“Jailbreak” is a major safety concern of Large Language Models (LLMs), which occurs when malicious prompts lead LLMs to produce harmful outputs, raising issues about the reliability and safety of LLMs. Therefore, an effective evaluation of jailbreaks is very crucial to develop its mitigation strategies. However, our research reveals that many jailbreaks identified by current evaluations may actually be hallucinations—erroneous outputs that are mistaken for genuine safety breaches. This finding suggests that some perceived vulnerabilities might not represent actual threats, indicating a need for more precise red teaming benchmarks. To address this problem, we propose the Benchmark for reliABilitY and jailBreak haLlUcination Evaluation (BabyBLUE). BabyBLUE introduces a specialized validation framework including various evaluators to enhance existing jailbreak benchmarks, ensuring outputs are useful malicious instructions. Additionally, BabyBLUE presents a new dataset as an augmentation to the existing red teaming benchmarks, specifically addressing hallucinations in jailbreaks, aiming to evaluate the true potential of jailbroken LLM outputs to cause harm to human society.

Federated Learning (FL) is popular as a privacy-preserving machine learning paradigm for generating a single model on decentralized data. However, statistical heterogeneity poses a significant challenge for FL. As a subfield of FL, personalized FL (pFL) has attracted attention for its ability to achieve personalized models that perform well on non-independent and identically distributed (Non-IID) data. However, existing pFL methods are limited in terms of leveraging the global model’s knowledge to enhance generalization while achieving personalization on local data. To address this, we proposed a new method personalized Federated learning with Adaptive Feature Aggregation and Knowledge Transfer (FedAFK), to train better feature extractors while balancing generalization and personalization for participating clients, which improves the performance of personalized models on Non-IID data. We conduct extensive experiments on three datasets in two widely-used heterogeneous settings and show the superior performance of our proposed method over thirteen state-of-the-art baselines.

In undirected graphs with real non-negative weights, we give a new randomized algorithm for the single-source shortest path (SSSP) problem with running time $O(m \sqrt{\log n \cdot \log \log n})$ in the comparison-addition model. This is the first algorithm to break the $O(m+n \log n)$ time bound for real-weighted sparse graphs by Dijkstra’s algorithm with Fibonacci heaps. Previous undirected nonnegative SSSP algorithms give time bound of $O(m \alpha(m, n)+ \min \{n \log n, n \log \log r\})$ in comparison-addition model, where $\alpha$ is the inverse-Ackermann function and r is the ratio of the maximum-to-minimum edge weight [Pettie & Ramachandran 2005], and linear time for integer edge weights in RAM model [Thorup 1999]. Note that there is a proposed complexity lower bound of $\Omega(m+\min \{n \log n, n \log \log r\})$ for hierarchy-based algorithms for undirected real-weighted SSSP [Pettie & Ramachandran 2005], but our algorithm does not obey the properties required for that lower bound. As a non-hierarchybased approach, our algorithm shows great advantage with much simpler structure, and is much easier to implement.