This paper is a technical overview of DeepMind and Google's recent work on reinforcement learning for controlling commercial cooling systems. Building on expertise that began with cooling Google's data centers more efficiently, we recently conducted live experiments on two real-world facilities in partnership with Trane Technologies, a building management system provider. These live experiments had a variety of challenges in areas such as evaluation, learning from offline data, and constraint satisfaction. Our paper describes these challenges in the hope that awareness of them will benefit future applied RL work. We also describe the way we adapted our RL system to deal with these challenges, resulting in energy savings of approximately 9% and 13% respectively at the two live experiment sites.

我们提出了一个混合工业冷却系统模型，该模型将分析解决方案嵌入多物理模拟中。该模型设计用于增强学习（RL）应用程序，并平衡简单性与模拟保真度和解释性。该模型的忠诚度根据大规模冷却系统的现实世界数据进行了评估。接下来是一个案例研究，说明如何将模型用于RL研究。为此，我们开发了一个工业任务套件，该套件允许指定不同的问题设置和复杂性水平，并使用它来评估不同RL算法的性能。

训练基金会模型（例如GPT-3和Palm）可能非常昂贵，通常涉及数以万计的GPU连续运行数月。这些模型通常经过专门的群集培训，这些群集具有快速，均匀的互连，并使用精心设计的软件系统来支持数据并行性和模型/管道并行性。这样的专用集群可能是昂贵且难以获得的。我们可以相反，可以利用更大量的分散，异质和较低的互连计算？先前的工作研究了可以纯粹以数据并行方式训练的相对较小模型的异质，分散的设置重点。模型平行基础模型培训（例如威震天）的最先进的方案仅考虑均匀的数据中心设置。在本文中，我们介绍了第一个研究大型基础模型的研究，该模型在异质网络上的去中心化制度中进行了模型并行性。我们的主要技术贡献是一种调度算法，该算法将不同的计算“任务”在培训基础模型中分配给通过缓慢的异质网络连接的一组分散的GPU设备。我们提供了正式的成本模型，并进一步提出了一种有效的进化算法，以找到最佳分配策略。我们进行了广泛的实验，这些实验代表了使用现实世界网络测量模拟的地理分布设备进行学习的不同方案。在最极端的情况下，在跨越3大洲的8个不同的城市中，我们的方法比以前的最新培训系统（Megatron）快4.8倍。

AI正在经历范式转变，随着模型的兴起（例如Bert，Dall-E，GPT-3），这些模型经过大规模的数据训练，并且可以适应广泛的下游任务。我们称这些模型基础模型来强调其至关重要但不完整的特征。该报告提供了基础模型的机会和风险的详尽说明，包括其功能（例如语言，愿景，机器人技术，推理，人类互动）和技术原则（例如，模型架构，培训程序，数据，系统，安全，安全性，评估，理论）对其应用（例如法律，医疗保健，教育）和社会影响（例如不平等，滥用，经济和环境影响，法律和道德考虑）。尽管基础模型基于标准的深度学习和转移学习，但它们的规模导致了新的新兴能力，以及它们在许多任务中的有效性都激发了同质化。同质化提供了强大的杠杆作用，但要求谨慎，因为基础模型的缺陷均由下游的所有适应模型继承。尽管即将广泛地部署基础模型，但我们目前对它们的工作方式，失败以及由于其新兴属性的影响而缺乏清晰的了解。为了解决这些问题，我们认为基础模型的许多批判性研究都需要与他们的基本社会技术性质相称。

Large language models (LLMs) have been shown to be able to perform new tasks based on a few demonstrations or natural language instructions. While these capabilities have led to widespread adoption, most LLMs are developed by resource-rich organizations and are frequently kept from the public. As a step towards democratizing this powerful technology, we present BLOOM, a 176B-parameter open-access language model designed and built thanks to a collaboration of hundreds of researchers. BLOOM is a decoder-only Transformer language model that was trained on the ROOTS corpus, a dataset comprising hundreds of sources in 46 natural and 13 programming languages (59 in total). We find that BLOOM achieves competitive performance on a wide variety of benchmarks, with stronger results after undergoing multitask prompted finetuning. To facilitate future research and applications using LLMs, we publicly release our models and code under the Responsible AI License.

We introduce Performers, Transformer architectures which can estimate regular (softmax) full-rank-attention Transformers with provable accuracy, but using only linear (as opposed to quadratic) space and time complexity, without relying on any priors such as sparsity or low-rankness. To approximate softmax attentionkernels, Performers use a novel Fast Attention Via positive Orthogonal Random features approach (FAVOR+), which may be of independent interest for scalable kernel methods. FAVOR+ can also be used to efficiently model kernelizable attention mechanisms beyond softmax. This representational power is crucial to accurately compare softmax with other kernels for the first time on large-scale tasks, beyond the reach of regular Transformers, and investigate optimal attention-kernels. Performers are linear architectures fully compatible with regular Transformers and with strong theoretical guarantees: unbiased or nearly-unbiased estimation of the attention matrix, uniform convergence and low estimation variance. We tested Performers on a rich set of tasks stretching from pixel-prediction through text models to protein sequence modeling. We demonstrate competitive results with other examined efficient sparse and dense attention methods, showcasing effectiveness of the novel attention-learning paradigm leveraged by Performers.

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.

Machine learning (ML) has found broad applicability in quantum information science in topics as diverse as experimental design, state classification, and even studies on quantum foundations. Here, we experimentally realize an approach for defining custom prior distributions that are automatically tuned using ML for use with Bayesian quantum state estimation methods. Previously, researchers have looked to Bayesian quantum state tomography due to its unique advantages like natural uncertainty quantification, the return of reliable estimates under any measurement condition, and minimal mean-squared error. However, practical challenges related to long computation times and conceptual issues concerning how to incorporate prior knowledge most suitably can overshadow these benefits. Using both simulated and experimental measurement results, we demonstrate that ML-defined prior distributions reduce net convergence times and provide a natural way to incorporate both implicit and explicit information directly into the prior distribution. These results constitute a promising path toward practical implementations of Bayesian quantum state tomography.

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.

Adversarial attacks hamper the decision-making ability of neural networks by perturbing the input signal. The addition of calculated small distortion to images, for instance, can deceive a well-trained image classification network. In this work, we propose a novel attack technique called Sparse Adversarial and Interpretable Attack Framework (SAIF). Specifically, we design imperceptible attacks that contain low-magnitude perturbations at a small number of pixels and leverage these sparse attacks to reveal the vulnerability of classifiers. We use the Frank-Wolfe (conditional gradient) algorithm to simultaneously optimize the attack perturbations for bounded magnitude and sparsity with $O(1/\sqrt{T})$ convergence. Empirical results show that SAIF computes highly imperceptible and interpretable adversarial examples, and outperforms state-of-the-art sparse attack methods on the ImageNet dataset.