Large language models (LLMs) are typically evaluated on the basis of task-based benchmarks such as MMLU. Such benchmarks do not examine responsible behaviour of LLMs in specific contexts. This is particularly true in the LGBTI+ context where social stereotypes may result in variation in LGBTI+ terminology. Therefore, domain-specific lexicons or dictionaries may be useful as a representative list of words against which the LLM's behaviour needs to be evaluated. This paper presents a methodology for evaluation of LLMs using an LGBTI+ lexicon in Indian languages. The methodology consists of four steps: formulating NLP tasks relevant to the expected behaviour, creating prompts that test LLMs, using the LLMs to obtain the output and, finally, manually evaluating the results. Our qualitative analysis shows that the three LLMs we experiment on are unable to detect underlying hateful content. Similarly, we observe limitations in using machine translation as means to evaluate natural language understanding in languages other than English. The methodology presented in this paper can be useful for LGBTI+ lexicons in other languages as well as other domain-specific lexicons. The work done in this paper opens avenues for responsible behaviour of LLMs, as demonstrated in the context of prevalent social perception of the LGBTI+ community.
相關內容
We advocate the idea of the natural-language-driven(NLD) simulation to efficiently produce the object interactions between multiple objects in the virtual road scenes, for teaching and testing the autonomous driving systems that should take quick action to avoid collision with obstacles with unpredictable motions. The NLD simulation allows the brief natural-language description to control the object interactions, significantly reducing the human efforts for creating a large amount of interaction data. To facilitate the research of NLD simulation, we collect the Language-to-Interaction(L2I) benchmark dataset with 120,000 natural-language descriptions of object interactions in 6 common types of road topologies. Each description is associated with the programming code, which the graphic render can use to visually reconstruct the object interactions in the virtual scenes. As a methodology contribution, we design SimCopilot to translate the interaction descriptions to the renderable code. We use the L2I dataset to evaluate SimCopilot's abilities to control the object motions, generate complex interactions, and generalize interactions across road topologies. The L2I dataset and the evaluation results motivate the relevant research of the NLD simulation.
The field of adversarial textual attack has significantly grown over the last few years, where the commonly considered objective is to craft adversarial examples (AEs) that can successfully fool the target model. However, the imperceptibility of attacks, which is also essential for practical attackers, is often left out by previous studies. In consequence, the crafted AEs tend to have obvious structural and semantic differences from the original human-written text, making them easily perceptible. In this work, we advocate leveraging multi-objectivization to address such issue. Specifically, we reformulate the problem of crafting AEs as a multi-objective optimization problem, where the attack imperceptibility is considered as an auxiliary objective. Then, we propose a simple yet effective evolutionary algorithm, dubbed HydraText, to solve this problem. To the best of our knowledge, HydraText is currently the only approach that can be effectively applied to both score-based and decision-based attack settings. Exhaustive experiments involving 44237 instances demonstrate that HydraText consistently achieves competitive attack success rates and better attack imperceptibility than the recently proposed attack approaches. A human evaluation study also shows that the AEs crafted by HydraText are more indistinguishable from human-written text. Finally, these AEs exhibit good transferability and can bring notable robustness improvement to the target model by adversarial training.
Small-scale models offer various computational advantages, and yet to which extent size is critical for problem-solving abilities remains an open question. Specifically for solving grade school math, the smallest model size so far required to break the 80\% barrier on the GSM8K benchmark remains to be 34B. Our work studies how high-quality datasets may be the key for small language models to acquire mathematical reasoning. We introduce \texttt{TinyGSM}, a synthetic dataset of 12.3M grade school math problems paired with Python solutions, generated fully by GPT-3.5. After finetuning on \texttt{TinyGSM}, we find that a duo of a 1.3B generation model and a 1.3B verifier model can achieve 81.5\% accuracy, outperforming existing models that are orders of magnitude larger. This also rivals the performance of the GPT-3.5 ``teacher'' model (77.4\%), from which our model's training data is generated. Our approach is simple and has two key components: 1) the high-quality dataset \texttt{TinyGSM}, 2) the use of a verifier, which selects the final outputs from multiple candidate generations.
The Segment Anything Model (SAM) has recently emerged as a significant breakthrough in foundation models, demonstrating remarkable zero-shot performance in object segmentation tasks. While SAM is designed for generalization, it exhibits limitations in handling specific medical imaging tasks that require fine-structure segmentation or precise boundaries. In this paper, we focus on the task of cardiac magnetic resonance imaging (cMRI) short-axis view segmentation using the SAM foundation model. We conduct a comprehensive investigation of the impact of different prompting strategies (including bounding boxes, positive points, negative points, and their combinations) on segmentation performance. We evaluate on two public datasets using the baseline model and models fine-tuned with varying amounts of annotated data, ranging from a limited number of volumes to a fully annotated dataset. Our findings indicate that prompting strategies significantly influence segmentation performance. Combining positive points with either bounding boxes or negative points shows substantial benefits, but little to no benefit when combined simultaneously. We further observe that fine-tuning SAM with a few annotated volumes improves segmentation performance when properly prompted. Specifically, fine-tuning with bounding boxes has a positive impact, while fine-tuning without bounding boxes leads to worse results compared to baseline.
As corporations rush to integrate large language models (LLMs) to their search offerings, it is critical that they provide factually accurate information that is robust to any presuppositions that a user may express. In this work, we introduce UPHILL, a dataset consisting of health-related queries with varying degrees of presuppositions. Using UPHILL, we evaluate the factual accuracy and consistency of InstructGPT, ChatGPT, and BingChat models. We find that while model responses rarely disagree with true health claims (posed as questions), they often fail to challenge false claims: responses from InstructGPT agree with 32% of the false claims, ChatGPT 26% and BingChat 23%. As we increase the extent of presupposition in input queries, the responses from InstructGPT and ChatGPT agree with the claim considerably more often, regardless of its veracity. Responses from BingChat, which rely on retrieved webpages, are not as susceptible. Given the moderate factual accuracy, and the inability of models to consistently correct false assumptions, our work calls for a careful assessment of current LLMs for use in high-stakes scenarios.
Sample selection models represent a common methodology for correcting bias induced by data missing not at random. It is well known that these models are not empirically identifiable without exclusion restrictions. In other words, some variables predictive of missingness do not affect the outcome model of interest. The drive to establish this requirement often leads to the inclusion of irrelevant variables in the model. A recent proposal uses adaptive LASSO to circumvent this problem, but its performance depends on the so-called covariance assumption, which can be violated in small to moderate samples. Additionally, there are no tools yet for post-selection inference for this model. To address these challenges, we propose two families of spike-and-slab priors to conduct Bayesian variable selection in sample selection models. These prior structures allow for constructing a Gibbs sampler with tractable conditionals, which is scalable to the dimensions of practical interest. We illustrate the performance of the proposed methodology through a simulation study and present a comparison against adaptive LASSO and stepwise selection. We also provide two applications using publicly available real data. An implementation and code to reproduce the results in this paper can be found at //github.com/adam-iqbal/selection-spike-slab
Smart meters are of the basic elements in the so-called Smart Grid. These devices, connected to the Internet, keep bidirectional communication with other devices in the Smart Grid structure to allow remote readings and maintenance. As any other device connected to a network, smart meters become vulnerable to attacks with different purposes, like stealing data or altering readings. Nowadays, it is becoming more and more popular to buy and plug-and-play smart meters, additionally to those installed by the energy providers, to directly monitor the energy consumption at home. This option inherently entails security risks that are under the responsibility of householders. In this paper, we focus on an open solution based on Smartpi 2.0 devices with two purposes. On the one hand, we propose a network configuration and different data flows to exchange data (energy readings) in the home. These flows are designed to support collaborative among the devices in order to prevent external attacks and attempts of corrupting the data. On the other hand, we check the vulnerability by performing two kind of attacks (denial of service and stealing and changing data by using a malware). We conclude that, as expected, these devices are vulnerable to these attacks, but we provide mechanisms to detect both of them and to solve, by applying cooperation techniques
Large language models (LLMs) can be used as accessible and intelligent chatbots by constructing natural language queries and directly inputting the prompt into the large language model. However, different prompt' constructions often lead to uncertainty in the answers and thus make it hard to utilize the specific knowledge of LLMs (like ChatGPT). To alleviate this, we use an interpretable structure to explain the prompt learning principle in LLMs, which certificates that the effectiveness of language models is determined by position changes of the task's related tokens. Therefore, we propose MTPrompt, a multi-dimensional task prompt learning method consisting based on task-related object, summary, and task description information. By automatically building and searching for appropriate prompts, our proposed MTPrompt achieves the best results on few-shot samples setting and five different datasets. In addition, we demonstrate the effectiveness and stability of our method in different experimental settings and ablation experiments. In interaction with large language models, embedding more task-related information into prompts will make it easier to stimulate knowledge embedded in large language models.
Incorporating prior knowledge into pre-trained language models has proven to be effective for knowledge-driven NLP tasks, such as entity typing and relation extraction. Current pre-training procedures usually inject external knowledge into models by using knowledge masking, knowledge fusion and knowledge replacement. However, factual information contained in the input sentences have not been fully mined, and the external knowledge for injecting have not been strictly checked. As a result, the context information cannot be fully exploited and extra noise will be introduced or the amount of knowledge injected is limited. To address these issues, we propose MLRIP, which modifies the knowledge masking strategies proposed by ERNIE-Baidu, and introduce a two-stage entity replacement strategy. Extensive experiments with comprehensive analyses illustrate the superiority of MLRIP over BERT-based models in military knowledge-driven NLP tasks.
For languages with no annotated resources, transferring knowledge from rich-resource languages is an effective solution for named entity recognition (NER). While all existing methods directly transfer from source-learned model to a target language, in this paper, we propose to fine-tune the learned model with a few similar examples given a test case, which could benefit the prediction by leveraging the structural and semantic information conveyed in such similar examples. To this end, we present a meta-learning algorithm to find a good model parameter initialization that could fast adapt to the given test case and propose to construct multiple pseudo-NER tasks for meta-training by computing sentence similarities. To further improve the model's generalization ability across different languages, we introduce a masking scheme and augment the loss function with an additional maximum term during meta-training. We conduct extensive experiments on cross-lingual named entity recognition with minimal resources over five target languages. The results show that our approach significantly outperforms existing state-of-the-art methods across the board.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191