This generates two (possibly invalid) child molecules. If valid molecules exist, the we choose one of them randomly. Details of seven different ways of modifying a molecule are as follows. The atom_addition connects a new atom to a single atom. The atom_insertion puts an atom between two atoms.

The challenge of discovering new molecules with desired properties is crucial in domains like drug discovery and material design. Recent advances in deep learning-based generative methods have shown promise but face the issue of sample efficiency due to the computational expense of evaluating the reward function. This paper proposes a novel algorithm for sample-efficient molecular optimization by distilling a powerful genetic algorithm into deep generative policy using GFlowNets training, the off-policy method for amortized inference. This approach enables the deep generative policy to learn from domain knowledge, which has been explicitly integrated into the genetic algorithm. Our method achieves state-of-the-art performance in the official molecular optimization benchmark, significantly outperforming previous methods. It also demonstrates effectiveness in designing inhibitors against SARS-CoV-2 with substantially fewer reward calls.

Recent studies developed jailbreaking attacks, which construct jailbreaking prompts to fool LLMs into responding to harmful questions.Early-stage jailbreaking attacks require access to model internals or significant human efforts. More advanced attacks utilize genetic algorithms for automatic and black-box attacks.However, the random nature of genetic algorithms significantly limits the effectiveness of these attacks.In this paper, we propose RLbreaker, a black-box jailbreaking attack driven by deep reinforcement learning (DRL).We model jailbreaking as a search problem and design an RL agent to guide the search, which is more effective and has less randomness than stochastic search, such as genetic algorithms.Specifically, we design a customized DRL system for the jailbreaking problem, including a novel reward function and a customized proximal policy optimization (PPO) algorithm.Through extensive experiments, we demonstrate that RLbreaker is much more effective than existing jailbreaking attacks against six state-of-the-art (SOTA) LLMs. We also show that RLbreaker is robust against three SOTA defenses and its trained agents can transfer across different LLMs.We further validate the key design choices of RLbreaker via a comprehensive ablation study.

The reliance of organisations on computer networks is enabled by network programmability, which is typically achieved through Service Function Chaining. These chains virtualise network functions, link them, and programmatically embed them on networking infrastructure. Optimal embedding of Service Function Chains is an NP-hard problem, with three sub-problems, chain composition, virtual network function embedding, and link embedding, that have to be optimised simultaneously, rather than sequentially, for optimal results. Genetic Algorithms have been employed for this, but existing approaches either do not optimise all three sub-problems or do not optimise all three sub-problems simultaneously. We propose a Genetic Algorithm-based approach called GENESIS, which evolves three sine-function-activated Neural Networks, and funnels their output to a Gaussian distribution and an A* algorithm to optimise all three sub-problems simultaneously. We evaluate GENESIS on an emulator across 48 different data centre scenarios and compare its performance to two state-of-the-art Genetic Algorithms and one greedy algorithm. GENESIS produces an optimal solution for 100% of the scenarios, whereas the second-best method optimises only 71% of the scenarios. Moreover, GENESIS is the fastest among all Genetic Algorithms, averaging 15.84 minutes, compared to an average of 38.62 minutes for the second-best Genetic Algorithm.

This paper explores adaptive problem solving with a game designed to support the development of problem-solving skills. Using an adaptive, AI-powered puzzle game, our adaptive problem-solving system dynamically generates pathfinding-based puzzles using a genetic algorithm, tailoring the difficulty of each puzzle to individual players in an online real-time approach. A player-modeling system records user interactions and informs the generation of puzzles to approximate a target difficulty level based on various metrics of the player. By combining procedural content generation with online adaptive difficulty adjustment, the system aims to maintain engagement, mitigate frustration, and maintain an optimal level of challenge. A pilot user study investigates the effectiveness of this approach, comparing different types of adaptive difficulty systems and interpreting players' responses. This work lays the foundation for further research into emotionally informed player models, advanced AI techniques for adaptivity, and broader applications beyond gaming in educational settings.

Large Language Models (LLMs) are widely used across research domains to tackle complex tasks, but their performance can vary significantly depending on the task at hand. Evolutionary algorithms, inspired by natural selection, can be used to refine solutions iteratively at inference-time. To the best of our knowledge, there has not been exploration on leveraging the collective capabilities of multi-source seeding for LLM-guided genetic algorithms. In this paper, we introduce a novel approach, MultiGA, which applies genetic algorithm principles to address complex natural language tasks and reasoning problems by sampling from a diverse population of LLMs to initialize the population. MultiGA generates a range of outputs from various parent LLMs, open source and closed source, and uses a neutral fitness function to evaluate them. Through an iterative recombination process, we mix and refine these generations until an optimal solution is achieved. We benchmark our approach using text-to-SQL code generation tasks, trip planning, GPQA benchmark for grad-level science questions, and the BBQ bias benchmark. Our results show that MultiGA converges to the accuracy of the LLM best fit for the task, and these insights lay the foundation for future research looking closer at integrating multiple LLMs for unexplored tasks in which selecting only one pre-trained model is unclear or suboptimal.

Multimodal Artificial Intelligence (AI) systems, particularly Vision-Language Models (VLMs), have become integral to critical applications ranging from autonomous decision-making to automated document processing. As these systems scale, they rely heavily on preprocessing pipelines to handle diverse inputs efficiently. However, this dependency on standard preprocessing operations, specifically image downscaling, creates a significant yet often overlooked security vulnerability. While intended for computational optimization, scaling algorithms can be exploited to conceal malicious visual prompts that are invisible to human observers but become active semantic instructions once processed by the model. Current adversarial strategies remain largely static, failing to account for the dynamic nature of modern agentic workflows. To address this gap, we propose Chameleon, a novel, adaptive adversarial framework designed to expose and exploit scaling vulnerabilities in production VLMs. Unlike traditional static attacks, Chameleon employs an iterative, agent-based optimization mechanism that dynamically refines image perturbations based on the target model's real-time feedback. This allows the framework to craft highly robust adversarial examples that survive standard downscaling operations to hijack downstream execution. We evaluate Chameleon against Gemini 2.5 Flash model. Our experiments demonstrate that Chameleon achieves an Attack Success Rate (ASR) of 84.5% across varying scaling factors, significantly outperforming static baseline attacks which average only 32.1%. Furthermore, we show that these attacks effectively compromise agentic pipelines, reducing decision-making accuracy by over 45% in multi-step tasks. Finally, we discuss the implications of these vulnerabilities and propose multi-scale consistency checks as a necessary defense mechanism.

Initial challenges, such as grade-school mathematics (GSM8K) and standard competition math (MATH dataset), have largely been surmounted, pushing the frontier of AI reasoning toward "grand challenge" problems, such as those found in the International Mathematical Olympiad (IMO). These problems, renowned for their demand for deep insight, creativity, and rigorous proof, expose a fascinating weakness in modern LLMs. While a model's performance on a single attempt (termed pass@1) may be very low, its ability to produce a correct answer within k attempts (pass@k) can be significantly higher. This pass@1 versus pass@k gap, especially pronounced when sampling with high temperature to produce diverse outputs, suggests that models possess a vast, latent capability that is not accessible in a single, high-confidence generation. Interestingly, to recover the full power of the model it is not sufficient to simply use multiple attempts. In fact, even the pass@k metric fails to capture the full story. On the most difficult problems, simply sampling k times and selecting the best answer (e.g., "best-of-32") still yields poor results. For instance, Huang and Yang (2025) report that a best-of-32 baseline on the IMO 2025 problems achieved an accuracy of only 31.6-38.1% for leading models [HY25]. This paradox lies at the heart of our work: the latent capability of LLMs is not merely a matter of selection (finding one correct needle in a haystack of k attempts), but one of synthesis.

Jailbreak attacks aim to exploit large language models (LLMs) by inducing them to generate harmful content, thereby revealing their vulnerabilities. Understanding and addressing these attacks is crucial for advancing the field of LLM safety. Previous jailbreak approaches have mainly focused on direct manipulations of harmful intent, with limited attention to the impact of persona prompts. In this study, we systematically explore the efficacy of persona prompts in compromising LLM defenses. We propose a genetic algorithm-based method that automatically crafts persona prompts to bypass LLM's safety mechanisms. Our experiments reveal that: (1) our evolved persona prompts reduce refusal rates by 50-70% across multiple LLMs, and (2) these prompts demonstrate synergistic effects when combined with existing attack methods, increasing success rates by 10-20%. Our code and data are available at https://github.com/CjangCjengh/Generic_Persona.

Context-based question answering (CBQA) models provide more accurate and relevant answers by considering the contextual information. They effectively extract specific information given a context, making them functional in various applications involving user support, information retrieval, and educational platforms. In this manuscript, we benchmarked the performance of 47 CBQA models from Hugging Face on eight different datasets. This study aims to identify the best-performing model across diverse datasets without additional fine-tuning. It is valuable for practical applications where the need to retrain models for specific datasets is minimized, streamlining the implementation of these models in various contexts. The best-performing models were trained on the SQuAD v2 or SQuAD v1 datasets. The best-performing model was ahotrod/electra_large_discriminator_squad2_512, which yielded 43\% accuracy across all datasets. We observed that the computation time of all models depends on the context length and the model size. The model's performance usually decreases with an increase in the answer length. Moreover, the model's performance depends on the context complexity. We also used the Genetic algorithm to improve the overall accuracy by integrating responses from other models. ahotrod/electra_large_discriminator_squad2_512 generated the best results for bioasq10b-factoid (65.92\%), biomedical\_cpgQA (96.45\%), QuAC (11.13\%), and Question Answer Dataset (41.6\%). Bert-large-uncased-whole-word-masking-finetuned-squad achieved an accuracy of 82\% on the IELTS dataset.