抽象推理是智能系统的关键能力。大型语言模型在抽象推理任务上实现了高度的性能，但表现出许多缺陷。但是，人类的抽象推理也是不完美的，并且取决于我们对推理问题内容的知识和信念。例如，人类对在日常情况下基于逻辑规则的逻辑规则比关于抽象属性的任意规则更可靠地理解。语言模型的培训经验类似地赋予了他们先前的期望，这些期望反映了人类的知识和信念。因此，我们假设语言模型会显示出类似人类的内容对抽象推理问题的影响。我们在三个逻辑推理任务中探讨了这一假设：自然语言推论，判断三段论的逻辑有效性和ison选择任务（Wason，1968）。我们发现，最新的大语言模型（具有7或700亿个参数； Hoffman等，2022）反映了这些任务中人类在人类中观察到的许多相同模式 - 像人类一样，模型对可信情况的理由更有效地理由不现实或抽象的。我们的发现对理解这些认知效应以及有助于语言模型表现的因素具有影响。

我们介绍了Sparrow，这是一个寻求信息的对话代理，与提示的语言模型基线相比，训练有素，更有帮助，正确和无害。我们使用从人类反馈中的强化学习来培训我们的模型，以帮助人类评估者判断代理人的行为。首先，为了使我们的代理人更有帮助和无害，我们将良好对话的要求分解为代理人应遵循的自然语言规则，并分别向评估者询问每个规则。我们证明，这种崩溃使我们能够收集对代理行为的更多针对性的人类判断，并允许更有效的规则条件奖励模型。其次，我们的代理商在收集对模型声明的偏好判决时提供了支持事实主张的来源的证据。对于事实问题，麻雀提供的证据支持了78％的时间。比基线比基线更享受麻雀，同时对人类的对抗性探测更具弹性，在探测时只有8％的时间违反了我们的规则。最后，我们进行了广泛的分析，表明尽管我们的模型学会遵守我们的规则，但它可以表现出分布偏见。

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.

Many real-world applications of language models (LMs), such as code autocomplete and writing assistance, involve human-LM interaction, but the main LM benchmarks are non-interactive, where a system produces output without human intervention. To evaluate human-LM interaction, we develop a framework, Human-AI Language-based Interaction Evaluation (H-LINE), that expands non-interactive evaluation along three dimensions, capturing (i) the interactive process, not only the final output; (ii) the first-person subjective experience, not just a third-party assessment; and (iii) notions of preference beyond quality. We then design five tasks ranging from goal-oriented to open-ended to capture different forms of interaction. On four state-of-the-art LMs (three variants of OpenAI's GPT-3 and AI21's J1-Jumbo), we find that non-interactive performance does not always result in better human-LM interaction and that first-person and third-party metrics can diverge, suggesting the importance of examining the nuances of human-LM interaction.

语言理解的概率模型是可解释和结构化的，例如隐喻理解的模型描述了有关潜在主题和特征的推论。但是，这些模型是为特定任务手动设计的。大型语言模型（LLMS）可以通过内在的学习来执行许多任务，但它们缺乏概率模型的清晰结构。在本文中，我们使用经过思考的提示将概率模型的结构引入LLMS。这些提示导致该模型推断潜在变量和有关其关系的理由，以选择隐喻的适当释义。所选择的潜在变量和关系是由认知心理学理解理论得出的。我们将这些提示应用于GPT-3的两个最大版本，并表明它们可以改善释义选择。

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.

鉴于大型语言模型的广泛能力，应该有可能朝着一般的文本的助手工作，这些助手与人类价值一致，这意味着它是有帮助，诚实的和无害的。在此方向上的初始遗传，我们研究简单的基线技术和评估，例如提示。我们发现，从模型规模增加适度的干预措施的好处，概括为各种对准评估，并不会损害大型模型的性能。接下来，我们调查与对齐，比较仿制，二进制歧视和排名偏好建模相关的几个培训目标的缩放趋势。我们发现排名优先级模型比模仿学习更好地表现得多，并且通常以模型大小更有利地缩放。相比之下，二进制歧视通常与模仿学习非常类似地执行和缩放。最后，我们研究了一种“偏好模型预训练阶段的培训阶段，其目的是在对人偏好的芬明时提高样本效率。

Large language models (LLMs) have demonstrated impressive capabilities in natural language understanding and generation, but the quality bar for medical and clinical applications is high. Today, attempts to assess models' clinical knowledge typically rely on automated evaluations on limited benchmarks. There is no standard to evaluate model predictions and reasoning across a breadth of tasks. To address this, we present MultiMedQA, a benchmark combining six existing open question answering datasets spanning professional medical exams, research, and consumer queries; and HealthSearchQA, a new free-response dataset of medical questions searched online. We propose a framework for human evaluation of model answers along multiple axes including factuality, precision, possible harm, and bias. In addition, we evaluate PaLM (a 540-billion parameter LLM) and its instruction-tuned variant, Flan-PaLM, on MultiMedQA. Using a combination of prompting strategies, Flan-PaLM achieves state-of-the-art accuracy on every MultiMedQA multiple-choice dataset (MedQA, MedMCQA, PubMedQA, MMLU clinical topics), including 67.6% accuracy on MedQA (US Medical License Exam questions), surpassing prior state-of-the-art by over 17%. However, human evaluation reveals key gaps in Flan-PaLM responses. To resolve this we introduce instruction prompt tuning, a parameter-efficient approach for aligning LLMs to new domains using a few exemplars. The resulting model, Med-PaLM, performs encouragingly, but remains inferior to clinicians. We show that comprehension, recall of knowledge, and medical reasoning improve with model scale and instruction prompt tuning, suggesting the potential utility of LLMs in medicine. Our human evaluations reveal important limitations of today's models, reinforcing the importance of both evaluation frameworks and method development in creating safe, helpful LLM models for clinical applications.

随着人工智能系统变得越来越强大和普遍，人们对机器的道德或缺乏道德的关注变得越来越关注。然而，向机器讲授道德是一项艰巨的任务，因为道德仍然是人类中最激烈的争论问题之一，更不用说AI了。但是，部署到数百万用户的现有AI系统已经在做出充满道德影响的决策，这构成了一个看似不可能的挑战：教学机器的道德意义，而人类继续努力努力。为了探索这一挑战，我们介绍了Delphi，这是一个基于深层神经网络的实验框架，直接训练了描述性道德判断，例如，“帮助朋友”通常是不错的，而“帮助朋友传播假新闻”不是。经验结果提供了对机器伦理的承诺和局限性的新见解。面对新的道德情况，德尔菲（Delphi）表现出强大的概括能力，而现成的神经网络模型表现出明显差的判断，包括不公正的偏见，证实了对明确教学机器的道德意义的必要性。然而，德尔菲并不完美，表现出对普遍性偏见和不一致的敏感性。尽管如此，我们还是展示了不完美的Delphi的积极用例，包括在其他不完美的AI系统中将其用作组件模型。重要的是，我们根据著名的道德理论来解释Delphi的运营化，这使我们提出了重要的未来研究问题。

最近已被证明大型语言模型在各种任务集中获得合理的零射普通化（Brown等，2020）。它已经假设这是语言模型的隐式多任务学习的结果，在语言模型中的预押（Radford等，2019）。可以通过明确的多任务学习直接引起零拍常规化？为了以缩放测试这个问题，我们开发一个系统，以便轻松地将任何自然语言任务映射到人类可读的提示表单中。我们转换一组大量的监督数据集，每个数据集都有多个提示，具有不同的措辞。这些提示的数据集允许基准测试模型执行完全看不见的任务的能力。我们介绍了一个普拉克尔编码器 - 解码器模型（Raffel等，2020; Lester等，2021），覆盖各种任务。该模型在多个标准数据集中达到强大的零点性能，通常优于其尺寸的型号超过16倍。此外，我们的方法对来自Big-替补基准测试的任务子集具有强烈性能，优于其尺寸的6倍。所有提示和培训的型号都可以在https://github.com/ bigscience-workshop / protectsource / httpsource / https：//huggingface.co/bigscience/t0pp。

There has been a recent resurgence in the area of explainable artificial intelligence as researchers and practitioners seek to make their algorithms more understandable. Much of this research is focused on explicitly explaining decisions or actions to a human observer, and it should not be controversial to say that looking at how humans explain to each other can serve as a useful starting point for explanation in artificial intelligence. However, it is fair to say that most work in explainable artificial intelligence uses only the researchers' intuition of what constitutes a 'good' explanation. There exists vast and valuable bodies of research in philosophy, psychology, and cognitive science of how people define, generate, select, evaluate, and present explanations, which argues that people employ certain cognitive biases and social expectations towards the explanation process. This paper argues that the field of explainable artificial intelligence should build on this existing research, and reviews relevant papers from philosophy, cognitive psychology/science, and social psychology, which study these topics. It draws out some important findings, and discusses ways that these can be infused with work on explainable artificial intelligence.

关于人类阅读的研究长期以来一直记录在阅读行为表明特定于任务的效果，但是建立一个通用模型来预测人类在给定任务中将显示什么的通用模型。我们介绍了Neat，这是人类阅读中注意力分配的计算模型，基于人类阅读优化了一项任务中关注经济和成功之间的权衡。我们的模型是使用当代神经网络建模技术实施的，并对注意力分配的分配方式在不同任务中如何变化做出明确的测试预测。我们在一项针对阅读理解任务的两个版本的眼影研究中对此进行了测试，发现我们的模型成功说明了整个任务的阅读行为。因此，我们的工作提供了证据表明，任务效果可以建模为对任务需求的最佳适应。

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。

Reasoning is a fundamental aspect of human intelligence that plays a crucial role in activities such as problem solving, decision making, and critical thinking. In recent years, large language models (LLMs) have made significant progress in natural language processing, and there is observation that these models may exhibit reasoning abilities when they are sufficiently large. However, it is not yet clear to what extent LLMs are capable of reasoning. This paper provides a comprehensive overview of the current state of knowledge on reasoning in LLMs, including techniques for improving and eliciting reasoning in these models, methods and benchmarks for evaluating reasoning abilities, findings and implications of previous research in this field, and suggestions on future directions. Our aim is to provide a detailed and up-to-date review of this topic and stimulate meaningful discussion and future work.

People constantly use language to learn about the world. Computational linguists have capitalized on this fact to build large language models (LLMs) that acquire co-occurrence-based knowledge from language corpora. LLMs achieve impressive performance on many tasks, but the robustness of their world knowledge has been questioned. Here, we ask: do LLMs acquire generalized knowledge about real-world events? Using curated sets of minimal sentence pairs (n=1215), we tested whether LLMs are more likely to generate plausible event descriptions compared to their implausible counterparts. We found that LLMs systematically distinguish possible and impossible events (The teacher bought the laptop vs. The laptop bought the teacher) but fall short of human performance when distinguishing likely and unlikely events (The nanny tutored the boy vs. The boy tutored the nanny). In follow-up analyses, we show that (i) LLM scores are driven by both plausibility and surface-level sentence features, (ii) LLMs generalize well across syntactic sentence variants (active vs passive) but less well across semantic sentence variants (synonymous sentences), (iii) some, but not all LLM deviations from ground-truth labels align with crowdsourced human judgments, and (iv) explicit event plausibility information emerges in middle LLM layers and remains high thereafter. Overall, our analyses reveal a gap in LLMs' event knowledge, highlighting their limitations as generalized knowledge bases. We conclude by speculating that the differential performance on impossible vs. unlikely events is not a temporary setback but an inherent property of LLMs, reflecting a fundamental difference between linguistic knowledge and world knowledge in intelligent systems.

The recent advent of large language models - large neural networks trained on a simple predictive objective over a massive corpus of natural language - has reinvigorated debate over whether human cognitive capacities might emerge in such generic models given sufficient training data. Of particular interest is the ability of these models to reason about novel problems zero-shot, without any direct training on those problems. In human cognition, this capacity is closely tied to an ability to reason by analogy. Here, we performed a direct comparison between human reasoners and a large language model (GPT-3) on a range of analogical tasks, including a novel text-based matrix reasoning task closely modeled on Raven's Progressive Matrices. We found that GPT-3 displayed a surprisingly strong capacity for abstract pattern induction, matching or even surpassing human capabilities in most settings. Our results indicate that large language models such as GPT-3 have acquired an emergent ability to find zero-shot solutions to a broad range of analogy problems.

语言是协调问题的强大解决方案：他们提供了稳定的，有关我们所说的单词如何对应于我们头脑中的信仰和意图的共同期望。然而，在变量和非静止社会环境中的语言使用需要语言表征来灵活：旧词在飞行中获取新的临时或合作伙伴特定含义。在本文中，我们介绍了柴（通过推理的连续分层适应），一个分层贝叶斯的协调理论和会议组织，旨在在这两个基本观察之间调和长期张力。我们认为，沟通的中央计算问题不仅仅是传输，如在经典配方中，而是在多个时间尺度上持续学习和适应。合作伙伴特定的共同点迅速出现在数型互动中的社会推论中，而社群范围内的社会公约是稳定的前锋，这些前锋已经抽象出与多个合作伙伴的互动。我们展示了新的实证数据，展示了我们的模型为多个现象提供了对先前账户挑战的计算基础：（1）与同一合作伙伴的重复互动的更有效的参考表达的融合（2）将合作伙伴特定的共同基础转移到陌生人，并（3）交际范围的影响最终会形成。

我们建议并探讨可以将语言模型作为社会科学研究中特定人类亚人群的有效代理进行研究的可能性。人工智能工具的实践和研究应用有时受到有问题的偏见（例如种族主义或性别歧视）的限制，这些偏见通常被视为模型的统一特性。我们表明，一个这样的工具中的“算法偏见”（GPT-3语言模型）既是细粒度又是人口统计相关的，这意味着适当的条件会导致其准确地仿真来自各种人类的响应分布亚组。我们将此属性称为“算法忠诚度”，并在GPT-3中探索其范围。我们通过将模型调节在美国进行的多项大型调查中的数千个社会人口统计背景故事中调节，从而创建“硅样本”。然后，我们比较硅和人类样品，以证明GPT-3中包含的信息远远超出了表面相似性。它是细微的，多方面的，并反映了特征人类态度的思想，态度和社会文化背景之间的复杂相互作用。我们建议，具有足够算法的忠诚度的语言模型构成了一种新颖而有力的工具，可以促进各种学科的人类和社会的理解。

Pragmatics is an essential part of communication, but it remains unclear what mechanisms underlie human pragmatic communication and whether NLP systems capture pragmatic language understanding. To investigate both these questions, we perform a fine-grained comparison of language models and humans on seven pragmatic phenomena, using zero-shot prompting on an expert-curated set of English materials. We ask whether models (1) select pragmatic interpretations of speaker utterances, (2) make similar error patterns as humans, and (3) use similar linguistic cues as humans to solve the tasks. We find that the largest models achieve high accuracy and match human error patterns: within incorrect responses, models favor the literal interpretation of an utterance over heuristic-based distractors. We also find evidence that models and humans are sensitive to similar linguistic cues. Our results suggest that even paradigmatic pragmatic phenomena may be solved without explicit representations of other agents' mental states, and that artificial models can be used to gain mechanistic insights into human pragmatic processing.

推理是人类认知和智力的关键支柱。在过去的十年中，我们目睹了自然语言处理的巨大收益和大型语言模型的前所未有的缩放。最近的工作表征了很少射击技术的能力，例如思想链，可以在大语言模型中模仿人类的推理。这个标志性的功能很少，连同不断扩展的语言模型相结合，打开了解决各种任务的可能性的远景，例如数学单词问题，代码完成和常识性推理。促使思想链（COT）通过提供中间步骤并敦促模型遵循相同的过程，从而进一步推动了模型的性能。尽管具有令人信服的性能，但在这些模型中推理能力的起源却很少探索。这项工作启动了对大语言模型中推理机制的更深入了解的初步步骤。我们的工作围绕查询模型，同时在提示中控制除一个组件以外的所有组件外：符号，模式和文本。然后，我们分析查询之间的性能差异。我们的结果表明，在提示中存在事实模式对于COT的成功并不是必需的。尽管如此，我们从经验上表明，仅依靠模式也不足以获得高质量的结果。我们认为文本具有常识性知识和意义。我们详尽的经验分析提供了定性的例子，说明了文本和模式之间的共生关系。这种对COT的系统理解使我们能够设计简洁的思想链，被称为CCOT，在其中修剪文本和模式只能保留其关键角色，同时以PAR或更高的求解任务率交付。