We present Contrastive Feature Masking Vision Transformer (CFM-ViT) - an image-text pretraining methodology that achieves simultaneous learning of image- and region-level representation for open-vocabulary object detection (OVD). Our approach combines the masked autoencoder (MAE) objective into the contrastive learning objective to improve the representation for localization tasks. Unlike standard MAE, we perform reconstruction in the joint image-text embedding space, rather than the pixel space as is customary with the classical MAE method, which causes the model to better learn region-level semantics. Moreover, we introduce Positional Embedding Dropout (PED) to address scale variation between image-text pretraining and detection finetuning by randomly dropping out the positional embeddings during pretraining. PED improves detection performance and enables the use of a frozen ViT backbone as a region classifier, preventing the forgetting of open-vocabulary knowledge during detection finetuning. On LVIS open-vocabulary detection benchmark, CFM-ViT achieves a state-of-the-art 33.9 AP$r$, surpassing the best approach by 7.6 points and achieves better zero-shot detection transfer. Finally, CFM-ViT acquires strong image-level representation, outperforming the state of the art on 8 out of 12 metrics on zero-shot image-text retrieval benchmarks.

Rapid advancements of large language models (LLMs) have enabled the processing, understanding, and generation of human-like text, with increasing integration into systems that touch our social sphere. Despite this success, these models can learn, perpetuate, and amplify harmful social biases. In this paper, we present a comprehensive survey of bias evaluation and mitigation techniques for LLMs. We first consolidate, formalize, and expand notions of social bias and fairness in natural language processing, defining distinct facets of harm and introducing several desiderata to operationalize fairness for LLMs. We then unify the literature by proposing three intuitive taxonomies, two for bias evaluation, namely metrics and datasets, and one for mitigation. Our first taxonomy of metrics for bias evaluation disambiguates the relationship between metrics and evaluation datasets, and organizes metrics by the different levels at which they operate in a model: embeddings, probabilities, and generated text. Our second taxonomy of datasets for bias evaluation categorizes datasets by their structure as counterfactual inputs or prompts, and identifies the targeted harms and social groups; we also release a consolidation of publicly-available datasets for improved access. Our third taxonomy of techniques for bias mitigation classifies methods by their intervention during pre-processing, in-training, intra-processing, and post-processing, with granular subcategories that elucidate research trends. Finally, we identify open problems and challenges for future work. Synthesizing a wide range of recent research, we aim to provide a clear guide of the existing literature that empowers researchers and practitioners to better understand and prevent the propagation of bias in LLMs.

In this work, we present an approach to identify sub-tasks within a demonstrated robot trajectory using language instructions. We identify these sub-tasks using language provided during demonstrations as guidance to identify sub-segments of a longer robot trajectory. Given a sequence of natural language instructions and a long trajectory consisting of image frames and discrete actions, we want to map an instruction to a smaller fragment of the trajectory. Unlike previous instruction following works which directly learn the mapping from language to a policy, we propose a language-conditioned change-point detection method to identify sub-tasks in a problem. Our approach learns the relationship between constituent segments of a long language command and corresponding constituent segments of a trajectory. These constituent trajectory segments can be used to learn subtasks or sub-goals for planning or options as demonstrated by previous related work. Our insight in this work is that the language-conditioned robot change-point detection problem is similar to the existing video moment retrieval works used to identify sub-segments within online videos. Through extensive experimentation, we demonstrate a $1.78_{\pm 0.82}\%$ improvement over a baseline approach in accurately identifying sub-tasks within a trajectory using our proposed method. Moreover, we present a comprehensive study investigating sample complexity requirements on learning this mapping, between language and trajectory sub-segments, to understand if the video retrieval-based methods are realistic in real robot scenarios.

We aim to better understand the emergence of `situational awareness' in large language models (LLMs). A model is situationally aware if it's aware that it's a model and can recognize whether it's currently in testing or deployment. Today's LLMs are tested for safety and alignment before they are deployed. An LLM could exploit situational awareness to achieve a high score on safety tests, while taking harmful actions after deployment. Situational awareness may emerge unexpectedly as a byproduct of model scaling. One way to better foresee this emergence is to run scaling experiments on abilities necessary for situational awareness. As such an ability, we propose `out-of-context reasoning' (in contrast to in-context learning). We study out-of-context reasoning experimentally. First, we finetune an LLM on a description of a test while providing no examples or demonstrations. At test time, we assess whether the model can pass the test. To our surprise, we find that LLMs succeed on this out-of-context reasoning task. Their success is sensitive to the training setup and only works when we apply data augmentation. For both GPT-3 and LLaMA-1, performance improves with model size. These findings offer a foundation for further empirical study, towards predicting and potentially controlling the emergence of situational awareness in LLMs. Code is available at: https://github.com/AsaCooperStickland/situational-awareness-evals.

LLMs have demonstrated great capabilities in various NLP tasks. Different entities can further improve the performance of those LLMs on their specific downstream tasks by fine-tuning LLMs. When several entities have similar interested tasks, but their data cannot be shared because of privacy concerns regulations, federated learning (FL) is a mainstream solution to leverage the data of different entities. However, fine-tuning LLMs in federated learning settings still lacks adequate support from existing FL frameworks because it has to deal with optimizing the consumption of significant communication and computational resources, data preparation for different tasks, and distinct information protection demands. This paper first discusses these challenges of federated fine-tuning LLMs, and introduces our package FS-LLM as a main contribution, which consists of the following components: (1) we build an end-to-end benchmarking pipeline, automizing the processes of dataset preprocessing, federated fine-tuning execution, and performance evaluation on federated LLM fine-tuning; (2) we provide comprehensive federated parameter-efficient fine-tuning algorithm implementations and versatile programming interfaces for future extension in FL scenarios with low communication and computation costs, even without accessing the full model; (3) we adopt several accelerating and resource-efficient operators for fine-tuning LLMs with limited resources and the flexible pluggable sub-routines for interdisciplinary study. We conduct extensive experiments to validate the effectiveness of FS-LLM and benchmark advanced LLMs with state-of-the-art parameter-efficient fine-tuning algorithms in FL settings, which also yields valuable insights into federated fine-tuning LLMs for the research community. To facilitate further research and adoption, we release FS-LLM at https://github.com/alibaba/FederatedScope/tree/llm.

This article introduces Mi-Go, a novel testing framework aimed at evaluating the performance and adaptability of general-purpose speech recognition machine learning models across diverse real-world scenarios. The framework leverages YouTube as a rich and continuously updated data source, accounting for multiple languages, accents, dialects, speaking styles, and audio quality levels. To demonstrate the effectiveness of the framework, the Whisper model, developed by OpenAI, was employed as a test object. The tests involve using a total of 124 YouTube videos to test all Whisper model versions. The results underscore the utility of YouTube as a valuable testing platform for speech recognition models, ensuring their robustness, accuracy, and adaptability to diverse languages and acoustic conditions. Additionally, by contrasting the machine-generated transcriptions against human-made subtitles, the Mi-Go framework can help pinpoint potential misuse of YouTube subtitles, like Search Engine Optimization.

Transformer based large language models with emergent capabilities are becoming increasingly ubiquitous in society. However, the task of understanding and interpreting their internal workings, in the context of adversarial attacks, remains largely unsolved. Gradient-based universal adversarial attacks have been shown to be highly effective on large language models and potentially dangerous due to their input-agnostic nature. This work presents a novel geometric perspective explaining universal adversarial attacks on large language models. By attacking the 117M parameter GPT-2 model, we find evidence indicating that universal adversarial triggers could be embedding vectors which merely approximate the semantic information in their adversarial training region. This hypothesis is supported by white-box model analysis comprising dimensionality reduction and similarity measurement of hidden representations. We believe this new geometric perspective on the underlying mechanism driving universal attacks could help us gain deeper insight into the internal workings and failure modes of LLMs, thus enabling their mitigation.

Automatic fact-checking plays a crucial role in combating the spread of misinformation. Large Language Models (LLMs) and Instruction-Following variants, such as InstructGPT and Alpaca, have shown remarkable performance in various natural language processing tasks. However, their knowledge may not always be up-to-date or sufficient, potentially leading to inaccuracies in fact-checking. To address this limitation, we propose combining the power of instruction-following language models with external evidence retrieval to enhance fact-checking performance. Our approach involves leveraging search engines to retrieve relevant evidence for a given input claim. This external evidence serves as valuable supplementary information to augment the knowledge of the pretrained language model. Then, we instruct-tune an open-sourced language model, called LLaMA, using this evidence, enabling it to predict the veracity of the input claim more accurately. To evaluate our method, we conducted experiments on two widely used fact-checking datasets: RAWFC and LIAR. The results demonstrate that our approach achieves state-of-the-art performance in fact-checking tasks. By integrating external evidence, we bridge the gap between the model's knowledge and the most up-to-date and sufficient context available, leading to improved fact-checking outcomes. Our findings have implications for combating misinformation and promoting the dissemination of accurate information on online platforms. Our released materials are accessible at: https://thcheung.github.io/factllama.

Recently, the remarkable capabilities of large language models (LLMs) have been illustrated across a variety of research domains such as natural language processing, computer vision, and molecular modeling. We extend this paradigm by utilizing LLMs for material property prediction by introducing our model Materials Informatics Transformer (MatInFormer). Specifically, we introduce a novel approach that involves learning the grammar of crystallography through the tokenization of pertinent space group information. We further illustrate the adaptability of MatInFormer by incorporating task-specific data pertaining to Metal-Organic Frameworks (MOFs). Through attention visualization, we uncover the key features that the model prioritizes during property prediction. The effectiveness of our proposed model is empirically validated across 14 distinct datasets, hereby underscoring its potential for high throughput screening through accurate material property prediction.

The demand for psychological counselling has grown significantly in recent years, particularly with the global outbreak of COVID-19, which has heightened the need for timely and professional mental health support. Online psychological counselling has emerged as the predominant mode of providing services in response to this demand. In this study, we propose the Psy-LLM framework, an AI-based assistive tool leveraging Large Language Models (LLMs) for question-answering in psychological consultation settings to ease the demand for mental health professions. Our framework combines pre-trained LLMs with real-world professional Q\&A from psychologists and extensively crawled psychological articles. The Psy-LLM framework serves as a front-end tool for healthcare professionals, allowing them to provide immediate responses and mindfulness activities to alleviate patient stress. Additionally, it functions as a screening tool to identify urgent cases requiring further assistance. We evaluated the framework using intrinsic metrics, such as perplexity, and extrinsic evaluation metrics, with human participant assessments of response helpfulness, fluency, relevance, and logic. The results demonstrate the effectiveness of the Psy-LLM framework in generating coherent and relevant answers to psychological questions. This article discusses the potential and limitations of using large language models to enhance mental health support through AI technologies.