wrong reasoning step. STIM offers a powerful tool for diagnosing and improving model reasoning and can generalize to other structured step-wise generation tasks.

cs.CL↗

arXiv2025

Search for the Optical Counterpart of Einstein Probe Discovered Fast X-ray Transients from Lulin Observatory

The launch of the Einstein Probe (EP) mission has revolutionized the detection and follow-up observations of fast X-ray transients (FXTs) by providing prompt and timely access to their precise localizations. In the first year of its operation, the EP-mission reports the discovery of 72 high signal-to-noise FXTs. Subjected to the visibility in the sky and weather conditions, we search for the optical counterparts of 42 EP-discovered FXTs from the Lulin Observatory. We successfully detect the optical counterparts of 12 FXTs, and five of those are first discovered by us from the Lulin Observatory. We find that the optical counterparts are generally faint ($r>20$\,mag) and decline rapidly ($>0.5$\,mag per day). We also find that 12 out of 42 FXTs show direct evidence of their association with Gamma-Ray Bursts (GRBs) through significant temporal and spatial overlapping. Furthermore, the luminosities and redshifts of FXTs with confirmed optical counterparts in our observations are fully consistent with the faintest end of the GRB population. However, the non-detection of any associated optical counterpart with a significant fraction of FXTs suggests that EP FXTs are likely a subset of so-called `dark FXTs', similar to `dark GRBs'. Additionally, the luminosities of {\bf two FXTs with confirmed redshifts} are also consistent with jetted tidal disruption events (TDEs). However, we find that the optical luminosities of FXTs differ significantly from typical supernova shock breakout or kilonova emissions. Thus, we conclude that a significant fraction of EP-discovered FXTs are associated with events having relativistic jets; either a GRB or a jetted TDE.

astro-ph.HE↗

arXiv2025

Amazon Nova AI Challenge -- Trusted AI: Advancing secure, AI-assisted software development

AI systems for software development are rapidly gaining prominence, yet significant challenges remain in ensuring their safety. To address this, Amazon launched the Trusted AI track of the Amazon Nova AI Challenge, a global competition among 10 university teams to drive advances in secure AI. In the challenge, five teams focus on developing automated red teaming bots, while the other five create safe AI assistants. This challenge provides teams with a unique platform to evaluate automated red-teaming and safety alignment methods through head-to-head adversarial tournaments where red teams have multi-turn conversations with the competing AI coding assistants to test their safety alignment. Along with this, the challenge provides teams with a feed of high quality annotated data to fuel iterative improvement. Throughout the challenge, teams developed state-of-the-art techniques, introducing novel approaches in reasoning-based safety alignment, robust model guardrails, multi-turn jail-breaking, and efficient probing of large language models (LLMs). To support these efforts, the Amazon Nova AI Challenge team made substantial scientific and engineering investments, including building a custom baseline coding specialist model for the challenge from scratch, developing a tournament orchestration service, and creating an evaluation harness. This paper outlines the advancements made by university teams and the Amazon Nova AI Challenge team in addressing the safety challenges of AI for software development, highlighting this collaborative effort to raise the bar for AI safety.

cs.AI↗

arXiv2025

Retrieval-Augmented Multi-Agent System for Rapid Statement of Work Generation

Drafting a Statement of Work (SOW) is a vital part of business and legal projects. It outlines key details like deliverables, timelines, responsibilities, and legal terms. However, creating these documents is often a slow and complex process. It usually involves multiple people, takes several days, and leaves room for errors or outdated content. This paper introduces a new AI-driven automation system that makes the entire SOW drafting process faster, easier, and more accurate. Instead of relying completely on humans, the system uses three intelligent components or 'agents' that each handle a part of the job. One agent writes the first draft, another checks if everything is legally correct, and the third agent formats the document and ensures everything is in order. Unlike basic online tools that just fill in templates, this system understands the meaning behind the content and customizes the SOW to match the needs of the project. It also checks legal compliance and formatting so that users can trust the result. The system was tested using real business examples. It was able to create a full SOW in under three minutes, compared to several hours or days using manual methods. It also performed well in accuracy and quality, showing that it can reduce legal risks and save a lot of time. This solution shows how artificial intelligence can be used to support legal and business professionals by taking care of routine work and helping them focus on more important decisions. It's a step toward making legal processes smarter, faster, and more reliable.

cs.MA↗

arXiv2025

Establishing Best Practices for Building Rigorous Agentic Benchmarks

Benchmarks are essential for quantitatively tracking progress in AI. As AI agents become increasingly capable, researchers and practitioners have introduced agentic benchmarks to evaluate agents on complex, real-world tasks. These benchmarks typically measure agent capabilities by evaluating task outcomes via specific reward designs. However, we show that many agentic benchmarks have issues in task setup or reward design. For example, SWE-bench Verified uses insufficient test cases, while TAU-bench counts empty responses as successful. Such issues can lead to under- or overestimation of agents' performance by up to 100% in relative terms. To make agentic evaluation rigorous, we introduce the Agentic Benchmark Checklist (ABC), a set of guidelines that we synthesized from our benchmark-building experience, a survey of best practices, and previously reported issues. When applied to CVE-Bench, a benchmark with a particularly complex evaluation design, ABC reduces the performance overestimation by 33%.

cs.AI↗

arXiv2025

Customize Multi-modal RAI Guardrails with Precedent-based predictions

A multi-modal guardrail must effectively filter image content based on user-defined policies, identifying material that may be hateful, reinforce harmful stereotypes, contain explicit material, or spread misinformation. Deploying such guardrails in real-world applications, however, poses significant challenges. Users often require varied and highly customizable policies and typically cannot provide abundant examples for each custom policy. Consequently, an ideal guardrail should be scalable to the multiple policies and adaptable to evolving user standards with minimal retraining. Existing fine-tuning methods typically condition predictions on pre-defined policies, restricting their generalizability to new policies or necessitating extensive retraining to adapt. Conversely, training-free methods struggle with limited context lengths, making it difficult to incorporate all the policies comprehensively. To overcome these limitations, we propose to condition model's judgment on "precedents", which are the reasoning processes of prior data points similar to the given input. By leveraging precedents instead of fixed policies, our approach greatly enhances the flexibility and adaptability of the guardrail. In this paper, we introduce a critique-revise mechanism for collecting high-quality precedents and two strategies that utilize precedents for robust prediction. Experimental results demonstrate that our approach outperforms previous methods across both few-shot and full-dataset scenarios and exhibits superior generalization to novel policies.

cs.LG↗

arXiv2025

An Intermediate Luminosity GRB 210210A: The early onset of the external forward shock in the X-ray?

We have analyzed the prompt and afterglow characteristics of the intermediate luminosity burst ``GRB 210210A". Our prompt emission analysis indicates that GRB 210210A is among the softest long GRBs detected by the Swift-BAT. The time-integrated prompt emission spectrum of GRB 210210A is aptly described by a power law function with an exponential cutoff. The spectral peak energy (E$_{p,z}$) in rest-frame and the E$_{\rm γ, iso}$ for this GRB marginally satisfy the 2$σ$ Amati correlation, a common feature observed in low/intermediate luminosity GRBs. Notably, an early bump is observed in the Swift-XRT light curve (a rare feature); the optical afterglow light curve, on the other hand, appears to follow a power law decay. However, due to the lack of sufficient early optical observations, we cannot completely rule out the possibility of an early bump in the optical light curve. For the bump observed in the early X-ray light curve, we calculated parameters such as peak time, rise time, decay time, and bulk Lorentz factor ($Γ_{0}$ $\sim$ 156), which perfectly satisfy the correlation between the parameters of the onset of the afterglow in GRBs. Both the optical and X-ray (including our observations) light curves exhibit a chromatic break in the late afterglow. Based on the prompt and afterglow parameters, we confirm that the intermediate luminosity GRB 210210A favors a collapsar scenario and is possibly powered by a magnetar.

astro-ph.HE↗

arXiv2025

Evaluating the Critical Risks of Amazon's Nova Premier under the Frontier Model Safety Framework

Nova Premier is Amazon's most capable multimodal foundation model and teacher for model distillation. It processes text, images, and video with a one-million-token context window, enabling analysis of large codebases, 400-page documents, and 90-minute videos in a single prompt. We present the first comprehensive evaluation of Nova Premier's critical risk profile under the Frontier Model Safety Framework. Evaluations target three high-risk domains -- Chemical, Biological, Radiological & Nuclear (CBRN), Offensive Cyber Operations, and Automated AI R&D -- and combine automated benchmarks, expert red-teaming, and uplift studies to determine whether the model exceeds release thresholds. We summarize our methodology and report core findings. Based on this evaluation, we find that Nova Premier is safe for public release as per our commitments made at the 2025 Paris AI Safety Summit. We will continue to enhance our safety evaluation and mitigation pipelines as new risks and capabilities associated with frontier models are identified.

cs.CR↗

arXiv2025

An End-to-End Assurance Framework for AI/ML Workloads in Datacenters

Modern machine learning workloads such as large language model training, fine-tuning jobs are highly distributed and span across hundreds of systems with multiple GPUs. Job completion time for these workloads is the artifact of the application, compute, network and storage performance. In case of failure or degraded performance it is imperative to understand the root cause and possible remediation for the problem for end-to-end assurance. This demo showcases SaaSbased observability and automated troubleshooting for AI/ML workload performance issues using cross-layer telemetry and logs (e.g., Application telemetry, Collective communication logs, GPU Health metrics, Network Flow Data, NIC ROCEv2 telemetry). Different use cases are demonstrated for end-to-end assurance such as Cross-layer Dependency Graph, Cross-layer Service Level Expectations, Automated Root Cause Analysis, GPU-toGPU application path tracing.

cs.NI↗

arXiv2025

Non-reciprocity in magnon mediated charge-spin-orbital current interconversion

In magnetic systems, angular momentum is carried by spin and orbital degrees of freedom. Nonlocal devices, comprising heavy-metal nanowires on magnetic insulators like yttrium iron garnet (YIG), enable angular momentum transport via magnons. These magnons are polarized by spin accumulation at the interface through the spin Hall effect (SHE) and detected via the inverse SHE (iSHE). The processes are generally reciprocal, as demonstrated by comparable efficiencies when reversing injector and detector roles. However, introducing Ru, which enables the orbital Hall effect (OHE), disrupts this reciprocity. In our system, magnons polarized through combined SHE and OHE and detected via iSHE are 35% more efficient than the reverse process. We attribute this nonreciprocity to nonzero spin vorticity, resulting from varying electron drift velocities across the Pt/Ru interface. This study highlights the potential of orbital transport mechanisms in influencing angular momentum transport and efficiency in nonlocal spintronic devices.

cond-mat.mtrl-sci↗