This paper introduces LongBench v2, a benchmark designed to assess the ability of LLMs to handle long-context problems requiring deep understanding and reasoning across real-world multitasks. LongBench v2 consists of 503 challenging multiple-choice questions, with contexts ranging from 8k to 2M words, across six major task categories: single-document QA, multi-document QA, long in-context learning, long-dialogue history understanding, code repository understanding, and long structured data understanding. To ensure the breadth and the practicality, we collect data from nearly 100 highly educated individuals with diverse professional backgrounds. We employ both automated and manual review processes to maintain high quality and difficulty, resulting in human experts achieving only 53.7% accuracy under a 15-minute time constraint. Our evaluation reveals that the best-performing model, when directly answers the questions, achieves only 50.1% accuracy. In contrast, the o1-preview model, which includes longer reasoning, achieves 57.7%, surpassing the human baseline by 4%. These results highlight the importance of enhanced reasoning ability and scaling inference-time compute to tackle the long-context challenges in LongBench v2.

The development of large language models (LLMs) depends on **trustworthy evaluation**. However, most current evaluations rely on public benchmarks, which are prone to data contamination issues that significantly compromise fairness. Previous researches have focused on constructing dynamic benchmarks to address contamination. However, continuously building new benchmarks is costly and cyclical.In this work, we aim to tackle contamination by analyzing the mechanisms of contaminated models themselves. Through our experiments, we discover that the overestimation of contaminated models is likely due to parameters acquiring shortcut solutions in training. We further propose a novel method for identifying shortcut neurons through **comparative and causal analysis**.Building on this, we introduce an evaluation method called **shortcut neuron patching** to suppress shortcut neurons. Experiments validate the effectiveness of our approach in mitigating contamination. Additionally, our evaluation results exhibit a strong linear correlation with MixEval, a recently released trustworthy benchmark, achieving a Spearman coefficient (𝜌) exceeding 0.95. This high correlation indicates that our method closely reveals true capabilities of the models and is trustworthy. We conduct further experiments to demonstrate the generalizability of our method across various benchmarks and hyperparameter settings. **Code**: https://github.com/GaryStack/Trustworthy-Evaluation.

To mitigate the potential misuse of large language models (LLMs), recent research has developed watermarking algorithms, which restrict the generation process to leave an invisible trace for watermark detection. Due to the two-stage nature of the task, most studies evaluate the generation and detection separately, thereby presenting a challenge in unbiased, thorough, and applicable evaluations. In this paper, we introduce WaterBench, the first comprehensive benchmark for LLM watermarks, in which we design three crucial factors: (1) For benchmarking procedure, to ensure an apples-to-apples comparison, we first adjust each watermarking method’s hyper-parameter to reach the same watermarking strength, then jointly evaluate their generation and detection performance. (2) For task selection, we diversify the input and output length to form a five-category taxonomy, covering 9 tasks. (3) For evaluation metric, we adopt the GPT4-Judge for automatically evaluating the decline of instruction-following abilities after watermarking. We evaluate 4 open-source watermarks on 2 LLMs under 2 watermarking strengths and observe the common struggles for current methods on maintaining the generation quality. The code and data are available at https://github.com/THU-KEG/WaterBench.

Multi-Document Summarization (MDS) commonly employs the 2-stage extract-then-abstract paradigm, which first extracts a relatively short meta-document, then feeds it into the deep neural networks to generate an abstract. Previous work usually takes the ROUGE score as the label for training a scoring model to evaluate source documents. However, the trained scoring model is prone to under-fitting for low-resource settings, as it relies on the training data. To extract documents effectively, we construct prompting templates that invoke the underlying knowledge in Pre-trained Language Model (PLM) to calculate the document and keyword’s perplexity, which can assess the document’s semantic salience. Our unsupervised approach can be applied as a plug-in to boost other metrics for evaluating a document’s salience, thus improving the subsequent abstract generation. We get positive results on 2 MDS datasets, 2 data settings, and 2 abstractive backbone models, showing our method’s effectiveness. Our code is available at https://github.com/THU-KEG/UPER

Wikipedia abstract generation aims to distill a Wikipedia abstract from web sources and has met significant success by adopting multi-document summarization techniques. However, previous works generally view the abstract as plain text, ignoring the fact that it is a description of a certain entity and can be decomposed into different topics. In this paper, we propose a two-stage model TWAG that guides the abstract generation with topical information. First, we detect the topic of each input paragraph with a classifier trained on existing Wikipedia articles to divide input documents into different topics. Then, we predict the topic distribution of each abstract sentence, and decode the sentence from topic-aware representations with a Pointer-Generator network. We evaluate our model on the WikiCatSum dataset, and the results show that TWAG outperforms various existing baselines and is capable of generating comprehensive abstracts.