As language models (LMs) scale, they develop many novel behaviors, good and bad, exacerbating the need to evaluate how they behave. Prior work creates evaluations with crowdwork (which is time-consuming and expensive) or existing data sources (which are not always available). Here, we automatically generate evaluations with LMs. We explore approaches with varying amounts of human effort, from instructing LMs to write yes/no questions to making complex Winogender schemas with multiple stages of LM-based generation and filtering. Crowdworkers rate the examples as highly relevant and agree with 90-100% of labels, sometimes more so than corresponding human-written datasets. We generate 154 datasets and discover new cases of inverse scaling where LMs get worse with size. Larger LMs repeat back a dialog user's preferred answer ("sycophancy") and express greater desire to pursue concerning goals like resource acquisition and goal preservation. We also find some of the first examples of inverse scaling in RL from Human Feedback (RLHF), where more RLHF makes LMs worse. For example, RLHF makes LMs express stronger political views (on gun rights and immigration) and a greater desire to avoid shut down. Overall, LM-written evaluations are high-quality and let us quickly discover many novel LM behaviors.

As AI systems become more capable, we would like to enlist their help to supervise other AIs. We experiment with methods for training a harmless AI assistant through self-improvement, without any human labels identifying harmful outputs. The only human oversight is provided through a list of rules or principles, and so we refer to the method as 'Constitutional AI'. The process involves both a supervised learning and a reinforcement learning phase. In the supervised phase we sample from an initial model, then generate self-critiques and revisions, and then finetune the original model on revised responses. In the RL phase, we sample from the finetuned model, use a model to evaluate which of the two samples is better, and then train a preference model from this dataset of AI preferences. We then train with RL using the preference model as the reward signal, i.e. we use 'RL from AI Feedback' (RLAIF). As a result we are able to train a harmless but non-evasive AI assistant that engages with harmful queries by explaining its objections to them. Both the SL and RL methods can leverage chain-of-thought style reasoning to improve the human-judged performance and transparency of AI decision making. These methods make it possible to control AI behavior more precisely and with far fewer human labels.

我们研究语言模型是否可以评估自己主张的有效性，并预测他们能够正确回答的问题。我们首先表明，当以正确的格式提供时，较大的模型在多样化的多项选择和True/False问题上进行了很好的校准。因此，我们可以通过要求模型首先提出答案，然后评估其答案正确的概率“ p（true）”来对开放式采样任务进行自我评估。我们发现在各种任务中，P（true）的表现，校准和缩放令人鼓舞。当我们允许模型考虑自己的许多样本之前，在预测一种特定可能性的有效性之前，自我评估的性能进一步改善。接下来，我们研究是否可以培训模型来预测“ P（ik）”，即“我知道”问题的概率，而无需参考任何特定提出的答案。模型在预测P（IK）方面表现良好，并且在跨任务中部分概括，尽管它们在新任务上的P（IK）校准方面遇到了困难。预测的p（IK）概率在存在相关的原始材料的情况下以及对数学单词问题解决方案的提示也适当增加。我们希望这些观察结果为培训更诚实的模型提供了基础，并研究了诚实对模型模仿人类写作以外的其他目标培训的案例的普遍性。

我们建议采用统计回归作为投影操作员，以使数据驱动以数据为基础的Mori-Zwanzig形式主义中的运营商学习。我们提出了一种原则性方法，用于为任何回归模型提取Markov和内存操作员。我们表明，线性回归的选择导致了基于Mori的投影操作员最近提出的数据驱动的学习算法，这是一种高阶近似Koopman学习方法。我们表明，更具表现力的非线性回归模型自然填补了高度理想化和计算有效的MORI投影操作符和最佳迄今为止计算上最佳的Zwanzig投影仪之间的差距。我们进行了数值实验，并提取了一系列基于回归的投影的运算符，包括线性，多项式，样条和基于神经网络的回归，随着回归模型的复杂性的增加而显示出渐进的改进。我们的命题提供了一个通用框架来提取内存依赖性校正，并且可以轻松地应用于文献中固定动力学系统的一系列数据驱动的学习方法。

While the capabilities of autonomous systems have been steadily improving in recent years, these systems still struggle to rapidly explore previously unknown environments without the aid of GPS-assisted navigation. The DARPA Subterranean (SubT) Challenge aimed to fast track the development of autonomous exploration systems by evaluating their performance in real-world underground search-and-rescue scenarios. Subterranean environments present a plethora of challenges for robotic systems, such as limited communications, complex topology, visually-degraded sensing, and harsh terrain. The presented solution enables long-term autonomy with minimal human supervision by combining a powerful and independent single-agent autonomy stack, with higher level mission management operating over a flexible mesh network. The autonomy suite deployed on quadruped and wheeled robots was fully independent, freeing the human supervision to loosely supervise the mission and make high-impact strategic decisions. We also discuss lessons learned from fielding our system at the SubT Final Event, relating to vehicle versatility, system adaptability, and re-configurable communications.

Research has shown that climate change creates warmer temperatures and drier conditions, leading to longer wildfire seasons and increased wildfire risks in the United States. These factors have in turn led to increases in the frequency, extent, and severity of wildfires in recent years. Given the danger posed by wildland fires to people, property, wildlife, and the environment, there is an urgency to provide tools for effective wildfire management. Early detection of wildfires is essential to minimizing potentially catastrophic destruction. In this paper, we present our work on integrating multiple data sources in SmokeyNet, a deep learning model using spatio-temporal information to detect smoke from wildland fires. Camera image data is integrated with weather sensor measurements and processed by SmokeyNet to create a multimodal wildland fire smoke detection system. We present our results comparing performance in terms of both accuracy and time-to-detection for multimodal data vs. a single data source. With a time-to-detection of only a few minutes, SmokeyNet can serve as an automated early notification system, providing a useful tool in the fight against destructive wildfires.

System identification, also known as learning forward models, transfer functions, system dynamics, etc., has a long tradition both in science and engineering in different fields. Particularly, it is a recurring theme in Reinforcement Learning research, where forward models approximate the state transition function of a Markov Decision Process by learning a mapping function from current state and action to the next state. This problem is commonly defined as a Supervised Learning problem in a direct way. This common approach faces several difficulties due to the inherent complexities of the dynamics to learn, for example, delayed effects, high non-linearity, non-stationarity, partial observability and, more important, error accumulation when using bootstrapped predictions (predictions based on past predictions), over large time horizons. Here we explore the use of Reinforcement Learning in this problem. We elaborate on why and how this problem fits naturally and sound as a Reinforcement Learning problem, and present some experimental results that demonstrate RL is a promising technique to solve these kind of problems.

By transferring knowledge from large, diverse, task-agnostic datasets, modern machine learning models can solve specific downstream tasks either zero-shot or with small task-specific datasets to a high level of performance. While this capability has been demonstrated in other fields such as computer vision, natural language processing or speech recognition, it remains to be shown in robotics, where the generalization capabilities of the models are particularly critical due to the difficulty of collecting real-world robotic data. We argue that one of the keys to the success of such general robotic models lies with open-ended task-agnostic training, combined with high-capacity architectures that can absorb all of the diverse, robotic data. In this paper, we present a model class, dubbed Robotics Transformer, that exhibits promising scalable model properties. We verify our conclusions in a study of different model classes and their ability to generalize as a function of the data size, model size, and data diversity based on a large-scale data collection on real robots performing real-world tasks. The project's website and videos can be found at robotics-transformer.github.io

Surgery is the only viable treatment for cataract patients with visual acuity (VA) impairment. Clinically, to assess the necessity of cataract surgery, accurately predicting postoperative VA before surgery by analyzing multi-view optical coherence tomography (OCT) images is crucially needed. Unfortunately, due to complicated fundus conditions, determining postoperative VA remains difficult for medical experts. Deep learning methods for this problem were developed in recent years. Although effective, these methods still face several issues, such as not efficiently exploring potential relations between multi-view OCT images, neglecting the key role of clinical prior knowledge (e.g., preoperative VA value), and using only regression-based metrics which are lacking reference. In this paper, we propose a novel Cross-token Transformer Network (CTT-Net) for postoperative VA prediction by analyzing both the multi-view OCT images and preoperative VA. To effectively fuse multi-view features of OCT images, we develop cross-token attention that could restrict redundant/unnecessary attention flow. Further, we utilize the preoperative VA value to provide more information for postoperative VA prediction and facilitate fusion between views. Moreover, we design an auxiliary classification loss to improve model performance and assess VA recovery more sufficiently, avoiding the limitation by only using the regression metrics. To evaluate CTT-Net, we build a multi-view OCT image dataset collected from our collaborative hospital. A set of extensive experiments validate the effectiveness of our model compared to existing methods in various metrics. Code is available at: https://github.com/wjh892521292/Cataract OCT.

Deep Reinforcement Learning (RL) agents are susceptible to adversarial noise in their observations that can mislead their policies and decrease their performance. However, an adversary may be interested not only in decreasing the reward, but also in modifying specific temporal logic properties of the policy. This paper presents a metric that measures the exact impact of adversarial attacks against such properties. We use this metric to craft optimal adversarial attacks. Furthermore, we introduce a model checking method that allows us to verify the robustness of RL policies against adversarial attacks. Our empirical analysis confirms (1) the quality of our metric to craft adversarial attacks against temporal logic properties, and (2) that we are able to concisely assess a system's robustness against attacks.