Generative AI is transforming business applications by enabling natural language interfaces and intelligent automation. However, the underlying large language models (LLMs) are evolving rapidly and so prompting them consistently is a challenge. This leads to inconsistent and unpredictable application behavior, undermining the reliability that businesses require for mission-critical workflows. In this paper, we introduce the concept of prompt migration as a systematic approach to stabilizing GenAI applications amid changing LLMs. Using the Tursio enterprise search application as a case study, we analyze the impact of successive GPT model upgrades, detail our migration framework including prompt redesign and a migration testbed, and demonstrate how these techniques restore application consistency. Our results show that structured prompt migration can fully recover the application reliability that was lost due to model drift. We conclude with practical lessons learned, emphasizing the need for prompt lifecycle management and robust testing to ensure dependable GenAI-powered business applications.

Large language models (LLMs) are rapidly being integrated into psychological research as research tools, evaluation targets, human simulators, and cognitive models. However, recent evidence reveals severe measurement unreliability: Personality assessments collapse under factor analysis, moral preferences reverse with punctuation changes, and theory-of-mind accuracy varies widely with trivial rephrasing. These "measurement phantoms"--statistical artifacts masquerading as psychological phenomena--threaten the validity of a growing body of research. Guided by the dual-validity framework that integrates psychometrics with causal inference, we present a six-stage workflow that scales validity requirements to research ambition--using LLMs to code text requires basic reliability and accuracy, while claims about psychological properties demand comprehensive construct validation. Researchers must (1) explicitly define their research goal and corresponding validity requirements, (2) develop and validate computational instruments through psychometric testing, (3) design experiments that control for computational confounds, (4) execute protocols with transparency, (5) analyze data using methods appropriate for non-independent observations, and (6) report findings within demonstrated boundaries and use results to refine theory. We illustrate the workflow through an example of model evaluation--"LLM selfhood"--showing how systematic validation can distinguish genuine computational phenomena from measurement artifacts. By establishing validated computational instruments and transparent practices, this workflow provides a path toward building a robust empirical foundation for AI psychology research.

Parameter-efficient fine-tuning (PEFT) has become a common method for fine-tuning large language models, where a base model can serve multiple users through PEFT module switching. To enhance user experience, base models require periodic updates. However, once updated, PEFT modules fine-tuned on previous versions often suffer substantial performance degradation on newer versions. Re-tuning these numerous modules to restore performance would incur significant computational costs. Through a comprehensive analysis of the changes that occur during base model updates, we uncover an interesting phenomenon: continual training primarily affects task-specific knowledge stored in Feed-Forward Networks (FFN), while having less impact on the task-specific pattern in the Attention mechanism. Based on these findings, we introduce Trans-PEFT, a novel approach that enhances the PEFT module by focusing on the task-specific pattern while reducing its dependence on certain knowledge in the base model. Further theoretical analysis supports our approach. Extensive experiments across 7 base models and 12 datasets demonstrate that Trans-PEFT trained modules can maintain performance on updated base models without re-tuning, significantly reducing maintenance overhead in real-world applications.

Reproducibility and reliability remain pressing challenges for generative AI systems whose behavior can drift with each model update or prompt revision. We introduce GPR-bench, a lightweight, extensible benchmark that operationalizes regression testing for general purpose use cases. GPR-bench couples an open, bilingual (English and Japanese) dataset covering eight task categories (e.g., text generation, code generation, and information retrieval) and 10 scenarios in each task categories (80 total test cases for each language) with an automated evaluation pipeline that employs "LLM-as-a-Judge" scoring of correctness and conciseness. Experiments across three recent model versions - gpt-4o-mini, o3-mini, and o4-mini - and two prompt configurations (default versus concise-writing instruction) reveal heterogeneous quality. Our results show that newer models generally improve correctness, but the differences are modest and not statistically significant, suggesting that GPR-bench may not be sufficiently challenging to differentiate between recent model versions. In contrast, the concise-writing instruction significantly enhances conciseness (+12.37 pp, Mann-Whitney U test: p < 0.001, effect size r = 0.2995) with minimal degradations on accuracy (-1.7 pp), demonstrating the effectiveness of prompt engineering. Released under the MIT License, GPR- bench lowers the barrier to initiating reproducibility monitoring and provides a foundation for community-driven extensions, while also raising important considerations about benchmark design for rapidly evolving language models.

Context: ChatGPT and other large language models (LLMs) are widely used across healthcare, business, economics, engineering, and software engineering (SE). Despite their popularity, concerns persist about their reliability, especially their error rates across domains and the software development lifecycle (SDLC). Objective: This study synthesizes and quantifies ChatGPT's reported error rates across major domains and SE tasks aligned with SDLC phases. It provides an evidence-based view of where ChatGPT excels, where it fails, and how reliability varies by task, domain, and model version (GPT-3.5, GPT-4, GPT-4-turbo, GPT-4o). Method: A Multivocal Literature Review (MLR) was conducted, gathering data from academic studies, reports, benchmarks, and grey literature up to 2025. Factual, reasoning, coding, and interpretive errors were considered. Data were grouped by domain and SE phase and visualized using boxplots to show error distributions. Results: Error rates vary across domains and versions. In healthcare, rates ranged from 8% to 83%. Business and economics saw error rates drop from ~50% with GPT-3.5 to 15-20% with GPT-4. Engineering tasks averaged 20-30%. Programming success reached 87.5%, though complex debugging still showed over 50% errors. In SE, requirements and design phases showed lower error rates (~5-20%), while coding, testing, and maintenance phases had higher variability (10-50%). Upgrades from GPT-3.5 to GPT-4 improved reliability. Conclusion: Despite improvements, ChatGPT still exhibits non-negligible error rates varying by domain, task, and SDLC phase. Full reliance without human oversight remains risky, especially in critical settings. Continuous evaluation and critical validation are essential to ensure reliability and trustworthiness.

Modern LLMs struggle with efficient updates, as each new pretrained model version requires repeating expensive alignment processes. This challenge also applies to domain- or language-specific models, where fine-tuning on specialized data must be redone for every new base model release. In this paper, we explore the transfer of fine-tuning updates between model versions. Specifically, we derive the diff vector from one source model version, which represents the weight changes from fine-tuning, and apply it to the base model of a different target version. Through empirical evaluations on various open-weight model versions, we show that transferring diff vectors can significantly improve the target base model, often achieving performance comparable to its fine-tuned counterpart. For example, reusing the fine-tuning updates from Llama 3.0 8B leads to an absolute accuracy improvement of 10.7% on GPQA over the base Llama 3.1 8B without additional training, surpassing Llama 3.1 8B Instruct. In a multilingual model development setting, we show that this approach can significantly increase performance on target-language tasks without retraining, achieving an absolute improvement of 4.7% and 15.5% on Global MMLU for Malagasy and Turkish, respectively, compared to Llama 3.1 8B Instruct. Our controlled experiments reveal that fine-tuning transfer is most effective when the source and target models are linearly connected in the parameter space. Additionally, we demonstrate that fine-tuning transfer offers a stronger and more computationally efficient starting point for further fine-tuning. Finally, we propose an iterative recycling-then-finetuning approach for continuous model development, which improves both efficiency and effectiveness. Our findings suggest that fine-tuning transfer is a viable strategy to reduce training costs while maintaining model performance.

The external evaluation of AI systems is increasingly recognised as a crucial approach for understanding their potential risks. However, facilitating external evaluation in practice faces significant challenges in balancing evaluators' need for system access with AI developers' privacy and security concerns. Additionally, evaluators have reason to protect their own privacy - for example, in order to maintain the integrity of held-out test sets. We refer to the challenge of ensuring both developers' and evaluators' privacy as one of providing mutual privacy. In this position paper, we argue that (i) addressing this mutual privacy challenge is essential for effective external evaluation of AI systems, and (ii) current methods for facilitating external evaluation inadequately address this challenge, particularly when it comes to preserving evaluators' privacy. In making these arguments, we formalise the mutual privacy problem; examine the privacy and access requirements of both model owners and evaluators; and explore potential solutions to this challenge, including through the application of cryptographic and hardware-based approaches.

As LLMs evolve, significant effort is spent on manually crafting prompts. While existing prompt optimization methods automate this process, they rely solely on learning from incorrect samples, leading to a sub-optimal performance. Additionally, an unexplored challenge in the literature is prompts effective for prior models may not perform well on newer versions or different languages. We propose the Learning from Contrastive Prompts (LCP) framework to address these gaps, enhancing both prompt optimization and adaptation. LCP employs contrastive learning to generate effective prompts by analyzing patterns in good and bad prompt examples. Our evaluation on the Big-Bench Hard dataset shows that LCP has a win rate of over 76% over existing methods in prompt optimization and demonstrates strong adaptability across different model versions, families, and languages. LCP offers a systematic approach to prompt engineering, reducing manual effort in deploying LLMs across varied contexts.

Large Language Models (LLMs) are frequently updated due to data or architecture changes to improve their performance. When updating models, developers often focus on increasing overall performance metrics with less emphasis on being compatible with previous model versions. However, users often build a mental model of the functionality and capabilities of a particular machine learning model they are interacting with. They have to adapt their mental model with every update -- a draining task that can lead to user dissatisfaction. In practice, fine-tuned downstream task adapters rely on pretrained LLM base models. When these base models are updated, these user-facing downstream task models experience instance regression or negative flips -- previously correct instances are now predicted incorrectly. This happens even when the downstream task training procedures remain identical. Our work aims to provide seamless model updates to a user in two ways. First, we provide evaluation metrics for a notion of compatibility to prior model versions, specifically for generative tasks but also applicable for discriminative tasks. We observe regression and inconsistencies between different model versions on a diverse set of tasks and model updates. Second, we propose a training strategy to minimize the number of inconsistencies in model updates, involving training of a compatibility model that can enhance task fine-tuned language models. We reduce negative flips -- instances where a prior model version was correct, but a new model incorrect -- by up to 40% from Llama 1 to Llama 2.

Federated Learning (FL) is a distributed machine learning paradigm that allows clients to train models on their data while preserving their privacy. FL algorithms, such as Federated Averaging (FedAvg) and its variants, have been shown to converge well in many scenarios. However, these methods require clients to upload their local updates to the server in a synchronous manner, which can be slow and unreliable in realistic FL settings. To address this issue, researchers have developed asynchronous FL methods that allow clients to continue training on their local data using a stale global model. However, most of these methods simply aggregate all of the received updates without considering their relative contributions, which can slow down convergence. In this paper, we propose a contribution-aware asynchronous FL method that takes into account the staleness and statistical heterogeneity of the received updates. Our method dynamically adjusts the contribution of each update based on these factors, which can speed up convergence compared to existing methods.