Safety and responsibility evaluations of advanced AI models are a critical but developing field of research and practice. In the development of Google DeepMind's advanced AI models, we innovated on and applied a broad set of approaches to safety evaluation. In this report, we summarise and share elements of our evolving approach as well as lessons learned for a broad audience. Key lessons learned include: First, theoretical underpinnings and frameworks are invaluable to organise the breadth of risk domains, modalities, forms, metrics, and goals. Second, theory and practice of safety evaluation development each benefit from collaboration to clarify goals, methods and challenges, and facilitate the transfer of insights between different stakeholders and disciplines. Third, similar key methods, lessons, and institutions apply across the range of concerns in responsibility and safety - including established and emerging harms. For this reason it is important that a wide range of actors working on safety evaluation and safety research communities work together to develop, refine and implement novel evaluation approaches and best practices, rather than operating in silos. The report concludes with outlining the clear need to rapidly advance the science of evaluations, to integrate new evaluations into the development and governance of AI, to establish scientifically-grounded norms and standards, and to promote a robust evaluation ecosystem.

Foundation models have become invaluable in advancing the medical field. Despite their promise, the strategic deployment of LLMs for effective utility in complex medical tasks remains an open question. Our novel framework, Medical Decision-making Agents (MDAgents) aims to address this gap by automatically assigning the effective collaboration structure for LLMs. Assigned solo or group collaboration structure is tailored to the complexity of the medical task at hand, emulating real-world medical decision making processes. We evaluate our framework and baseline methods with state-of-the-art LLMs across a suite of challenging medical benchmarks: MedQA, MedMCQA, PubMedQA, DDXPlus, PMC-VQA, Path-VQA, and MedVidQA, achieving the best performance in 5 out of 7 benchmarks that require an understanding of multi-modal medical reasoning. Ablation studies reveal that MDAgents excels in adapting the number of collaborating agents to optimize efficiency and accuracy, showcasing its robustness in diverse scenarios. We also explore the dynamics of group consensus, offering insights into how collaborative agents could behave in complex clinical team dynamics. Our code can be found at https://github.com/mitmedialab/MDAgents.

While vertical thinking relies on logical and commonsense reasoning, lateral thinking requires systems to defy commonsense associations and overwrite them through unconventional thinking. Lateral thinking has been shown to be challenging for current models but has received little attention. A recent benchmark, BRAINTEASER, aims to evaluate current models' lateral thinking ability in a zero-shot setting. In this paper, we split the original benchmark to also support fine-tuning setting and present SemEval Task 9: BRAIN-TEASER(S), the first task at this competition designed to test the system's reasoning and lateral thinking ability. As a popular task, BRAINTEASER(S)'s two subtasks receive 483 team submissions from 182 participants during the competition. This paper provides a fine-grained system analysis of the competition results, together with a reflection on what this means for the ability of the systems to reason laterally. We hope that the BRAINTEASER(S) subtasks and findings in this paper can stimulate future work on lateral thinking and robust reasoning by computational models.

Content Warning: This paper contains examples of misgendering and erasure that could be offensive and potentially triggering. Misgendering, the act of incorrectly addressing someone's gender, inflicts serious harm and is pervasive in everyday technologies, yet there is a notable lack of research to combat it. We are the first to address this lack of research into interventions for misgendering by conducting a survey of gender-diverse individuals in the US to understand perspectives about automated interventions for text-based misgendering. Based on survey insights on the prevalence of misgendering, desired solutions, and associated concerns, we introduce a misgendering interventions task and evaluation dataset, MisgenderMender. We define the task with two sub-tasks: (i) detecting misgendering, followed by (ii) correcting misgendering where misgendering is present in domains where editing is appropriate. MisgenderMender comprises 3790 instances of social media content and LLM-generations about non-cisgender public figures, annotated for the presence of misgendering, with additional annotations for correcting misgendering in LLM-generated text. Using this dataset, we set initial benchmarks by evaluating existing NLP systems and highlighting challenges for future models to address. We release the full dataset, code, and demo at https://tamannahossainkay.github.io/misgendermender/.

Large language models (LLMs) have shown significant potential for robotics applications, particularly task planning, by harnessing their language comprehension and text generation capabilities. However, in applications such as household robotics, a critical gap remains in the personalization of these models to individual user preferences. We introduce LLM-Personalize, a novel framework with an optimization pipeline designed to personalize LLM planners for household robotics. Our LLM-Personalize framework features an LLM planner that performs iterative planning in multi-room, partially-observable household scenarios, making use of a scene graph constructed with local observations. The generated plan consists of a sequence of high-level actions which are subsequently executed by a controller. Central to our approach is the optimization pipeline, which combines imitation learning and iterative self-training to personalize the LLM planner. In particular, the imitation learning phase performs initial LLM alignment from demonstrations, and bootstraps the model to facilitate effective iterative self-training, which further explores and aligns the model to user preferences. We evaluate LLM-Personalize on Housekeep, a challenging simulated real-world 3D benchmark for household rearrangements, and show that LLM-Personalize achieves more than a 30 percent increase in success rate over existing LLM planners, showcasing significantly improved alignment with human preferences. Project page: https://donggehan.github.io/projectllmpersonalize/.

Next Point-of-interest (POI) recommendation provides valuable suggestions for users to explore their surrounding environment. Existing studies rely on building recommendation models from large-scale users' check-in data, which is task-specific and needs extensive computational resources. Recently, the pretrained large language models (LLMs) have achieved significant advancements in various NLP tasks and have also been investigated for recommendation scenarios. However, the generalization abilities of LLMs still are unexplored to address the next POI recommendations, where users' geographical movement patterns should be extracted. Although there are studies that leverage LLMs for next-item recommendations, they fail to consider the geographical influence and sequential transitions. Hence, they cannot effectively solve the next POI recommendation task. To this end, we design novel prompting strategies and conduct empirical studies to assess the capability of LLMs, e.g., ChatGPT, for predicting a user's next check-in. Specifically, we consider several essential factors in human movement behaviors, including user geographical preference, spatial distance, and sequential transitions, and formulate the recommendation task as a ranking problem. Through extensive experiments on two widely used real-world datasets, we derive several key findings. Empirical evaluations demonstrate that LLMs have promising zero-shot recommendation abilities and can provide accurate and reasonable predictions. We also reveal that LLMs cannot accurately comprehend geographical context information and are sensitive to the order of presentation of candidate POIs, which shows the limitations of LLMs and necessitates further research on robust human mobility reasoning mechanisms.

While Large Language Models (LLMs) exhibit remarkable capabilities in zero-shot and few-shot scenarios, they often require computationally prohibitive sizes. Conversely, smaller Masked Language Models (MLMs) like BERT and RoBERTa achieve state-of-the-art results through fine-tuning but struggle with extending to few-shot and zero-shot settings due to their architectural constraints. Hence, we propose Statement-Tuning, a technique that models discriminative tasks as a set of finite statements and trains an Encoder model to discriminate between the potential statements to determine the label. We do Statement-Tuning on multiple tasks to enable cross-task generalization. Experimental results demonstrate that Statement Tuning achieves competitive performance compared to state-of-the-art LLMs with significantly fewer parameters. Moreover, the study investigates the impact of several design choices on few-shot and zero-shot generalization, revealing that Statement Tuning can achieve sufficient performance with modest training data and benefits from task and statement diversity for unseen task generalizability.

Zero-shot cross-lingual knowledge transfer enables a multilingual pretrained language model, finetuned on a task in one language, make predictions for this task in other languages. While being broadly studied for natural language understanding tasks, the described setting is understudied for generation. Previous works notice a frequent problem of generation in a wrong language and propose approaches to address it, usually using mT5 as a backbone model. In this work we compare various approaches proposed from the literature in unified settings, also including alternative backbone models, namely mBART and NLLB-200. We first underline the importance of tuning learning rate used for finetuning, which helps to substantially alleviate the problem of generation in the wrong language. Then, we show that with careful learning rate tuning, the simple full finetuning of the model acts as a very strong baseline and alternative approaches bring only marginal improvements. Finally, we find that mBART performs similarly to mT5 of the same size, and NLLB-200 can be competitive in some cases. Our final zero-shot models reach the performance of the approach based on data translation which is usually considered as an upper baseline for zero-shot cross-lingual transfer in generation.

While artificial intelligence has the potential to process vast amounts of data, generate new insights, and unlock greater productivity, its widespread adoption may entail unforeseen consequences. We identify conditions under which AI, by reducing the cost of access to certain modes of knowledge, can paradoxically harm public understanding. While large language models are trained on vast amounts of diverse data, they naturally generate output towards the 'center' of the distribution. This is generally useful, but widespread reliance on recursive AI systems could lead to a process we define as "knowledge collapse", and argue this could harm innovation and the richness of human understanding and culture. However, unlike AI models that cannot choose what data they are trained on, humans may strategically seek out diverse forms of knowledge if they perceive them to be worthwhile. To investigate this, we provide a simple model in which a community of learners or innovators choose to use traditional methods or to rely on a discounted AI-assisted process and identify conditions under which knowledge collapse occurs. In our default model, a 20% discount on AI-generated content generates public beliefs 2.3 times further from the truth than when there is no discount. An empirical approach to measuring the distribution of LLM outputs is provided in theoretical terms and illustrated through a specific example comparing the diversity of outputs across different models and prompting styles. Finally, based on the results, we consider further research directions to counteract such outcomes.

Sense of ownership in writing confines our investment of thoughts, time, and contribution, leading to attachment to the output. However, using writing assistants introduces a mental dilemma, as some content isn't directly our creation. For instance, we tend to credit Large Language Models (LLMs) more in creative tasks, even though all tasks are equal for them. Additionally, while we may not claim complete ownership of LLM-generated content, we freely claim authorship. We conduct a short survey to examine these issues and understand underlying cognitive processes in order to gain a better knowledge of human-computer interaction in writing and improve writing aid systems.