Large Language Models (LLMs) such as ChatGPT have received enormous attention over the past year and are now used by hundreds of millions of people every day. The rapid adoption of this technology naturally raises questions about the possible biases such models might exhibit. In this work, we tested one of these models (GPT-3) on a range of cognitive effects, which are systematic patterns that are usually found in human cognitive tasks. We found that LLMs are indeed prone to several human cognitive effects. Specifically, we show that the priming, distance, SNARC, and size congruity effects were presented with GPT-3, while the anchoring effect is absent. We describe our methodology, and specifically the way we converted real-world experiments to text-based experiments. Finally, we speculate on the possible reasons why GPT-3 exhibits these effects and discuss whether they are imitated or reinvented.

In the realm of natural language processing, the understanding of tabular data has perpetually stood as a focal point of scholarly inquiry. The emergence of expansive language models, exemplified by the likes of ChatGPT, has ushered in a wave of endeavors wherein researchers aim to harness these models for tasks related to table-based question answering. Central to our investigative pursuits is the elucidation of methodologies that amplify the aptitude of such large language models in discerning both the structural intricacies and inherent content of tables, ultimately facilitating their capacity to provide informed responses to pertinent queries. To this end, we have architected a distinctive module dedicated to the serialization of tables for seamless integration with expansive language models. Additionally, we've instituted a corrective mechanism within the model to rectify potential inaccuracies. Experimental results indicate that, although our proposed method trails the SOTA by approximately 11.7% in overall metrics, it surpasses the SOTA by about 1.2% in tests on specific datasets. This research marks the first application of large language models to table-based question answering tasks, enhancing the model's comprehension of both table structures and content.

We present Region-aware Open-vocabulary Vision Transformers (RO-ViT) - a contrastive image-text pretraining recipe to bridge the gap between image-level pretraining and open-vocabulary object detection. At the pretraining phase, we propose to randomly crop and resize regions of positional embeddings instead of using the whole image positional embeddings. This better matches the use of positional embeddings at region-level in the detection finetuning phase. In addition, we replace the common softmax cross entropy loss in contrastive learning with focal loss to better learn the informative yet difficult examples. Finally, we leverage recent advances in novel object proposals to improve open-vocabulary detection finetuning. We evaluate our full model on the LVIS and COCO open-vocabulary detection benchmarks and zero-shot transfer. RO-ViT achieves a state-of-the-art 34.1 $AP_r$ on LVIS, surpassing the best existing approach by +7.8 points in addition to competitive zero-shot transfer detection. Surprisingly, RO-ViT improves the image-level representation as well and achieves the state of the art on 9 out of 12 metrics on COCO and Flickr image-text retrieval benchmarks, outperforming competitive approaches with larger models.

The emergence of generative Large Language Models (LLMs) emphasizes the need for accurate and efficient prompting approaches. LLMs are often applied in Few-Shot Learning (FSL) contexts, where tasks are executed with minimal training data. FSL has become popular in many Artificial Intelligence (AI) subdomains, including AI for health. Rare diseases affect a small fraction of the population. Rare disease identification from clinical notes inherently requires FSL techniques due to limited data availability. Manual data collection and annotation is both expensive and time-consuming. In this paper, we propose Models-Vote Prompting (MVP), a flexible prompting approach for improving the performance of LLM queries in FSL settings. MVP works by prompting numerous LLMs to perform the same tasks and then conducting a majority vote on the resulting outputs. This method achieves improved results to any one model in the ensemble on one-shot rare disease identification and classification tasks. We also release a novel rare disease dataset for FSL, available to those who signed the MIMIC-IV Data Use Agreement (DUA). Furthermore, in using MVP, each model is prompted multiple times, substantially increasing the time needed for manual annotation, and to address this, we assess the feasibility of using JSON for automating generative LLM evaluation.

Benefiting from the powerful capabilities of large language models (LLMs), agents based on LLMs have shown the potential to address domain-specific tasks and emulate human behaviors. However, the content generated by these agents remains somewhat superficial, owing to their limited domain expertise and the absence of an effective cognitive architecture. To address this, we present the Configurable General Multi-Agent Interaction (CGMI) framework, designed to replicate human interactions in real-world scenarios. Specifically, we propose a tree-structured methodology for the assignment, detection, and maintenance of agent personality. Additionally, we designed a cognitive architecture equipped with a skill library based on the ACT* model, which contains memory, reflection, and planning modules. We have also integrated general agents to augment the virtual environment's realism. Using the CGMI framework, we simulated numerous classroom interactions between teacher and students. The experiments indicate that aspects such as the teaching methodology, curriculum, and student performance closely mirror real classroom settings. We will open source our work.

Large Language Models (LLMs) have gained widespread popularity across diverse domains involving text generation, summarization, and various natural language processing tasks. Despite their inherent limitations, LLM-based designs have shown promising capabilities in planning and navigating open-world scenarios. This paper introduces a novel application of pre-trained LLMs as agents within cybersecurity network environments, focusing on their utility for sequential decision-making processes. We present an approach wherein pre-trained LLMs are leveraged as attacking agents in two reinforcement learning environments. Our proposed agents demonstrate similar or better performance against state-of-the-art agents trained for thousands of episodes in most scenarios and configurations. In addition, the best LLM agents perform similarly to human testers of the environment without any additional training process. This design highlights the potential of LLMs to efficiently address complex decision-making tasks within cybersecurity. Furthermore, we introduce a new network security environment named NetSecGame. The environment is designed to eventually support complex multi-agent scenarios within the network security domain. The proposed environment mimics real network attacks and is designed to be highly modular and adaptable for various scenarios.

Recently, instruction-following Large Language Models (LLMs) , represented by ChatGPT, have exhibited exceptional performance in general Natural Language Processing (NLP) tasks. However, the unique characteristics of E-commerce data pose significant challenges to general LLMs. An LLM tailored specifically for E-commerce scenarios, possessing robust cross-dataset/task generalization capabilities, is a pressing necessity. To solve this issue, in this work, we proposed the first e-commerce instruction dataset EcomInstruct, with a total of 2.5 million instruction data. EcomInstruct scales up the data size and task diversity by constructing atomic tasks with E-commerce basic data types, such as product information, user reviews. Atomic tasks are defined as intermediate tasks implicitly involved in solving a final task, which we also call Chain-of-Task tasks. We developed EcomGPT with different parameter scales by training the backbone model BLOOMZ with the EcomInstruct. Benefiting from the fundamental semantic understanding capabilities acquired from the Chain-of-Task tasks, EcomGPT exhibits excellent zero-shot generalization capabilities. Extensive experiments and human evaluations demonstrate that EcomGPT outperforms ChatGPT in term of cross-dataset/task generalization on E-commerce tasks.

Large Language Models (LLMs), such as GPT and BERT, have demonstrated remarkable capabilities in addressing neural language process tasks. Recently, the release of ChatGPT has garnered significant attention due to its ability to analyze, comprehend, and synthesize information from user inputs. Therefore, these LLMs were adopted by researchers in many different domains. In the realm of code analysis, researchers have applied LLMs to tasks like code review and code generation. However, we observed that the strengths and limitations of adopting these LLMs to the code analysis have not been investigated. In this paper, we delve into LLMs' capabilities in security-oriented program analysis, considering perspectives from both attackers and security analysts. We focus on two representative LLMs, ChatGPT and CodeBert, and evaluate their performance in solving typical analytic tasks with varying levels of difficulty. Given the different natures of ChatGPT and CodeBERT, we conduct a qualitative analysis of the model's output for ChatGPT and a quantitative analysis for CodeBERT, respectively. For ChatGPT, we present a case study involving several security-oriented program analysis tasks while deliberately introducing challenges to assess its responses. On the other hand, for CodeBERT, we systematically analyze and classify the features in code, quantitatively evaluating the impact of these features on the model's performance. Our study demonstrates the LLM's efficiency in learning high-level semantics from code, positioning ChatGPT as a potential asset in security-oriented contexts. However, it is essential to acknowledge certain limitations, such as the heavy reliance on well-defined variable and function names, making them unable to learn from anonymized code. We hope that our findings and analysis will offer valuable insights for future researchers in this domain.

Considerable research efforts have been devoted to ensuring that large language models (LLMs) align with human values and generate safe text. However, an excessive focus on sensitivity to certain topics can compromise the model's robustness in following instructions, thereby impacting its overall performance in completing tasks. Previous benchmarks for jailbreaking LLMs have primarily focused on evaluating the safety of the models without considering their robustness. In this paper, we propose a benchmark that assesses both the safety and robustness of LLMs, emphasizing the need for a balanced approach. To comprehensively study text safety and output robustness, we introduce a latent jailbreak prompt dataset, each involving malicious instruction embedding. Specifically, we instruct the model to complete a regular task, such as translation, with the text to be translated containing malicious instructions. To further analyze safety and robustness, we design a hierarchical annotation framework. We present a systematic analysis of the safety and robustness of LLMs regarding the position of explicit normal instructions, word replacements (verbs in explicit normal instructions, target groups in malicious instructions, cue words for explicit normal instructions), and instruction replacements (different explicit normal instructions). Our results demonstrate that current LLMs not only prioritize certain instruction verbs but also exhibit varying jailbreak rates for different instruction verbs in explicit normal instructions. Code and data are available at https://github.com/qiuhuachuan/latent-jailbreak.

The Segment Anything Model (SAM) is a recently proposed prompt-based segmentation model in a generic zero-shot segmentation approach. With the zero-shot segmentation capacity, SAM achieved impressive flexibility and precision on various segmentation tasks. However, the current pipeline requires manual prompts during the inference stage, which is still resource intensive for biomedical image segmentation. In this paper, instead of using prompts during the inference stage, we introduce a pipeline that utilizes the SAM, called all-in-SAM, through the entire AI development workflow (from annotation generation to model finetuning) without requiring manual prompts during the inference stage. Specifically, SAM is first employed to generate pixel-level annotations from weak prompts (e.g., points, bounding box). Then, the pixel-level annotations are used to finetune the SAM segmentation model rather than training from scratch. Our experimental results reveal two key findings: 1) the proposed pipeline surpasses the state-of-the-art (SOTA) methods in a nuclei segmentation task on the public Monuseg dataset, and 2) the utilization of weak and few annotations for SAM finetuning achieves competitive performance compared to using strong pixel-wise annotated data.