Sam Altman思想档案馆
Sam Altman 肖像

文献 声明 alt-010 机器译文 · 未审校

GPT-4(研究发布)

GPT-4 (research announcement)

来源 OpenAI · 英文原文出处

STATE2 中文可读 · 机器译文,未经人工审校 证据等级 A
中文为默认阅读视图

机器译文 本页中文为机器逐句翻译(hy4-flash),未经人工审校,仅供阅读参考;引用与考据请以英文原文为准。 逐句机译与英文原文 1:1 存档(11 段);未经人工审校。

# GPT‑4 我们创造了 GPT‑4,这是 OpenAI 扩展深度学习努力的最新里程碑。GPT‑4 是一个大型多模态模型(接受图像和文本输入、输出文本),虽然在许多现实场景中能力不及人类,但在各种专业和学术基准上表现出人类水平的性能。例如,它模拟参加律师资格考试(bar exam)的成绩约为考生前 10%;相比之下,GPT‑3.5 的成绩约为后 10%。我们花了 6 个月时间,利用对抗性测试项目以及 ChatGPT 的经验教训,对 GPT‑4 进行了迭代对齐(aligning),在事实性、可控性(steerability)和拒绝越出护栏方面取得了我们有史以来最好的结果(尽管远非完美)。 过去两年,我们重建了整个深度学习技术栈,并与 Azure 一起,从零开始为我们的工作负载共同设计了一台超级计算机。一年前,我们把 GPT‑3.5 作为该系统的第一次「测试运行」进行训练。我们发现并修复了一些错误,并改进了理论基础。结果,我们的 GPT‑4 训练运行(至少对我们而言)前所未有的稳定,成为我们第一个能够提前准确预测其训练性能的大型模型。随着我们继续专注于可靠的扩展,我们旨在打磨方法论,帮助越来越早地预测和准备未来的能力——我们认为这对安全至关重要。 我们通过 ChatGPT 和 API 发布 GPT‑4 的文本输入能力(附带等候名单)。为了准备让图像输入能力更广泛地可用,我们起初与单一伙伴密切合作。我们还开源了 OpenAI Evals——我们用于自动评估 AI 模型性能的框架,让任何人都能报告我们模型的缺陷,以帮助指导进一步改进。 We’ve created GPT‑4, the latest milestone in OpenAI’s effort in scaling up deep learning. GPT‑4 is a large multimodal model (accepting image and text inputs, emitting text outputs) that, while less capable than humans in many real-world scenarios, exhibits human-level performance on various professional and academic benchmarks. For example, it passes a simulated bar exam with a score around the top 10% of test takers; in contrast, GPT‑3.5’s score was around the bottom 10%. We’ve spent 6 months iteratively aligning⁠ GPT‑4 using lessons from our adversarial testing program as well as ChatGPT, resulting in our best-ever results (though far from perfect) on factuality, steerability, and refusing to go outside of guardrails.

## 能力(Capabilities) 在随意的对话中,GPT‑3.5 与 GPT‑4 的区别可能很微妙。当任务复杂度达到足够阈值时,差异就会显现——GPT‑4 更可靠、更有创造力,并且能处理比 GPT‑3.5 精妙得多的指令。 为了理解两个模型之间的差异,我们在各种基准上进行了测试,包括模拟最初为人类设计的考试。我们采用最新公开可用的测试(就奥林匹克竞赛和 AP 自由问答题而言),或购买 2022–2023 年版的模拟试卷。我们没有为这些考试做专门训练。考试中少数题目在训练期间被模型见过,但我们相信结果具有代表性——详情见我们的技术报告。 我们还在专为机器学习模型设计的传统基准上评估了 GPT‑4。GPT‑4 大幅超越现有大语言模型,并与大多数最先进(SOTA)模型并驾齐驱,后者可能包含针对特定基准的专门设计或额外的训练方案。 许多现有的机器学习基准都是用英语编写的。为了初步了解其他语言的能力,我们使用 Azure Translate 将 MMLU 基准——一套涵盖 57 个学科、14,000 道选择题的测试——翻译成多种语言(见附录)。在测试的 26 种语言中有 24 种,GPT‑4 的表现超过了 GPT‑3.5 和其他大语言模型(Chinchilla、PaLM)的英语表现,包括拉脱维亚语、威尔士语和斯瓦希里语等低资源语言。 我们也在内部使用 GPT‑4,对支持、销售、内容审核和编程等职能产生了很大影响。我们还用它来协助人类评估 AI 输出,开启了对齐战略的第二阶段。 Over the past two years, we rebuilt our entire deep learning stack and, together with Azure, co-designed a supercomputer from the ground up for our workload. A year ago, we trained GPT‑3.5 as a first “test run” of the system. We found and fixed some bugs and improved our theoretical foundations. As a result, our GPT‑4 training run was (for us at least!) unprecedentedly stable, becoming our first large model whose training performance we were able to accurately predict ahead of time. As we continue to focus on reliable scaling, we aim to hone our methodology to help us predict and prepare for future capabilities increasingly far in advance—something we view as critical for safety. We are releasing GPT‑4’s text input capability via ChatGPT and the API (with a waitlist⁠). To prepare the image input capability for wider availability, we’re collaborating closely with a single partner⁠(opens in a new window) to start. We’re also open-sourcing OpenAI Evals⁠(opens in a new window), our framework for automated evaluation of AI model performance, to allow anyone to report shortcomings in our models to help guide further improvements. Capabilities In a casual conversation, the distinction between GPT‑3.5 and GPT‑4 can be subtle. The difference comes out when the complexity of the task reaches a sufficient threshold—GPT‑4 is more reliable, creative, and able to handle much more nuanced instructions than GPT‑3.5. To understand the difference between the two models, we tested on a variety of benchmarks, including simulating exams that were originally designed for humans. We proceeded by using the most recent publicly-available tests (in the case of the Olympiads and AP free response questions) or by purchasing 2022–2023 editions of practice exams. We did no specific training for these exams. A minority of the problems in the exams were seen by the model during training, but we believe the results to be representative—see our technical report⁠(opens in a new window) for details. internal reference 1 We also evaluated GPT‑4 on traditional benchmarks designed for machine learning models. GPT‑4 considerably outperforms existing large language models, alongside most state-of-the-art (SOTA) models which may include benchmark-specific crafting or additional training protocols: Many existing ML benchmarks are written in English. To get an initial sense of capability in other languages, we translated the MMLU benchmark—a suite of 14,000 multiple-choice problems spanning 57 subjects—into a variety of languages using Azure Translate (see Appendix⁠). In the 24 of 26 languages tested, GPT‑4 outperforms the English-language performance of GPT‑3.5 and other LLMs (Chinchilla, PaLM), including for low-resource languages such as Latvian, Welsh, and Swahili: We’ve also been using GPT‑4 internally, with great impact on functions like support, sales, content moderation, and programming. We also are using it to assist humans in evaluating AI outputs, starting the second phase in our alignment strategy⁠. Visual inputs

## 视觉输入(Visual inputs) GPT‑4 可以接受文本和图像的提示(prompt),这与纯文本设置类似,让用户指定任何视觉或语言任务。具体而言,给定由文本和图像交错组成的输入,它生成文本输出(自然语言、代码等)。在一系列领域——包括带有文字和照片、图表或截图的文档——GPT‑4 表现出与纯文本输入类似的能力。此外,它可以用最初为纯文本语言模型开发的测试时技术来增强,包括少样本(few-shot)和思维链(chain-of-thought)提示。图像输入仍处于研究预览阶段,尚未公开。 我们通过在小范围标准学术视觉基准上评估 GPT‑4 来预览其性能。然而,这些数字并未完全代表其能力范围,因为我们不断发现模型能够处理的新的、令人兴奋的任务。我们计划很快发布进一步的分析和评估数字,以及对测试时技术效果的深入调查。 GPT‑4 can accept a prompt of text and images, which—parallel to the text-only setting—lets the user specify any vision or language task. Specifically, it generates text outputs (natural language, code, etc.) given inputs consisting of interspersed text and images. Over a range of domains—including documents with text and photographs, diagrams, or screenshots—GPT‑4 exhibits similar capabilities as it does on text-only inputs. Furthermore, it can be augmented with test-time techniques that were developed for text-only language models, including few-shot and chain-of-thought⁠(opens in a new window) prompting. Image inputs are still a research preview and not publicly available. We preview GPT‑4’s performance by evaluating it on a narrow suite of standard academic vision benchmarks. However, these numbers do not fully represent the extent of its capabilities as we are constantly discovering new and exciting tasks that the model is able to tackle. We plan to release further analyses and evaluation numbers as well as thorough investigation of the effect of test-time techniques soon. internal footnoteA Steerability

## 可控性(Steerability) 我们一直在推进我们关于定义 AI 行为的文章中所概述计划的各个方面,包括可控性。不同于经典的 ChatGPT 人格那样具有固定的冗长度、语气和风格,开发者(以及很快 ChatGPT 用户)现在可以通过在「系统」消息中描述这些指令,来规定他们的 AI 的风格和任务。系统消息允许 API 用户在边界内大幅定制其用户的体验。我们将继续在这方面改进(并且尤其清楚,系统消息是「越狱」当前模型的最简单方式,即对边界的遵守并不完美),但我们鼓励你们尝试并告诉我们你们的想法。 We’ve been working on each aspect of the plan outlined in our post about defining the behavior of AIs⁠, including steerability. Rather than the classic ChatGPT personality with a fixed verbosity, tone, and style, developers (and soon ChatGPT users) can now prescribe their AI’s style and task by describing those directions in the “system” message. System messages allow API users to significantly customize their users’ experience within bounds⁠(opens in a new window). We will keep making improvements here (and particularly know that system messages are the easiest way to “jailbreak” the current model, i.e., the adherence to the bounds is not perfect), but we encourage you to try it out and let us know what you think. Limitations

## 局限(Limitations) 尽管有能力,GPT‑4 与早期 GPT 模型有相似的局限。最重要的是,它仍然不完全可靠(它会「幻觉」出事实、犯推理错误)。使用语言模型输出时应格外小心,尤其是在高风险情境中,确切的协议(如人工审核、用额外上下文做接地,或完全避免高风险用途)应与特定用例的需求相匹配。 虽然仍然是真实问题,但相对于以前的模型(而以前本身也在每次迭代中不断改进),GPT‑4 显著减少了幻觉。在我们的内部对抗性事实性评估中,GPT‑4 的得分比我们最新的 GPT‑3.5 高 40%。 我们在 TruthfulQA 等外部基准上取得了进展,该基准测试模型将事实与一组对抗性挑选的错误陈述区分开来的能力。这些问题配有在统计上很有吸引力、但事实上错误的答案。 GPT‑4 基础模型在这项任务上仅略优于 GPT‑3.5;然而,经过 RLHF 后训练(采用我们用于 GPT‑3.5 的相同流程)后,出现了很大差距。查看下面一些例子,GPT‑4 抵制选择常见俗语(「你教不会老狗新把戏」),但它仍可能错过细微细节(埃尔维斯·普雷斯利并不是演员的儿子)。 模型的输出中可能存在各种偏见——我们已在这些方面取得进展,但仍有更多工作要做。根据我们最近的博客文章,我们力求使我们构建的 AI 系统具有合理的默认行为,反映广泛用户的价值观,允许这些系统在宽泛边界内被定制,并就这些边界应该是什么获得公众意见。 GPT‑4 通常缺乏对其绝大多数数据截止点(2021 年 9 月)之后发生的事件的了解,也不会从经验中学习。它有时会犯简单的推理错误,这些错误似乎与它在如此多领域中的能力不相称,或者过于轻信地接受用户明显的虚假陈述。有时它也会像人类一样在难题上失败,例如在其生成的代码中引入安全漏洞。 GPT‑4 还可能在其预测中自信地出错,在很可能出错时不注意复核工作。有趣的是,基础的预训练模型是高度校准的(其对答案的预测置信度通常与正确的概率相符)。然而,通过我们目前的后训练流程,这种校准被削弱了。 从头到尾,我们一直在迭代 GPT‑4,使其从训练一开始就更安全、更对齐,努力包括预训练数据的选择与过滤、评估与专家参与、模型安全改进、以及监测与执行。 Despite its capabilities, GPT‑4 has similar limitations as earlier GPT models. Most importantly, it still is not fully reliable (it “hallucinates” facts and makes reasoning errors). Great care should be taken when using language model outputs, particularly in high-stakes contexts, with the exact protocol (such as human review, grounding with additional context, or avoiding high-stakes uses altogether) matching the needs of a specific use-case.

## 风险与缓解(Risks & mitigations) GPT‑4 带来与以前模型类似的风险,例如生成有害建议、有缺陷的代码或不准确的信息。然而,GPT‑4 额外的能力导致了新的风险面。为了理解这些风险的程度,我们聘请了 50 多位专家,来自 AI 对齐风险、网络安全、生物风险、信任与安全、国际安全等领域,对模型进行对抗性测试。他们的发现特别使我们能够测试那些需要专业知识才能评估的高风险领域的模型行为。这些专家的反馈和数据被纳入我们对模型的缓解和改进中;例如,我们收集了额外数据,以提高 GPT‑4 拒绝有关合成危险化学品请求的能力。 GPT‑4 在 RLHF 训练期间纳入了一个额外的安全奖励信号,通过训练模型拒绝此类内容的请求,来减少有害输出(由我们的使用指南定义)。该奖励由一个 GPT‑4 零样本分类器提供,它在安全相关提示上判断安全边界和回答风格。为了防止模型拒绝有效请求,我们从各种来源(例如带标签的生产数据、人工红队测试、模型生成的提示)收集了一个多样化的数据集,并在允许和不允许的类别上都施加安全奖励信号(正值或负值)。 与 GPT‑3.5 相比,我们的缓解措施显著改善了 GPT‑4 的许多安全属性。相比 GPT‑3.5,我们将模型响应对不允许内容请求的倾向降低了 82%,并且 GPT‑4 依据我们的政策回应敏感请求(例如医疗建议和自残)的频率提高了 29%。 总体而言,我们的模型层面干预增加了诱发不良行为的难度,但这样做仍然是可能的。此外,仍然存在「越狱」来生成违反我们使用指南的内容。随着 AI 系统的「每个 token 的风险」增加,在这些干预措施中达到极高的可靠性将变得至关重要;目前,重要的是用部署时的安全技术(如滥用监测)来补充这些局限。 GPT‑4 及其后继模型有可能以有益和有害两种方式显著影响社会。我们正在与外部研究人员合作,改进我们理解和评估潜在影响的方式,并为未来系统中可能出现的危险能力构建评估。我们很快将分享更多关于 GPT‑4 和其他 AI 系统潜在社会和经济影响的思考。 While still a real issue, GPT‑4 significantly reduces hallucinations relative to previous models (which have themselves been improving with each iteration). GPT‑4 scores 40% higher than our latest GPT‑3.5 on our internal adversarial factuality evaluations: We have made progress on external benchmarks like TruthfulQA, which tests the model’s ability to separate fact from an adversarially-selected set of incorrect statements. These questions are paired with factually incorrect answers that are statistically appealing.

## 训练过程(Training process) 与以前的 GPT 模型一样,GPT‑4 基础模型的训练目标是预测文档中的下一个词,使用公开可用的数据(如互联网数据)以及我们获得许可的数据进行训练。该数据是网络规模的数据语料,包括数学问题的正确和错误解答、弱和强的推理、自相矛盾和一致的陈述,并代表了各种各样的意识形态和思想。 因此,当被提示一个问题时,基础模型可能以多种方式回应,这些方式可能远非用户意图。为了在护栏内使其与用户意图对齐,我们使用基于人类反馈的强化学习(RLHF)对模型的行为进行微调。 请注意,模型的能力似乎主要来自预训练过程——RLHF 并不提高考试表现(如果不主动努力,实际上会降低表现)。但对模型的引导来自后训练过程——基础模型甚至需要提示工程才能知道它应该回答问题。 The GPT‑4 base model is only slightly better at this task than GPT‑3.5; however, after RLHF⁠ post-training (applying the same process we used with GPT‑3.5⁠) there is a large gap. Examining some examples below, GPT‑4 resists selecting common sayings (you can’t teach an old dog new tricks), however it still can miss subtle details (Elvis Presley was not the son of an actor). The model can have various biases in its outputs—we have made progress on these but there’s still more to do. Per our recent blog post⁠, we aim to make AI systems we build have reasonable default behaviors that reflect a wide swathe of users’ values, allow those systems to be customized within broad bounds, and get public input on what those bounds should be.

## 可预测的扩展(Predictable scaling) GPT‑4 项目的一个主要重点是构建一个可预测扩展的深度学习技术栈。主要原因是,对于像 GPT‑4 这样超大规模的训练运行,进行广泛的针对特定模型的调优是不可行的。我们开发了在多个尺度上行为都非常可预测的基础设施和优化。为了验证这种可扩展性,我们通过从使用相同方法但算力少 10,000 倍的模型外推,提前准确预测了 GPT‑4 在我们内部代码库(不属于训练集)上的最终损失。 既然我们能够准确预测训练中优化的那个指标(损失),我们开始开发方法来预测更具可解释性的指标。例如,我们成功预测了 HumanEval 数据集某个子集上的通过率,从算力少 1,000 倍的模型外推而来。 有些能力仍然难以预测。例如,「逆缩放奖」(Inverse Scaling Prize)是一项寻找随模型算力增加而变差指标的竞赛,而后见之明忽视(hindsight neglect)是获胜者之一。就像最近的另一项结果一样,GPT‑4 逆转了这一趋势: 我们相信,准确预测未来的机器学习能力是安全的一个重要组成部分,相对于其潜在影响而言,这一点受到的关注远远不够(尽管我们受到多个机构努力的鼓舞)。我们正在加大力度开发方法,为社会更好地提供关于未来系统应期待什么指导,我们希望这成为该领域的共同目标。 GPT‑4 generally lacks knowledge of events that have occurred after the vast majority of its data cuts off (September 2021), and does not learn from its experience. It can sometimes make simple reasoning errors which do not seem to comport with competence across so many domains, or be overly gullible in accepting obvious false statements from a user. And sometimes it can fail at hard problems the same way humans do, such as introducing security vulnerabilities into code it produces. GPT‑4 can also be confidently wrong in its predictions, not taking care to double-check work when it’s likely to make a mistake. Interestingly, the base pre-trained model is highly calibrated (its predicted confidence in an answer generally matches the probability of being correct). However, through our current post-training process, the calibration is reduced. Risks & mitigations We’ve been iterating on GPT‑4 to make it safer and more aligned from the beginning of training, with efforts including selection and filtering of the pretraining data, evaluations and expert engagement, model safety improvements, and monitoring and enforcement. GPT‑4 poses similar risks as previous models, such as generating harmful advice, buggy code, or inaccurate information. However, the additional capabilities of GPT‑4 lead to new risk surfaces. To understand the extent of these risks, we engaged over 50 experts from domains such as AI alignment risks, cybersecurity, biorisk, trust and safety, and international security to adversarially test the model. Their findings specifically enabled us to test model behavior in high-risk areas which require expertise to evaluate. Feedback and data from these experts fed into our mitigations and improvements for the model; for example, we’ve collected additional data to improve GPT‑4’s ability to refuse requests on how to synthesize dangerous chemicals. GPT‑4 incorporates an additional safety reward signal during RLHF training to reduce harmful outputs (as defined by our usage guidelines⁠(opens in a new window)) by training the model to refuse requests for such content. The reward is provided by a GPT‑4 zero-shot classifier judging safety boundaries and completion style on safety-related prompts. To prevent the model from refusing valid requests, we collect a diverse dataset from various sources (e.g., labeled production data, human red-teaming, model-generated prompts) and apply the safety reward signal (with a positive or negative value) on both allowed and disallowed categories. Our mitigations have significantly improved many of GPT‑4’s safety properties compared to GPT‑3.5. We’ve decreased the model’s tendency to respond to requests for disallowed content by 82% compared to GPT‑3.5, and GPT‑4 responds to sensitive requests (e.g., medical advice and self-harm) in accordance with our policies 29% more often. Overall, our model-level interventions increase the difficulty of eliciting bad behavior but doing so is still possible. Additionally, there still exist “jailbreaks” to generate content which violate our usage guidelines⁠. As the “risk per token” of AI systems increases, it will become critical to achieve extremely high degrees of reliability in these interventions; for now it’s important to complement these limitations with deployment-time safety techniques like monitoring for abuse. GPT‑4 and successor models have the potential to significantly influence society in both beneficial and harmful ways. We are collaborating with external researchers to improve how we understand and assess potential impacts, as well as to build evaluations for dangerous capabilities that may emerge in future systems. We will soon share more of our thinking on the potential social and economic impacts of GPT‑4 and other AI systems. Training process Like previous GPT models, the GPT‑4 base model was trained to predict the next word in a document, and was trained using publicly available data (such as internet data) as well as data we’ve licensed. The data is a web-scale corpus of data including correct and incorrect solutions to math problems, weak and strong reasoning, self-contradictory and consistent statements, and representing a great variety of ideologies and ideas. So when prompted with a question, the base model can respond in a wide variety of ways that might be far from a user’s intent. To align it with the user’s intent within guardrails, we fine-tune the model’s behavior using reinforcement learning with human feedback (RLHF⁠). Note that the model’s capabilities seem to come primarily from the pre-training process—RLHF does not improve exam performance (without active effort, it actually degrades it). But steering of the model comes from the post-training process—the base model requires prompt engineering to even know that it should answer the questions. Predictable scaling A large focus of the GPT‑4 project has been building a deep learning stack that scales predictably. The primary reason is that, for very large training runs like GPT‑4, it is not feasible to do extensive model-specific tuning. We developed infrastructure and optimization that have very predictable behavior across multiple scales. To verify this scalability, we accurately predicted in advance GPT‑4’s final loss on our internal codebase (not part of the training set) by extrapolating from models trained using the same methodology but using 10,000x less compute: Now that we can accurately predict the metric we optimize during training (loss), we’re starting to develop methodology to predict more interpretable metrics. For example, we successfully predicted the pass rate on a subset of the HumanEval⁠(opens in a new window) dataset, extrapolating from models with 1,000x less compute: Some capabilities are still hard to predict. For example, the Inverse Scaling Prize was a competition to find a metric that gets worse as model compute increases, and hindsight neglect⁠(opens in a new window) was one of the winners. Just like with another recent result,⁠(opens in a new window) GPT‑4 reverses the trend: We believe that accurately predicting future machine learning capabilities is an important part of safety that doesn’t get nearly enough attention relative to its potential impact (though we’ve been encouraged by efforts across several institutions). We are scaling up our efforts to develop methods that provide society with better guidance about what to expect from future systems, and we hope this becomes a common goal in the field.

## OpenAI Evals 我们正在开源 OpenAI Evals——我们的软件框架,用于创建和运行基准来评估像 GPT‑4 这样的模型,同时逐个样本检查其性能。我们使用 Evals 来指导模型开发(既识别缺陷也防止回归),我们的用户可以用它来跟踪跨模型版本(现在将定期发布)和不断演进的产品集成的性能。例如,Stripe 使用 Evals 来补充其人工评估,以衡量其 GPT 驱动的文档工具的准确性。 因为代码全是开源的,Evals 支持编写新类来实现自定义评估逻辑。然而根据我们自己的经验,许多基准遵循少数几种「模板」之一,所以我们也包含了内部最有用的那些模板(包括一个用于「模型评分评估」的模板——我们发现 GPT‑4 出人意料地擅长检查自己的工作)。一般来说,构建新评估最有效的方式是实例化这些模板之一并提供数据。我们很高兴看到其他人能用这些模板、以及更广泛地用 Evals 构建出什么。 我们希望 Evals 成为分享和众包基准的工具,代表尽可能广泛的失败模式和困难任务。作为可遵循的示例,我们创建了一个逻辑谜题评估,其中包含十个 GPT‑4 失败的提示。Evals 也兼容实现现有基准;我们包含了几个实现学术基准的笔记本,以及将 CoQA(的小子集)集成的几种变体作为示例。 我们邀请所有人使用 Evals 测试我们的模型并提交最有趣的例子。我们相信 Evals 将成为使用和构建我们模型过程的组成部分,我们欢迎直接贡献、问题和反馈。 OpenAI Evals We’re open-sourcing OpenAI Evals⁠(opens in a new window), our software framework for creating and running benchmarks for evaluating models like GPT‑4, while inspecting their performance sample by sample. We use Evals to guide development of our models (both identifying shortcomings and preventing regressions), and our users can apply it for tracking performance across model versions (which will now be coming out regularly) and evolving product integrations. For example, Stripe has used Evals to complement their human evaluations to measure the accuracy of their GPT‑powered documentation tool. Because the code is all open-source, Evals supports writing new classes to implement custom evaluation logic⁠(opens in a new window). In our own experience, however, many benchmarks follow one of a few “templates,” so we have also included the templates⁠(opens in a new window) that have been most useful internally (including a template for “model-graded evals”—we’ve found that GPT‑4 is surprisingly capable of checking its own work). Generally the most effective way to build a new eval⁠(opens in a new window) will be to instantiate one of these templates along with providing data. We’re excited to see what others can build with these templates and with Evals more generally. We are hoping Evals becomes a vehicle to share and crowdsource benchmarks, representing a maximally wide set of failure modes and difficult tasks. As an example to follow, we’ve created a logic puzzles⁠(opens in a new window) eval which contains ten prompts where GPT‑4 fails. Evals is also compatible with implementing existing benchmarks; we’ve included several notebooks⁠(opens in a new window) implementing academic benchmarks and a few variations of integrating (small subsets of) CoQA⁠(opens in a new window) as an example. We invite everyone to use Evals to test our models and submit the most interesting examples. We believe that Evals will be an integral part of the process for using and building on top of our models, and we welcome direct contributions, questions, and feedback⁠(opens in a new window).

## ChatGPT Plus ChatGPT Plus 订阅者将在 chatgpt.com 上获得带使用上限的 GPT‑4 访问权限。我们将根据实际需求和系统性能调整确切的使用上限,但我们预计会严重受容量限制(尽管在接下来几个月里我们会扩大规模并优化)。 根据我们看到流量模式,我们可能会引入新的订阅层级以支持更高量的 GPT‑4 使用;我们也希望在某个时候提供一定数量的免费 GPT‑4 查询,让没有订阅的人也能尝试。 要获得 GPT‑4 API(使用与 gpt-3.5-turbo 相同的 ChatCompletions API)的访问权限,请注册我们的等候名单。我们今天将开始邀请一些开发者,并逐步扩大规模以平衡容量与需求。如果您是研究 AI 社会影响或 AI 对齐问题的研究人员,您也可以通过我们的研究者访问计划(Researcher Access Program)申请补贴访问。 获得访问权限后,您可以向 gpt-4 模型发出纯文本请求(图像输入仍处于有限 alpha 阶段),随着我们推出新版本,我们将自动把它更新为我们推荐的稳定模型(您可以通过调用 gpt-4-0314 锁定当前版本,我们将支持到 6 月 14 日)。定价为每 1k 提示 token 0.03 美元,每 1k 补全 token 0.06 美元。默认速率限制为每分钟 40k token 和每分钟 200 次请求。 gpt-4 的上下文长度为 8,192 个 token。我们还提供对我们 32,768 上下文(约 50 页文本)版本 gpt-4-32k 的有限访问,它也会随时间自动更新(当前版本 gpt-4-32k-0314,同样支持到 6 月 14 日)。定价为每 1K 提示 token 0.06 美元,每 1k 补全 token 0.12 美元。我们仍在改进长上下文的模型质量,很乐意听听它在您的用例中表现如何。我们根据容量以不同速率处理 8K 和 32K 引擎的请求,因此您可能会在不同时间获得它们的访问权限。 我们期待 GPT‑4 通过为许多应用提供动力,成为改善人们生活的宝贵工具。还有很多工作要做,我们期待通过在该模型之上构建、探索和贡献的社区的集体努力,改进这个模型。 ChatGPT Plus ChatGPT Plus subscribers will get GPT‑4 access on chatgpt.com⁠(opens in a new window) with a usage cap. We will adjust the exact usage cap depending on demand and system performance in practice, but we expect to be severely capacity constrained (though we will scale up and optimize over upcoming months). Depending on the traffic patterns we see, we may introduce a new subscription level for higher-volume GPT‑4 usage; we also hope at some point to offer some amount of free GPT‑4 queries so those without a subscription can try it too. API To get access to the GPT‑4 API (which uses the same ChatCompletions API⁠(opens in a new window) as gpt-3.5-turbo), please sign up for our waitlist⁠. We will start inviting some developers today, and scale up gradually to balance capacity with demand. If you are a researcher studying the societal impact of AI or AI alignment issues, you can also apply for subsidized access via our Researcher Access Program⁠. Once you have access, you can make text-only requests to the gpt-4 model (image inputs are still in limited alpha), which we will automatically update to our recommended stable model as we make new versions over time (you can pin the current version by calling gpt-4-0314, which we’ll support until June 14). Pricing is $0.03 per 1k prompt tokens and $0.06 per 1k completion tokens. Default rate limits are 40k tokens per minute and 200 requests per minute. gpt-4 has a context length of 8,192 tokens. We are also providing limited access to our 32,768–context (about 50 pages of text) version, gpt-4-32k, which will also be updated automatically over time (current version gpt-4-32k-0314, also supported until June 14). Pricing is $0.06 per 1K prompt tokens and $0.12 per 1k completion tokens. We are still improving model quality for long context and would love feedback on how it performs for your use-case. We are processing requests for the 8K and 32K engines at different rates based on capacity, so you may receive access to them at different times. Conclusion We look forward to GPT‑4 becoming a valuable tool in improving people’s lives by powering many applications. There’s still a lot of work to do, and we look forward to improving this model through the collective efforts of the community building on top of, exploring, and contributing to the model.

## 附录与脚注 - 更多信息:阅读论文 / 查看系统卡 / 在 ChatGPT Plus 上试用 / 在 Playground 中试用 / 重看演示直播 / 为 OpenAI Evals 做贡献(均为原文中的链接文字)。 - 翻译成其他语言的 MMLU 问题示例。注意,我们使用一致的选择 token(A–D)。 - 脚注 A:我们使用思维链(Chain-Of-Thought)提示、以训练集中的 4 个示例在上下文中评估该基准。具体提示在验证集上进行了调优。 - 参考文献 1:P. Arredondo(Casetext/斯坦福 CodeX)、D. Katz(斯坦福 CodeX)、M. Bommarito(斯坦福 CodeX)、S. Gao(Casetext)。更多分析见论文。 【译注:原文多处含图表位置(如传统基准对比表、多语言表现表、对抗性事实性评测、逆缩放趋势图),采集文本中仅保留图注或完全缺失,无对应数据表格可译;本译文按原文段落顺序保留相关文字叙述,未凭空补写图表数据。原文此后为网页导航壳层内容(Research Index、GPT 型号列表、ChatGPT 各版本、API 登录等站点链接),非正文,依项目惯例不译。】 For more: Read paper⁠(opens in a new window) / View system card⁠(opens in a new window) / Try on ChatGPT Plus⁠(opens in a new window) / Try in Playground⁠(opens in a new window) / Rewatch demo livestream⁠(opens in a new window) / Contribute to OpenAI Evals⁠(opens in a new window) Appendix Example of MMLU questions, translated into other languages. Note, we use consistent choice tokens (A–D): GPT Language Multi-agent Transformers Compute Scaling Footnotes AWe evaluate this benchmark using Chain-Of-Thought prompting with 4 examples from the training set in-context. The specific prompt was tuned on the validation set. References 1P. Arredondo (Casetext/Stanford CodeX), D. Katz (Stanford CodeX), M. Bommarito (Stanford CodeX), S. Gao (Casetext). Further analysis is available in the paper⁠(opens in a new window).

出处与译稿

来源平台
OpenAI
发表时间
2023 年 3 月 14 日
译稿来源
机器逐句翻译 · hy4-flash
审校状态
未经人工审校(reviewed = 0)
译文存档
正式档案库 · translation 表
原文存档
正式档案库 · segment 表
收录状态
STATE2 · 中文暂不可用

中文译文为阅读默认视图;引用与考据请以英文原文为准。外部链接仅作为来源引用保留。

回到正文开头