Vamshi Krishna Bonagiri

Krishak Aneja, Manas Mittal, Anmol Goel et al.

Fine-tuning Large Language Models (LLMs) on benign narrow data can sometimes induce broad harmful behaviors, a vulnerability termed emergent misalignment (EM). While prior work links these failures to specific directions in the activation space, their relationship to the model's broader persona remains unexplored. We map the latent personality space of LLMs through established psychometric profiles like the Big Five, Dark Triad, and LLM-specific behaviors (e.g. evil, sycophancy), and show that the semantic geometry is highly stable across aligned models and their corrupted fine-tunes. Through causal interventions, we find that directions isolating social valence, such as the 'Evil' persona vector, and a Semantic Valence Vector (SVV) that we introduce, function as intrinsic guardrails: ablating them drives the misalignment rates above $40$%, while amplifying them suppresses the failure mode to less than $3$%. Leveraging the structural stability of the personality space, we show that vectors extracted $\textit{a priori}$ from an instruct-tuned model transfer zero-shot to successfully regulate EM in corrupted fine-tunes. Overall, our findings suggest that harmful fine-tuning does not overwrite a model's internal representation of personality, allowing conserved representations to serve as robust, cross-distribution guardrails.

Priyanshul Govil, Hemang Jain, Vamshi Krishna Bonagiri et al.

Large Language Models (LLMs) often inherit biases from the web data they are trained on, which contains stereotypes and prejudices. Current methods for evaluating and mitigating these biases rely on bias-benchmark datasets. These benchmarks measure bias by observing an LLM's behavior on biased statements. However, these statements lack contextual considerations of the situations they try to present. To address this, we introduce a contextual reliability framework, which evaluates model robustness to biased statements by considering the various contexts in which they may appear. We develop the Context-Oriented Bias Indicator and Assessment Score (COBIAS) to measure a biased statement's reliability in detecting bias, based on the variance in model behavior across different contexts. To evaluate the metric, we augmented 2,291 stereotyped statements from two existing benchmark datasets by adding contextual information. We show that COBIAS aligns with human judgment on the contextual reliability of biased statements (Spearman's $ρ= 0.65, p = 3.4 * 10^{-60}$) and can be used to create reliable benchmarks, which would assist bias mitigation works.

Vamshi Krishna Bonagiri, Ponnurangam Kumaragurum, Khanh Nguyen et al.

As Large Language Model (LLM) agents increasingly operate in complex environments with real-world consequences, their safety becomes critical. While uncertainty quantification is well-studied for single-turn tasks, multi-turn agentic scenarios with real-world tool access present unique challenges where uncertainties and ambiguities compound, leading to severe or catastrophic risks beyond traditional text generation failures. We propose using "quitting" as a simple yet effective behavioral mechanism for LLM agents to recognize and withdraw from situations where they lack confidence. Leveraging the ToolEmu framework, we conduct a systematic evaluation of quitting behavior across 12 state-of-the-art LLMs. Our results demonstrate a highly favorable safety-helpfulness trade-off: agents prompted to quit with explicit instructions improve safety by an average of +0.39 on a 0-3 scale across all models (+0.64 for proprietary models), while maintaining a negligible average decrease of -0.03 in helpfulness. Our analysis demonstrates that simply adding explicit quit instructions proves to be a highly effective safety mechanism that can immediately be deployed in existing agent systems, and establishes quitting as an effective first-line defense mechanism for autonomous agents in high-stakes applications.

Ishwar B Balappanawar, Vamshi Krishna Bonagiri, Anish R Joishy et al.

Large Language Models (LLMs) demonstrate impressive reasoning capabilities in familiar contexts, but struggle when the context conflicts with their parametric knowledge. To investigate this phenomenon, we introduce CounterLogic, a dataset containing 1,800 examples across 9 logical schemas, explicitly designed to evaluate logical reasoning through counterfactual (hypothetical knowledge-conflicting) scenarios. Our systematic evaluation of 11 LLMs across 6 different datasets reveals a consistent performance degradation, with accuracies dropping by 27% on average when reasoning through counterfactual information. We propose Self-Segregate, a prompting method enabling metacognitive awareness (explicitly identifying knowledge conflicts) before reasoning. Our method dramatically narrows the average performance gaps from 27% to just 11%, while significantly increasing the overall accuracy (+7.5%). We discuss the implications of these findings and draw parallels to human cognitive processes, particularly on how humans disambiguate conflicting information during reasoning tasks. Our findings offer practical insights for understanding and enhancing LLMs reasoning capabilities in real-world applications, especially where models must logically reason independently of their factual knowledge.

Prashant Kodali, Anmol Goel, Likhith Asapu et al.

Current computational approaches for analysing or generating code-mixed sentences do not explicitly model ``naturalness'' or ``acceptability'' of code-mixed sentences, but rely on training corpora to reflect distribution of acceptable code-mixed sentences. Modelling human judgement for the acceptability of code-mixed text can help in distinguishing natural code-mixed text and enable quality-controlled generation of code-mixed text. To this end, we construct Cline - a dataset containing human acceptability judgements for English-Hindi~(en-hi) code-mixed text. Cline is the largest of its kind with 16,642 sentences, consisting of samples sourced from two sources: synthetically generated code-mixed text and samples collected from online social media. Our analysis establishes that popular code-mixing metrics such as CMI, Number of Switch Points, Burstines, which are used to filter/curate/compare code-mixed corpora have low correlation with human acceptability judgements, underlining the necessity of our dataset. Experiments using Cline demonstrate that simple Multilayer Perceptron (MLP) models when trained solely using code-mixing metrics as features are outperformed by fine-tuned pre-trained Multilingual Large Language Models (MLLMs). Specifically, among Encoder models XLM-Roberta and Bernice outperform IndicBERT across different configurations. Among Encoder-Decoder models, mBART performs better than mT5, however Encoder-Decoder models are not able to outperform Encoder-only models. Decoder-only models perform the best when compared to all other MLLMS, with Llama 3.2 - 3B models outperforming similarly sized Qwen, Phi models. Comparison with zero and fewshot capabilitites of ChatGPT show that MLLMs fine-tuned on larger data outperform ChatGPT, providing scope for improvement in code-mixed tasks. Zero-shot transfer from En-Hi to En-Te acceptability judgments are better than random baselines.

Vamshi Krishna Bonagiri, Sreeram Vennam, Priyanshul Govil et al.

Despite recent advancements showcasing the impressive capabilities of Large Language Models (LLMs) in conversational systems, we show that even state-of-the-art LLMs are morally inconsistent in their generations, questioning their reliability (and trustworthiness in general). Prior works in LLM evaluation focus on developing ground-truth data to measure accuracy on specific tasks. However, for moral scenarios that often lack universally agreed-upon answers, consistency in model responses becomes crucial for their reliability. To address this issue, we propose an information-theoretic measure called Semantic Graph Entropy (SaGE), grounded in the concept of "Rules of Thumb" (RoTs) to measure a model's moral consistency. RoTs are abstract principles learned by a model and can help explain their decision-making strategies effectively. To this extent, we construct the Moral Consistency Corpus (MCC), containing 50K moral questions, responses to them by LLMs, and the RoTs that these models followed. Furthermore, to illustrate the generalizability of SaGE, we use it to investigate LLM consistency on two popular datasets -- TruthfulQA and HellaSwag. Our results reveal that task-accuracy and consistency are independent problems, and there is a dire need to investigate these issues further.

Vamshi Krishna Bonagiri, Sreeram Vennam, Manas Gaur et al.

A Large Language Model (LLM) is considered consistent if semantically equivalent prompts produce semantically equivalent responses. Despite recent advancements showcasing the impressive capabilities of LLMs in conversational systems, we show that even state-of-the-art LLMs are highly inconsistent in their generations, questioning their reliability. Prior research has tried to measure this with task-specific accuracy. However, this approach is unsuitable for moral scenarios, such as the trolley problem, with no "correct" answer. To address this issue, we propose a novel information-theoretic measure called Semantic Graph Entropy (SGE) to measure the consistency of an LLM in moral scenarios. We leverage "Rules of Thumb" (RoTs) to explain a model's decision-making strategies and further enhance our metric. Compared to existing consistency metrics, SGE correlates better with human judgments across five LLMs. In the future, we aim to investigate the root causes of LLM inconsistencies and propose improvements.