Task automation has been greatly empowered by the recent advances in Large Language Models (LLMs) via Python code, where the tasks range from software engineering development to general-purpose reasoning. While current benchmarks have shown that LLMs can solve tasks using programs like human developers, the majority of their evaluations are limited to short and self-contained algorithmic tasks or standalone function calls. Solving challenging and practical tasks requires the capability of utilizing **diverse function calls as tools** to efficiently implement functionalities like data analysis and web development. In addition, using multiple tools to solve a task needs compositional reasoning by accurately understanding **complex instructions**. Fulfilling both of these characteristics can pose a great challenge for LLMs. To assess how well LLMs can solve challenging and practical tasks via programs, we introduce BigCodeBench, a benchmark that challenges LLMs to invoke multiple function calls as tools from 139 libraries and 7 domains for 1,140 fine-grained tasks. To evaluate LLMs rigorously, each task encompasses 5.6 test cases with an average branch coverage of 99%. In addition, we propose a natural-language-oriented variant of BigCodeBench, BigCodeBench-Instruct, that automatically transforms the original docstrings into short instructions containing only essential information. Our extensive evaluation of 60 LLMs shows that **LLMs are not yet capable of following complex instructions to use function calls precisely, with scores up to 60%, significantly lower than the human performance of 97%**. The results underscore the need for further advancements in this area.

With the increasing interconnection of smart devices, users often desire to adopt the same app on quite different devices for identical tasks, such as watching the same movies on both their smartphones and TVs. However, the significant differences in screen size, aspect ratio, and interaction styles make it challenging to adapt Graphical User Interfaces (GUIs) across these devices. Although there are millions of apps available on Google Play, only a few thousand are designed to support smart TV displays. Existing techniques to map a mobile app GUI to a TV either adopt a responsive design, which struggles to bridge the substantial gap between phone and TV, or use mirror apps for improved video display, which requires hardware support and extra engineering efforts. Instead of developing another app for supporting TVs, we propose a semi-automated approach to generate corresponding adaptive TV GUIs, given the phone GUIs as the input. Based on our empirical study of GUI pairs for TVs and phones in existing apps, we synthesize a list of rules for grouping and classifying phone GUIs, converting them to TV GUIs, and generating dynamic TV layouts and source code for the TV display. Our tool is not only beneficial to developers but also to GUI designers, who can further customize the generated GUIs for their TV app development. An evaluation and user study demonstrate the accuracy of our generated GUIs and the usefulness of our tool.

Powered by the rising popularity of deep learning techniques on smartphones, on-device deep learning models are being used in vital fields such as finance, social media, and driving assistance. Because of the transparency of the Android platform and the on-device models inside, on-device models on Android smartphones have been proven to be extremely vulnerable. However, due to the challenge in accessing and analyzing iOS app files, despite iOS being a mobile platform as popular as Android, there are no relevant works on on-device models in iOS apps. Since the functionalities of the same app on Android and iOS platforms are similar, the same vulnerabilities may exist on both platforms. In this article, we present the first empirical study about on-device models in iOS apps, including their adoption of deep learning frameworks, structure, functionality, and potential security issues. We study why current developers use different on-device models for one app between iOS and Android. We propose a more general attack against white-box models that does not rely on pre-trained models and a new adversarial attack approach based on our findings to target iOS’s gray-box on-device models. Our results show the effectiveness of our approaches. Finally, we successfully exploit the vulnerabilities of on-device models to attack real-world iOS apps.

Mobile apps are ubiquitous in our daily lives for supporting different tasks such as reading and chatting. Despite the availability of many GUI testing tools, app testers still struggle with low testing code coverage due to tools frequently getting stuck in loops or overlooking activities with concealed entries. This results in a significant amount of testing time being spent on redundant and repetitive exploration of a few GUI pages. To address this, we utilize Android’s deep links, which assist in triggering Android intents to lead users to specific pages and introduce a deep link-enhanced exploration method. This approach, integrated into the testing tool Monkey, gives rise to Delm (Deep Link-enhanced Monkey). Delm oversees the dynamic exploration process, guiding the tool out of meaningless testing loops to unexplored GUI pages. We provide a rigorous activity context mock-up approach for triggering existing Android intents to discover more activities with hidden entrances. We conduct experiments to evaluate Delm’s effectiveness on activity context mock-up, activity coverage, method coverage, and crash detection. The findings reveal that Delm can mock up more complex activity contexts and significantly outperform state-of-the-art baselines with 27.2% activity coverage, 21.13% method coverage, and 23.81% crash detection.

Large-scale language models have achieved tremendous success across various natural language processing (NLP) applications. Nevertheless, language models are vulnerable to backdoor attacks, which inject stealthy triggers into models for steering them to undesirable behaviors. Most existing backdoor attacks, such as data poisoning, require further (re)training or fine-tuning language models to learn the intended backdoor patterns. The additional training process however diminishes the stealthiness of the attacks, as training a language model usually requires long optimization time, a massive amount of data, and considerable modifications to the model parameters.

Code summarization aims at generating a code comment given a block of source code and it is normally performed by training machine learning algorithms on existing code block-comment pairs. Code comments in practice have different intentions. For example, some code comments might explain how the methods work, while others explain why some methods are written. Previous works have shown that a relationship exists between a code block and the category of a comment associated with it. In this article, we aim to investigate to which extent we can exploit this relationship to improve code summarization performance. We first classify comments into six intention categories and manually label 20,000 code-comment pairs. These categories include “what,” “why,” “how-to-use,” “how-it-is-done,” “property,” and “others.” Based on this dataset, we conduct an experiment to investigate the performance of different state-of-the-art code summarization approaches on the categories. We find that the performance of different code summarization approaches varies substantially across the categories. Moreover, the category for which a code summarization model performs the best is different for the different models. In particular, no models perform the best for “why” and “property” comments among the six categories. We design a composite approach to demonstrate that comment category prediction can boost code summarization to reach better results. The approach leverages classified code-category labeled data to train a classifier to infer categories. Then it selects the most suitable models for inferred categories and outputs the composite results. Our composite approach outperforms other approaches that do not consider comment categories and obtains a relative improvement of 8.57% and 16.34% in terms of ROUGE-L and BLEU-4 score, respectively.