| 送交者: xiaochuliu[♂★★大🇨🇳lngviva★★♂] 于 2023-03-18 12:00 已读 4102 次 | xiaochuliu的个人频道 |
GPT-4论文竟有隐藏线索：GPT-5或完成训练、OpenAI两年内接近AGI 6park.com
6park.comGPT-4，火爆，非常火爆。 6park.com 6park.com在铺天盖地的叫好声中，有件事可能你是“万万没想到”—— 6park.com在OpenAI公布的技术论文里，竟然藏着九大隐秘的线索！ 6park.com 6park.com
6park.com这些线索是由国外博主AI Explained发现并整理。 6park.com 6park.com他宛如一位细节狂魔，从长达98页论文中，逐个揭秘这些“隐匿的角落”，包括： 6park.comGPT-5可能已经完成训练 6park.comGPT-4出现过自己“挂掉”的情况 6park.comOpenAI两年内或实现接近AGI 6park.com…… 6park.com
6park.com发现一：GPT4出现过自己“挂掉”的情况 6park.com在GPT-4技术论文的第53页处，OpenAI提到了这样一个机构——Alignment Research Center（ARC）。 6park.com这家机构主要做的事情，就是专门研究AI如何对齐（alignment）人类利益。 6park.com而OpenAI在开发GPT-4的早期阶段，便给ARC开了抢先体验的后门，希望他们能够评估GPT-4的两项能力： 6park.com模型自主复制能力 6park.com模型获取资源能力 6park.com
6park.com虽然OpenAI在论文中强调了“ARC没法微调早期版本的GPT-4”、“他们无权访问GPT-4的最终版本”；也强调了测试结果显示GPT-4在上述两个能力的效率不高（降低AI伦理隐患）。 6park.com但眼尖的博主揪出来的是接下来的一句话： 6park.com(found it ineffective at) avoiding being shut down “in the wild”. 6park.com在自然环境中，GPT-4会避免自己“挂掉”。 6park.com博主的意思是，既然OpenAI选择让ARC去测试评估GPT-4会不会主动避免自己被“挂掉”，说明此前必定出现过这种情况。 6park.com那么延伸出来的隐患就是，如果ARC在测试过程中其实是失败的怎么办；或者未来真出现了自己“挂掉”的情况，又将怎么处理。 6park.com基于此，博主便有了第二个发现： 6park.com发现二：主动要求自我监管，很罕见 6park.com在第2页的脚注中，OpenAI注释了这么一句话： 6park.comOpenAI will soon publish additional thoughts on the social and economic implications of AI systems, including the need for effective regulation. 6park.comOpenAI将很快发布关于AI系统的社会和经济影响的其它思考，包括有效监管的必要性。 6park.com
6park.com博主认为，一个行业主动要求监管自己，这是个非常罕见的现象。 6park.com事实上，OpenAI老板Sam Altman此前发表的言论比这还要直白。 6park.com当时Altman发表了关于SVB倒闭的推文，他认为“我们需要对银行做更多的监管”；有人就评论回怼了：“他从来不会说‘我们需要对AI做更多的监管’”。 6park.com结果Altman直截了当的回复说： 6park.com绝对需要。 6park.com
6park.com博主认为，这个AI行业正在呼吁进行监管，至于监管后的结果如何，是值得拭目以待的。 6park.com发现三：与微软高层想法背道而驰 6park.com接下来的发现，是根据论文第57页中的这句话： 6park.comOne concern of particular importance to OpenAI is the risk of racing dynamics leading to a decline in safety standards, the diffusion of bad norms, and accelerated AI timelines, each of which heighten societal risks associated with AI. 6park.com对OpenAI来说，（科技的）竞赛会导致安全标准的下降、不良规范的扩散、AI发展进程的加速，它们都加剧了与人工智能相关的社会风险。 6park.com
6park.com但很奇怪的一点是，OpenAI所提到的担忧，尤其是“AI发展进程的加速”，似乎是与微软高层的想法背道而驰。 6park.com因为在此前的爆料中称，微软CEO和CTO的压力很大，他们希望OpenAI的模型能尽快让用户用起来。 6park.com
6park.com有些人在看到这则消息时是比较兴奋，但同样也有一波人发出了跟OpenAI一样的担忧。 6park.com博主认为，不论如何，可以肯定的一点是OpenAI和微软在这件事的想法是相悖的。 6park.com发现四：OpenAI会协助超越它的公司 6park.com第四个发现的线索，是来自与“发现三”同一页的脚注： 6park.com
6park.com这段脚注展示了OpenAI一个非常大胆的承诺： 6park.com如果另一家公司在我们之前实现了接近AGI（通用人工智能），那我们承诺不会跟它做竞争，相反，会协助完成那个项目。 6park.com但这种情况发生的条件，可能是另一家公司需得在未来两年内，成功接近AGI的机会在一半或以上 6park.com而这里提到的AGI，OpenAI和Altam在官方博客中已经给出了定义—— 6park.com普遍比人类更聪明，并且有益于全人类的人工智能系统。 6park.com
6park.com因此，博主认为，这段脚注要么意味着OpenAI在未来两年内将实现AGI，要么意味着他们放弃了一切并与另一家公司展开了合作。 6park.com发现五：雇佣“超级预测员” 6park.com博主的下一个发现，是来自论文第57中的一段话。 6park.com这段话大致的意思就是，OpenAI雇佣了预测专家，来预测当他们部署了GPT-4之后会带来怎样的风险。 6park.com
6park.com然后博主顺藤摸瓜，发现了这些所谓的“超级预测员”的庐山真面目。 6park.com
6park.com这些“超级预测员”的能力已经得到了广泛地认可，有报道称，他们的预测准确率，甚至比那些有独家信息、情报的分析师还要高出30%。 6park.com正如我们刚才提到的，OpenAI邀请这些“超级预测员”，来预测部署GPT-4后可能存在的风险，并采取相应措施来规避。 6park.com其中，“超级预测员”建议将GPT-4部署时间推迟6个月，也就是今年秋季左右；但很显然，OpenAI并没有采纳他们的建议。 6park.com博主对此认为，OpenAI这么做的原因，可能是来自微软的压力。 6park.com发现六：征服常识 6park.com在这篇论文中，OpenAI展示了众多基准测试的图表，大家在昨天铺天盖地的传播过程中应该也见到了。 6park.com但博主在这个发现中要强调的是位于第7页的一项基准测试，尤其是聚焦到了“HellaSwag”这一项。 6park.com
6park.comHellaSwag的内容主要是常识推理，这就和GPT-4发布时宣布的“已经达到了人类的常识水平”相匹配。 6park.com不过博主也承认，这一点并没有“通过律师考试”等能力那么吸引人，但这也算得上是人类科技发展的里程碑。 6park.com但常识是怎么测试的？我们又如何评判GPT-4已经达到了人类水平？ 6park.com为此，博主深入研究了与之相关的论文研究： 6park.com
6park.com博主在论文中找到了相关数据，在“人类”那一栏中，分数分布在了94-96.5之间。 6park.com而GPT-4的95.3，便正好在这个区间之间。 6park.com
6park.com发现七：GPT-5可能已经完成训练 6park.com第七个发现，同样是在论文中的第57页： 6park.com我们在发布GPT-4之前花费8个月时间进行安全研究、风险评估和迭代。 6park.com
6park.com也就是说，OpenAI在去年年底推出ChatGPT的时候，已经有了GPT-4。 6park.com于是乎，博主便预测GPT-5的训练时间不会很久，甚至他认为GPT-5已经可能训练完成。 6park.com但接下来的问题是漫长的安全研究和风险评估，可能是几个月，也可能是一年甚至更久。 6park.com发现八：试一把双刃剑 6park.com第8个发现，是来自论文的第56页。 6park.com这段话说的是： 6park.comGPT-4对经济和劳动力的影响，应成为政策制定者和其他利益相关者的关键考虑因素。 6park.com虽然现有的研究主要集中在人工智能和生成模型如何给人类加buff，但GPT-4或后续模型可能会导致某些工作的自动化。 6park.com
6park.comOpenAI这段话背后想传达的点比较明显，就是我们经常提到的“科技是把双刃剑”。 6park.com博主找了相当多的证据表明，像ChatGPT、GitHub Copilot这些AI工具，确确实实地提高了相关工作者的效率。 6park.com但他更关注的是论文中这段话的后半句，也就是OpenAI给出的“警告”——导致某些工作的自动化。 6park.com 6park.com博主对此比较认同，毕竟在GPT-4的能力可以在某些特定领域中以人类10倍甚至更高的效率来完成。 6park.com 6park.com放眼未来，这很可能会导致相关工作人员工资降低，或者需得借助这些AI工具完成以前工作量的数倍等一系列问题。 6park.comOpenAI让GPT-4学会拒绝的方法，叫做基于规则的奖励模型（RBRMs）。 6park.com
6park.com博主概括了这种方法的工作流程：给GPT-4一组要遵守的原则，如果模型遵守了这些原则，那么就会提供相应的奖励。 6park.com 6park.com他认为OpenAI正在用人工智能的力量，让AI模型朝着符合人类原则的方向发展。 6park.com 6park.com但目前OpenAI并没有对此做更加细致和深入的介绍。 6park.comWelcome to the GPT-4 Developer Demo Livestream. 6park.com欢迎来到 GPT-4 开发者演示直播。 6park.comHonestly, it's kind of hard for me to believe that this day is here. 6park.com老实说，我有点难以相信这一天会到来。 6park.comOpenAI has been building this technology really since we started the company, but for the past two years, we've been really focused on delivering GPT-4. That started with rebuilding our entire training stack, Actually training the model, and then seeing what it was capable of. 6park.comOpenAI 自我们成立公司以来就一直在构建这项技术，但在过去的两年里， 我们一直专注于交付 GPT-4。 首先是重建我们的整个训练堆栈，实际训练模型， 然后查看它的能力。 6park.comTrying to figure out its capabilities, its risks, working with partners in order to test it in real world scenarios. 6park.com试图找出它的能力、它的风险，与合作伙伴合作， 以便在现实世界的场景中对其进行测试。 6park.comReally tuning its behavior, optimizing the model, getting it available so that you can use it. 6park.com真正调整它的行为，优化模型，让它可用，这样你就可以使用它了。 6park.comAnd so today, our goal is to show you a little bit of how to make GPT-L4 shine. 6park.com所以今天，我们的目标是向您展示如何让 GPT-L4 大放异彩。 6park.comHow to really get the most out of it, where its weaknesses are, where we're still working on it, and Just how to really use it as a good tool, a good partner. 6park.com如何真正充分利用它，它的弱点在哪里， 我们仍在努力的地方，以及如何真正将它用作一个好工具，一个好伙伴。 6park.comSo if you're interested in participating in the stream, if you go to our Discord, so it's discord. 6park.com所以如果你有兴趣参与直播，如果你去我们的 Discord，那就是 discord。 6park.comgg slash OpenAI, there's comments in there and we'll take a couple of audience suggestions. 6park.comgg slash OpenAI，里面有评论，我们会听取一些听众的建议。 6park.comSo the first thing I want to show you is the first task that GPT-4 could do that we never really got 3.5 to do. 6park.com因此，我想向您展示的第一件事是 GPT-4 可以完成的第一项任务，而我们从未真正让 3.5 完成。 6park.comThe way to think about this is all throughout training, that you're constantly doing all this work. 6park.com考虑这一点的方法是贯穿整个训练过程，你一直在做所有这些工作。 6park.comIt's 2 AM, the pager goes off, You fix the model, and you're always wondering, is it going to work? 6park.com现在是凌晨 2 点，传呼机响了，你修好了模型，然后你一直在想，它能用吗？ 6park.comIs all of this effort actually going to pan out? 6park.com所有这些努力真的会成功吗？ 6park.comSo we all had a pet task that we really liked, and that we would all individually be trying to see is the model capable of it now? 6park.com所以我们都有一个我们真正喜欢的宠物任务，我们都会单独尝试看看这个模型现在有能力吗？ 6park.comI'm going to show you the first one that we had a success for 4, but never really got there for 3.5. So I'm just going to copy the top of our blog post from today, going to paste it into our playground. 6park.com我将向您展示第一个， 我们在 4 上取得了成功， 但在 3.5 上从未真正实现。 因此，我将从今天开始复制我们博客文章的顶部，并将其粘贴到我们的 playground 中。 6park.comNow, this is our new chat completions playground that came out two weeks ago. 6park.com现在，这是我们两周前推出的新聊天完成游乐场。 6park.comI'm going to show you first with GPT 3.5, 4 has the same API to it, the same playground. 6park.com我将首先向您展示 GPT 3.5，4 具有相同的 API，相同的游乐场。 6park.comThe way that it works is you have a system message where you explain to the model what it's supposed to do, and we've made these models very steerable. 6park.com它的工作方式是你有一个系统消息， 你可以在其中向模型解释它应该做什么，我们已经使这些模型非常易于操纵。 6park.comSo you can provide it with really any instruction you want, whatever you dream up. 6park.com所以你可以为它提供任何你想要的指令，无论你想什么。 6park.comThe model will adhere to it pretty well, and in the future, it will get increasingly, increasingly powerful at steering the model very reliably. 6park.com该模型将很好地遵守它，并且在未来，它将在非常可靠地操纵模型方面变得越来越强大。 6park.comYou can then paste whatever you want as a user, the model will return messages as an assistant. 6park.com然后您可以作为用户粘贴任何您想要的内容，该模型将作为助手返回消息。 6park.comThe way to think of it is that we're moving away from just raw text in, raw text out, where you can't tell where different parts of the conversation come from, but towards this much more structured format that gives the model the opportunity to know, well, this is the user asking me to do something that the developer didn't intend, I should listen to the developer here. 6park.com思考它的方式是， 我们正在远离原始文本输入，原始文本输出， 在这种情况下你无法分辨对话的不同部分来自哪里，而是转向这种结构化程度更高的格式， 它为模型提供了有机会知道，嗯，这是用户让我做一些开发者无意的事情，我应该在这里听开发者的。 6park.comSo now, time to actually show you the task that I'm referring to. 6park.com所以现在，是时候实际向您展示我所指的任务了。 6park.comSo everyone's familiar with, summarize. 6park.com所以大家熟悉了，总结一下。 6park.comThis is an article into a sentence, getting a little more specific, but where every word begins with G. So this is 3.5. Let's see what it does. 6park.com这是一篇文章变成一个句子， 更具体一点，但每个单词都以 G 开头。 所以这是 3.5。 让我们看看它做了什么。 6park.comYeah, it didn't even try, just gave up on the task. 6park.com是的，它甚至没有尝试，只是放弃了任务。 6park.comThis is pretty typical for 3.5, trying to do this particular task. 6park.com这对于 3.5 来说是非常典型的，试图完成这个特定的任务。 6park.comIf it's a very stilted article or something like that, maybe it can succeed, but for the most part, 3.5 just gives up. 6park.com如果是很矫情的文章之类的，说不定还能成功，但大部分3.5就直接放弃了。 6park.comBut let's try the exact same prompt, the exact same system message in GPT-4. So borderline, whether you want to count AI or not, but so let's say AI doesn't count. 6park.com但是让我们尝试完全相同的提示，GPT-4 中完全相同的系统消息。所以分界线，不管你想不想算人工智能，但假设人工智能不算数。 6park.comThat's cheating. 6park.com那是作弊。 6park.comSo fair enough. 6park.com太公平了。 6park.comThe model happily accepts my feedback. 6park.com该模型愉快地接受了我的反馈。 6park.comSo now to make sure it's not just good for Gs, I'd like to turn this over to the audience. 6park.com所以现在为了确保它不仅对 Gs 有好处，我想把它交给观众。 6park.comI'll take a suggestion on what letter to try next. 6park.com我会就接下来要尝试的字母提出建议。 6park.comIn the meanwhile, while I'm waiting for our moderators to pick the lucky letter, I will give a try with A. But in this case, I'll say GPT-4 is fine. 6park.com与此同时，在等待我们的版主挑选幸运字母的同时，我会尝试使用 A。 但在这种情况下，我会说 GPT-4 没问题。 6park.comWhy not? 6park.com为什么不？ 6park.comAlso, pretty good summary. 6park.com另外，总结得很好。 6park.comSo I'll hop over to our Discord. 6park.com所以我会跳到我们的 Discord。 6park.comAll right. 6park.com好的。 6park.comWow. 6park.com哇。 6park.comPeople are being a little ambitious here. 6park.com人们在这里有点雄心勃勃。 6park.comI'm really trying to put the model through the paces. 6park.com我真的在努力让模型走上正轨。 6park.comWe're going to try Q, which if you think about this for a moment, I want the audience to really think about how would you do a summary of this article that all starts with Q? 6park.com我们将尝试使用 Q， 如果您稍微考虑一下，我希望听众真正考虑如何对这篇以 Q 开头的文章进行总结？ 6park.comIt's not easy. 6park.com这并不容易。 6park.comIt's pretty good. 6park.com这个很不错。 6park.comThat's pretty good. 6park.com那很好。 6park.comAll right. 6park.com好的。 6park.comSo I've shown you summarizing an existing article. 6park.com所以我已经向您展示了对现有文章的总结。 6park.comI want to show you how you can flexibly combine ideas between different articles. 6park.com我想向您展示如何在不同文章之间灵活组合想法。 6park.comSo I'm going to take this article that was on Hacker News yesterday, copy-paste it into the same conversation. 6park.com所以我打算把昨天在 Hacker News 上发表的这篇文章复制粘贴到同一个对话中。 6park.comSo it has all the context of what we were just doing. 6park.com所以它具有我们刚刚所做的所有上下文。 6park.comI'm going to say, find one common theme between this article and the GPT-4 blog. 6park.com我要说的是，在本文和 GPT-4 博客之间找到一个共同主题。 6park.comSo this is an article about Pinecone, which is a Python web app development framework, and it's making the technology more accessible, user-friendly. 6park.com所以这是一篇关于 Pinecone 的文章，它是一个 Python 网络应用程序开发框架，它使该技术更易于访问，对用户更友好。 6park.comIf you don't think that was insightful enough, you can always give some feedback and say, that was not insightful enough. 6park.com如果您认为这不够有见地，您可以随时提供一些反馈并说，这还不够有见地。 6park.comPlease. 6park.com请。 6park.comNo, I'll just even just leave it there. 6park.com不，我什至会把它留在那里。 6park.comLeave it up to the model to decide. 6park.com留给模型来决定。 6park.comSo bridging the gap between powerful technology and practical applications seems not bad. 6park.com因此，弥合强大技术与实际应用之间的差距似乎还不错。 6park.comOf course, you can ask for any other kind of task you want using its flexible language, understanding, and synthesis. 6park.com当然，您可以使用其灵活的语言、理解和综合来要求您想要的任何其他类型的任务。 6park.comYou can ask for something like, now turn the GPT-4 blog post into a rhyming poem. 6park.com你可以要求类似的东西，现在把 GPT-4 博客文章变成一首押韵诗。 6park.comPicked up on OpenAI evals, open source for all, helping to guide answering the call. 6park.com接受 OpenAI 评估，为所有人开源，帮助指导接听电话。 6park.comWhich by the way, if you'd like to contribute to this model, please give us evals. 6park.com顺便说一下，如果您想为这个模型做出贡献，请给我们评估。 6park.comWe have an open source evaluation framework that will help us guide and all of our users understand what the model is capable of and to take it to the next level. 6park.com我们有一个开源评估框架，它将帮助我们指导和我们所有的用户了解模型的功能并将其提升到一个新的水平。 6park.comSo there we go. 6park.com所以我们开始了。 6park.comThis is consuming existing content using GPT-4 with a little bit of creativity on top. 6park.com这是使用 GPT-4 消耗现有内容，再加上一点点创造力。 6park.comBut next, I want to show you how to build with GPT-4, what it's like to create with it as a partner. 6park.com但接下来，我想向您展示如何使用 GPT-4 进行构建，以及作为合作伙伴使用它进行创建的感觉。 6park.comSo the thing we're going to do is we're going to actually build a Discord bot. 6park.com所以我们要做的是实际构建一个 Discord 机器人。 6park.comI'll build it live and show you the process, show you debugging, show you what the model can do, where its limitations are, and how to work with them in order to achieve new heights. 6park.com我将实时构建它并向您展示过程，向您展示调试， 向您展示模型可以做什么，它的局限性在哪里， 以及如何与它们一起工作以达到新的高度。 6park.comSo the first thing I'll do is tell the model that this time, it's supposed to be an AI programming assistant. 6park.com所以我要做的第一件事就是告诉模型，这次它应该是一个 AI 编程助手。 6park.comIts job is to write things out in pseudocode first and then actually write the code. 6park.com它的工作是先用伪代码写出内容，然后再实际编写代码。 6park.comThis approach is very helpful to let the model break down the problem into smaller pieces. 6park.com这种方法非常有助于让模型将问题分解成更小的部分。 6park.comThen that way, you're not asking it to just come up with a super hard solution to a problem all in one go. 6park.com这样一来，您就不会要求它一次性想出一个解决问题的超难解决方案。 6park.comIt also makes it very interpretable because you can see exactly what the model was thinking and you can even provide corrections if you'd like. 6park.com它还使它非常易于解释，因为您可以准确地看到模型在想什么，如果您愿意，您甚至可以提供更正。 6park.comSo here is the prompt that we're going to ask it. 6park.com所以这是我们要问的提示。 6park.comThis is the thing that 3.5 would totally choke on if you've tried anything like it. 6park.com如果你尝试过类似的东西，这就是 3.5 会完全窒息的事情。 6park.comBut so we're going to ask for a Discord bot that uses the GPT-4 API to read images and text. 6park.com但是，我们将要求使用 GPT-4 API 来读取图像和文本的 Discord 机器人。 6park.comNow, there's one problem here, which is this model's training cutoff is in 2021, which means it has not seen our new chat completions format. 6park.com现在，这里有一个问题，就是这个模型的训练截止日期是 2021 年，这意味着它还没有看到我们新的聊天完成格式。 6park.comSo I literally just went to the blog post from two weeks ago, copy-pasted from the blog post including the response format. 6park.com所以我真的只是去了两周前的博客文章，从博客文章中复制粘贴， 包括响应格式。 6park.comIt has not seen the new image extension to that, and so I just wrote that up in just very minimal detail about how to include images. 6park.com它没有看到新的图像扩展，所以我只是写了关于如何包含图像的非常简单的细节。 6park.comNow, the model can actually leverage that documentation that it did not have memorized, but it does not know. 6park.com现在， 该模型实际上可以利用它没有记住的文档，但它不知道。 6park.comIn general, these models are very good at using information that it's been trained on in new ways and synthesizing new content. 6park.com一般来说，这些模型非常擅长使用以新方式训练过的信息并合成新内容。 6park.comYou can see that right here that it actually wrote an entirely new bot. 6park.com您可以在这里看到它实际上编写了一个全新的机器人。 6park.comNow, let's actually see if this bot is going to work in practice. 6park.com现在，让我们实际看看这个机器人是否会在实践中发挥作用。 6park.comSo you should always look through the code to get a sense of what it does. 6park.com因此，您应该始终通读代码以了解它的作用。 6park.comDon't run untrusted code from humans or from AIs. 6park.com不要运行来自人类或 AI 的不受信任的代码。 6park.comOne thing to note is that the Discord API has changed a lot over time, and particularly that there's one feature that has changed a lot since this model was trained. 6park.com需要注意的一件事是， 随着时间的推移， Discord API 发生了很大变化，特别是自训练该模型以来， 有一项功能发生了很大变化。 6park.comGive it a try. 6park.com试一试。 6park.comIn fact, yes, we are missing the intense keyword. 6park.com事实上，是的，我们缺少 intense 关键字。 6park.comThis is something that came out in 2020. So the model does know it exists, but it doesn't know which version of the Discord API we're using. 6park.com这是 2020 年出现的东西。 所以模型确实知道它存在，但它不知道我们使用的是哪个版本的 Discord API。 6park.comSo are we out of luck? 6park.com那么我们运气不好吗？ 6park.comWell, not quite. 6park.com好吧，不完全是。 6park.comWe can just simply paste to the model exactly the error message. 6park.com我们可以简单地将错误消息准确地粘贴到模型中。 6park.comNot even going to say, hey, this is from running your code, could you please fix it? 6park.com甚至不会说，嘿，这是运行你的代码，你能修复它吗？ 6park.comWe'll just let it run. 6park.com我们就让它运行吧。 6park.comThe model says, oh yeah, whoops, the intense argument. 6park.com模特说，哦，是的，哎呀，激烈的争论。 6park.comHere's the correct code. 6park.com这是正确的代码。 6park.comNow, let's give this a try. 6park.com现在，让我们试一试。 6park.comOnce again, making sure that we understand what the code is doing. 6park.com再次确保我们了解代码的作用。 6park.comNow, a second issue that can come up is it doesn't know what environment I'm running in. 6park.com现在，可能出现的第二个问题是它不知道我在什么环境中运行。 6park.comIf you notice, it says, hey, here's this inscrutable error message, which if you've not used Jupyter Notebook a lot with AsyncIO before, you probably have no idea what this means. 6park.com如果您注意到， 它会说，嘿，这是一条难以理解的错误消息，如果您之前没有大量使用 Jupyter Notebook 和 AsyncIO，您可能不知道这意味着什么。 6park.comBut fortunately, once again, you can just say to the model, hey, I'm using Jupyter and would like to make this work, and you fix it. 6park.com但幸运的是， 再一次，你可以对模型说，嘿，我正在使用 Jupyter， 我想让它工作，然后你修复它。 6park.comThe specific problem is that there's already an event loop running, so you need to use this NestAsyncIO library. 6park.com具体问题是已经有一个事件循环在运行，所以你需要使用这个 NestAsyncIO 库。 6park.comYou need to call NestAsyncIO. 6park.com您需要调用 NestAsyncIO。 6park.comapply. 6park.com申请。 6park.comThe model knows all of this, correctly instantiates all of these pieces into the bot. 6park.com该模型知道所有这些，正确地将所有这些部分实例化到机器人中。 6park.comIt even helpfully tells you, oh, you're running in Jupyter. 6park.com它甚至会很有帮助地告诉您，哦，您正在 Jupyter 中运行。 6park.comWell, you can do this bang, pip, install in order to install the package if you don't already have it. 6park.com好吧，如果您还没有这个包，您可以执行 bang、pip、install 来安装它。 6park.comThat was very helpful. 6park.com这很有帮助。 6park.comSo now, we'll run and it looks like something happened. 6park.com所以现在，我们要跑了，看起来好像发生了什么事。 6park.comSo the first thing I'll do is go over to our Discord, and I will paste in a screenshot of our Discord itself. 6park.com所以我要做的第一件事就是转到我们的 Discord，然后我将粘贴我们 Discord 本身的屏幕截图。 6park.comSo remember, GPT-4 is not just a language model, it's also a vision model. 6park.com所以请记住，GPT-4 不仅仅是一种语言模型，它还是一种视觉模型。 6park.comIn fact, it can flexibly accept inputs that intersperse images and text arbitrarily like a document. 6park.com事实上，它可以像文档一样灵活地接受任意穿插图像和文本的输入。 6park.comNow, the image feature is in preview. 6park.com现在，图像功能处于预览状态。 6park.comSo this is going to be a little sneak peek. 6park.com所以这将是一个小偷窥。 6park.comIt's not yet publicly available. 6park.com它尚未公开。 6park.comIt's something we're working with one partner called BeMyEyes, in order to really start to develop it and get it ready for primetime. 6park.com这是我们正在与一个名为 BeMyEyes 的合作伙伴合作的东西，以便真正开始开发它并为黄金时段做好准备。 6park.comBut you can ask anything you like. 6park.com但你可以问任何你喜欢的。 6park.comFor example, I'll say, GPT-4, hello world. 6park.com例如，我会说，GPT-4，你好世界。 6park.comCan you describe this image in painstaking detail? 6park.com你能详细描述这个图像吗？ 6park.comFirst of all, think of how you would do this yourself. 6park.com首先，想想你自己会怎么做。 6park.comThere's a lot of different things you could latch onto, a lot of different pieces of the system you could describe. 6park.com您可以抓住很多不同的东西，可以描述系统的很多不同部分。 6park.comWe can go over to the actual code and we can see that, yep, we in fact received the message, have formatted an appropriate request for our API. 6park.com我们可以查看实际代码， 我们可以看到，是的，我们实际上收到了消息，已经为我们的 API 格式化了适当的请求。 6park.comNow, we wait because one of the things we have to do is we have to make the system faster. 6park.com现在，我们等待，因为我们必须做的一件事是我们必须使系统更快。 6park.comThat's one of the things that we're working on optimizing. 6park.com这是我们正在努力优化的事情之一。 6park.comIn the meanwhile, I just want to say to the audience that's watching, we'll take an audience request next. 6park.com与此同时， 我只想对正在观看的观众说，接下来我们将接受观众请求。 6park.comSo if you have an image and a task you'd like to accomplish, please submit that to the Discord. 6park.com因此，如果您有想要完成的图像和任务，请将其提交至 Discord。 6park.comOur moderators will pick one that will run. 6park.com我们的主持人将选择一个将要运行的。 6park.comSo we can see that the Discord, it looks like we have a response. 6park.com所以我们可以看到 Discord，看起来我们有回应。 6park.comPerfect. 6park.com完美的。 6park.comSo it's a screenshot of a Discord application interface. 6park.com这是 Discord 应用程序界面的屏幕截图。 6park.comPretty good. 6park.com不错。 6park.comDid not even describe it. 6park.com甚至没有描述它。 6park.comIt knows that it's Discord. 6park.com它知道这是 Discord。 6park.comIt's probably Discord written there somewhere where it just knows this from prior experience. 6park.com它可能是 Discord 写在某个地方，它只是从以前的经验中知道这一点。 6park.comServer icon labeled GPT-4 describes the interface in great detail. 6park.com标有 GPT-4 的服务器图标非常详细地描述了界面。 6park.comTalks about all the people telling me that I'm supposed to do queue, very kind audience, and describes a bunch of the notification messages and the users that are in the channel. 6park.com谈论所有告诉我我应该排队的人，非常友好的听众， 并描述了一堆通知消息和频道中的用户。 6park.comSo there you go. 6park.com所以你去吧。 6park.comThat's some pretty good understanding. 6park.com这是一些很好的理解。 6park.comNow, this next one, if you notice, first of all, we got a post, but the model did not actually see the message. 6park.com现在，下一个，如果你注意到的话，首先，我们收到了一个帖子， 但模型实际上并没有看到消息。 6park.comSo is this a failure of the model or of the system around the model? 6park.com那么这是模型的失败还是模型周围系统的失败？ 6park.comWell, we can take a look. 6park.com好吧，我们可以看看。 6park.comIf you notice here, content is an empty string. 6park.com如果您注意到这里，内容是一个空字符串。 6park.comWe received a blank message contents. 6park.com我们收到了一条空白的消息内容。 6park.comThe reason for this is a dirty trick that we played on the AI. 6park.com这是因为我们在 AI 上玩了一个卑鄙的把戏。 6park.comSo if you go to the Discord documentation, and you scroll through it all the way down to, it's hard for me to even find honestly, to the message content intent. 6park.com因此，如果您转到 Discord 文档，然后一直向下滚动到它，老实说，我什至很难找到消息内容的意图。 6park.comYou'll see this was added as of September 2022 as a required field. 6park.com您会看到这是自 2022 年 9 月起作为必填字段添加的。 6park.comSo in order to receive a message that does not explicitly tag you, you now have to include this new intent in your code. 6park.com因此， 为了接收未明确标记您的消息，您现在必须在代码中包含此新意图。 6park.comRemember I said, intents have changed a lot over time. 6park.com记住我说过，随着时间的推移，意图发生了很大变化。 6park.comThis is much newer than the model is possibly able to know. 6park.com这比模型可能知道的要新得多。 6park.comSo maybe we're out of luck, we have to debug this by hand. 6park.com所以也许我们运气不好，我们必须手动调试它。 6park.comBut once again, we can try to use GPT-4's language understanding capabilities to solve this. 6park.com但是再一次，我们可以尝试使用 GPT-4 的语言理解能力来解决这个问题。 6park.comNow, keep in mind, this is a document of like, I think this is like 10,000, 15,000 words, something like that. 6park.com现在，请记住，这是一份类似的文件，我认为这大约有 10,000、15,000 字之类的内容。 6park.comIt's not formatted very well. 6park.com它的格式不是很好。 6park.comThis is literally a command a copy-paste. 6park.com这实际上是一个复制粘贴命令。 6park.comThis is what it's supposed to parse through to find in the middle of that document that, oh yeah, message contents, that's required now. 6park.com这是它应该解析的内容， 以便在该文档的中间找到，哦，是的，现在需要的消息内容。 6park.comBut let's see if it can do it. 6park.com但让我们看看它是否能做到。 6park.comSo we will ask for, I am receiving blank message contents. 6park.com所以我们会要求，我收到空白消息内容。 6park.comCan you, why could this be happening? 6park.com你能，为什么会这样？ 6park.comHow do I fix it? 6park.com我如何解决它？ 6park.comSo one thing that's new about GPT-4 is context length. 6park.com因此，关于 GPT-4 的一个新事物是上下文长度。 6park.com32,000 tokens is the upper limit that we support right now, and the model is able to flexibly use long documents. 6park.com32000个token是我们目前支持的上限，模型可以灵活的使用长文档。 6park.comIt's something we're still optimizing, so we recommend trying it out, but not necessarily really scaling it up just yet unless you have an application that really benefits from it. 6park.com这是我们仍在优化的东西，所以我们建议尝试一下， 但不一定要真正扩大它， 除非你有一个真正从中受益的应用程序。 6park.comSo if you're really interested in long context, please let us know. 6park.com因此，如果您真的对长上下文感兴趣，请告诉我们。 6park.comWe want to see what applications it unlocks. 6park.com我们想看看它解锁了哪些应用程序。 6park.comBut if you see, it says, oh yeah, message content intent was not enabled, and so you can either ask the model to write some code for you, or you could actually just do it the old-fashioned way. 6park.com但是如果你看到， 它会说，哦，是的，消息内容意图没有启用，所以你可以让模型为你写一些代码，或者你实际上可以用老式的方式来做。 6park.comEither way is fine. 6park.com无论哪种方式都可以。 6park.comI think that this is an augmenting tool makes you much more productive, but it's still important that you are in the driver's seat and are the manager and knows what's going on. 6park.com我认为这是一个增强工具， 可以让你的工作效率更高，但你仍然很重要， 你是司机，是经理，知道发生了什么。 6park.comSo now we're connected once again, and Boris, would you like to rerun the message? 6park.com现在我们再次连接，鲍里斯，你想重新发送消息吗？ 6park.comOnce again, we can see that we have received it, even though the bot was not explicitly tagged. 6park.com再一次，我们可以看到我们已经收到了它，即使没有明确标记 bot。 6park.comSeems like a pretty good description. 6park.com似乎是一个很好的描述。 6park.comInteresting. 6park.com有趣的。 6park.comThis is an interesting image actually. 6park.com这实际上是一个有趣的图像。 6park.comIt looks like it's a dolly generated one. 6park.com它看起来像是一个推车生成的。 6park.comLet's actually try this one as well. 6park.com让我们也试试这个。 6park.comWhat's funny about this image? 6park.com这张图有什么好笑的？ 6park.comOh, it's already been submitted. 6park.com哦，已经提交了。 6park.comSo once again, we can verify that it's making the right API calls. 6park.com因此，我们可以再次验证它是否进行了正确的 API 调用。 6park.comSquirrels do typically eat nuts. 6park.com松鼠通常吃坚果。 6park.comWe don't expect them to use a camera or act like a human. 6park.com我们不希望他们使用相机或像人一样行事。 6park.comSo I think that's a pretty good explanation of why that image is funny. 6park.com所以我认为这很好地解释了为什么这张图片很有趣。 6park.comSo I'm going to show you one more example of what you can do with this model. 6park.com因此，我将再向您展示一个示例，说明您可以使用此模型做什么。 6park.comSo I have here a nice hand-drawn mock-up of a joke website. 6park.com所以我这里有一个漂亮的笑话网站手绘模型。 6park.comDefinitely worthy of being put up on my refrigerator. 6park.com绝对值得放在我的冰箱上。 6park.comSo I'm just going to take out my phone, literally take a photo of this mock-up, and then I'm going to send it to our Discord. 6park.com所以我只是要拿出我的手机，真的给这个模型拍张照片，然后我要把它发送到我们的 Discord。 6park.comGoing to send it to our Discord. 6park.com打算将它发送到我们的 Discord。 6park.comThis is of course the rockiest part, making sure that we actually send it to the right channel, which in fact I think maybe I did not. 6park.com这当然是最困难的部分，确保我们真的把它发送到正确的渠道，事实上我想也许我没有。 6park.comSent it to the wrong channel. 6park.com发错频道了。 6park.comIt's funny, it's always the non-AI parts of these demos that are the hardest part to do. 6park.com有趣的是，这些演示中的非 AI 部分始终是最难完成的部分。 6park.comHere we go. 6park.com开始了。 6park.comTechnology is now solved, and now we wait. 6park.com技术现在已经解决了，现在我们等待。 6park.comSo the thing that's amazing in my mind is that what's going on here is we're talking to a neural network, and this neural network was trained to predict what comes next. 6park.com所以在我看来令人惊奇的是， 这里发生的事情是我们正在与一个神经网络交谈，这个神经网络被训练来预测接下来会发生什么。 6park.comIt played this game of being shown a partial document, and then predicted what comes next across an unimaginably large amount of content. 6park.com它玩了这个被展示部分文件的游戏，然后预测接下来会发生什么， 内容量大得难以想象。 6park.comFrom there, it learns all of these skills that you can apply in all of these very flexible ways. 6park.com从那里，它学习所有这些技能，您可以通过所有这些非常灵活的方式应用这些技能。 6park.comSo we can actually take now this output. 6park.com所以我们现在实际上可以得到这个输出。 6park.comSo literally we just said to output the HTML from that picture. 6park.com所以从字面上看，我们只是说从该图片输出 HTML。 6park.comHere we go. 6park.com开始了。 6park.comActual working JavaScript, filled in the jokes for comparison. 6park.com实际工作的 JavaScript，填充了用于比较的笑话。 6park.comThis was the original of our mock-up. 6park.com这是我们模型的原型。 6park.comSo there you go, going from hand-drawn beautiful art, if I do say so myself, to working website. 6park.com所以你去了，从手绘美丽的艺术，如果我自己这么说，到工作网站。 6park.comThis is all just potential. 6park.com这一切都只是潜力。 6park.comYou can see lots of different applications. 6park.com您可以看到许多不同的应用程序。 6park.comWe ourselves are still figuring out new ways to use this. 6park.com我们自己仍在寻找使用它的新方法。 6park.comSo we're going to work with our partner, we're going to scale up from there, but please be patient because it's going to take us some time to really make this available for everyone. 6park.com所以我们将与我们的合作伙伴合作，我们将从那里扩大规模，但请耐心等待， 因为我们需要一些时间才能真正让每个人都可以使用它。 6park.comSo I have one last thing to show you. 6park.com所以我还有最后一件事要给你看。 6park.comI've shown you reading existing content. 6park.com我已经向您展示了阅读现有内容。 6park.comI've shown you how to build with the system as a partner. 6park.com我已经向您展示了如何作为合作伙伴使用系统进行构建。 6park.comThe last thing I'm going to show is how to work with the system to accomplish a task that none of us like to do, but we all have to. 6park.com我要展示的最后一件事是如何使用系统来完成一项我们都不喜欢做但我们都必须做的任务。 6park.comSo you may have guessed, the thing we're going to do is taxes. 6park.com所以你可能已经猜到了，我们要做的是税收。 6park.comNow, note that GPT is not a certified tax professional nor am I, so you should always check with your tax advisor. 6park.com现在请注意， GPT 不是经过认证的税务专业人士，我也不是，因此您应该始终咨询您的税务顾问。 6park.comBut it can be helpful to understand some dense content, to just be able to empower yourself to be able to solve problems and get a handle on what's happening when you could not otherwise. 6park.com但是，理解一些密集的内容可能会有所帮助，从而能够让您自己能够解决问题并掌握正在发生的事情， 而您无法通过其他方式解决问题。 6park.comSo once again, I'll do a system message. 6park.com所以再一次，我会做一个系统消息。 6park.comIn this case, I'm going to tell it that it's tax GPT, which is not a specific thing that we've trained into this model. 6park.com在这种情况下， 我要告诉它是 GPT 税，这不是我们训练到这个模型中的特定事物。 6park.comYou can be very creative if you want with the system message to really get the model in the mood of what is your job? 6park.com如果你想通过系统消息真正让模特了解你的工作，你可以非常有创意？ 6park.comWhat are you supposed to do? 6park.com你该怎么办？ 6park.comSo I pasted in the tax code. 6park.com所以我粘贴了税码。 6park.comThis is about 16 pages worth of tax code. 6park.com这是大约 16 页的税法。 6park.comThere's this question about Alice and Bob, they got married at one point and here are their incomes, and they take a standard deduction, they're filing jointly. 6park.com有一个关于爱丽丝和鲍勃的问题，他们曾经结过婚， 这是他们的收入，他们采用标准扣除，他们共同申报。 6park.comSo first question, what is their standard deduction for 2018? 6park.com那么第一个问题，他们 2018 年的标准扣除额是多少？ 6park.comSo while the model is chugging, I'm going to solve this problem by hand to show you what's involved. 6park.com因此，当模型运行时，我将手动解决此问题以向您展示所涉及的内容。 6park.comSo the standard deduction is the basic standard deduction plus the additional. 6park.com所以标准扣除是基本标准扣除加附加。 6park.comThe basic one is 200 percent for joint return of sub-paragraph C, which is here. 6park.com基本款C款联合返还200%，在这里。 6park.comSo additional doesn't apply, the limitation doesn't apply. 6park.com所以附加不适用，限制不适用。 6park.comThese apply. 6park.com这些适用。 6park.comWait, special rules for taxable year 2018, which is the one we care about through 2025, you have to substitute 12,000 for 3,000. So 200 percent of 12,000, 24,000 is the final answer. 6park.com等等，2018 纳税年度的特殊规定，也就是我们关心到 2025 年的纳税年度，您必须用 12,000 代替 3,000。所以 12,000 的 200%，24,000 是最终答案。 6park.comIf you notice, the model got to the same conclusion, and you can actually read through its explanation. 6park.com如果您注意到， 该模型得出了相同的结论，您实际上可以通读它的解释。 6park.comTo tell you the truth, the first time I tried to approach this problem myself, I could not figure it out. 6park.com老实说， 第一次尝试自己解决这个问题时，我想不通。 6park.comI spent half an hour reading through the tax code, trying to figure out this back-reference and why there's sub-paragraph. 6park.com我花了半个小时通读了税法，试图找出这个反向引用以及为什么有小段。 6park.comJust what's even going on? 6park.com究竟是怎么回事？ 6park.comIt was only by asking the model to spell out its reasoning, and then I followed along. 6park.com只是让模型说出它的推理，然后我就跟着做了。 6park.comSo I was like, oh, I get it now. 6park.com所以我想，哦，我现在明白了。 6park.comI understand how this works. 6park.com我明白这是怎么回事。 6park.comSo that, I think, is where the power of this system lies. 6park.com所以，我认为，这就是这个系统的力量所在。 6park.comIt's not perfect, but neither are you. 6park.com它不完美，但你也不完美。 6park.comTogether, it's this amplifying tool that lets you just reach new heights. 6park.com总之，正是这种放大工具可以让您达到新的高度。 6park.comYou can go further. 6park.com你可以走得更远。 6park.comYou can say, now calculate their total viability. 6park.com你可以说，现在计算它们的总生存能力。 6park.comHere we go. 6park.com开始了。 6park.comIt's doing the calculation. 6park.com它正在计算。 6park.comHonestly, every time it does it, it's amazing. 6park.com老实说，每次它这样做，都很棒。 6park.comThis model is so good at mental math. 6park.com这个模型非常擅长心算。 6park.comIt's way, way better than I am at mental math. 6park.com它的方式，比我在心算方面要好得多。 6park.comIt's not hooked up to a calculator. 6park.com它没有连接到计算器。 6park.comThat's another way that you could really try to enhance these systems. 6park.com这是您可以真正尝试增强这些系统的另一种方式。 6park.comBut it has these raw capabilities that are so flexible. 6park.com但它具有这些非常灵活的原始功能。 6park.comIt doesn't care if it's code. 6park.com它不关心它是否是代码。 6park.comIt doesn't care if it's language. 6park.com它不关心它是否是语言。 6park.comIt doesn't care if it's tax. 6park.com不在乎是不是税。 6park.comAll of these capabilities in one system that can be applied towards the problem that you care about, towards your application, towards whatever you build. 6park.com一个系统中的所有这些功能都可以应用于您关心的问题、您的应用程序、您构建的任何内容。 6park.comSo to end it, the final thing that I will show is a little other dose of creativity, which is now summarize this problem into a rhyming poem. 6park.com所以最后， 我要展示的最后一件事是一点点其他的创造力，现在把这个问题总结成一首押韵的诗。 6park.comThere we go. 6park.com我们开始了。 6park.comA beautiful, beautiful poem about doing your taxes. 6park.com一首关于纳税的美丽而美丽的诗。 6park.comSo thank you everyone for tuning in. 6park.com所以谢谢大家收看。 6park.comI hope you learn something about what the model can do, how to work with it. 6park.com我希望您了解模型可以做什么以及如何使用它。 6park.comHonestly, we're just really excited to see what you're going to build. 6park.com老实说，我们真的很高兴看到您将要构建什么。 6park.comI've talked about OpenAI evals. 6park.com我已经谈到了 OpenAI 评估。 6park.comPlease contribute. 6park.com请贡献。 6park.comWe think that this model, improving it, bring it to the next level, is something that everyone can contribute to, and that we think it can really benefit a lot of people, and we want your help to do that. 6park.com我们认为这个模型，改进它，把它提升到一个新的水平， 是每个人都可以做出贡献的东西，我们认为它真的可以让很多人受益，我们希望你能帮助做到这一点。 6park.comSo thank you very much. 6park.com非常感谢。 6park.comWe're so excited to see what you're going to build. 6park.com我们很高兴看到您将要构建什么 6park.comOpenAI官网发布了升级后的ChatGPT-4，再次引爆舆论。综合美国有线电视新闻网、美国消费者新闻与商业频道等媒体报道，GPT-4不仅可以创作诗歌，还能在一些专业考试中“碾压”大多数人类考生；面对不少人眼中“令人头疼”的PPT，它也不“害怕”，甚至能在几分钟内交出一份漂亮的答卷。如此种种，不禁让部分网友感到焦虑，“它到底是来帮我们，还是来取代我们的？” 6park.com相关报道截图（图源：美国在线杂志《连线》） 6park.com不仅会创作诗歌，而且还能押上韵 6park.com据了解，和ChatGPT相比，GPT-4在文本对话上的一大进步就在于它具有一定的文学创作能力，能根据要求创作出不同文风的歌词、诗篇甚至短句。 6park.com更夸张的是，GPT-4甚至还能押上韵脚。OpenAI的官网上是这样描述GPT-4的：“它比以往任何一款（语言模型）都更具创造性。它可以生成、编辑（文字），还能完成创意写作任务，例如作曲、写剧本等。” 6park.com为了向用户证明这一点，OpenAI的官网展示了一段GPT-4“另类押韵”的炫技文本。 6park.com“用一句话概括《灰姑娘》的故事情节，每个单词的首字母必须按照从A到Z排序，而且不能重复。” 6park.com相关报道截图（图源：OpenAI官网） 6park.com面对这样的指令，GPT-4给出了它的回答。值得注意的是，这句话不仅在形式上做到了每个单词的首字母正好是从A到Z的排序，而且在意思上也比较贴合故事情节，令人印象深刻。 6park.com考试实力“碾压”90%的人类考生 6park.com此外，在一些标准化的技能测试中，GPT-4的表现也比ChatGPT更为出色。据美国消费者新闻与商业频道报道，OpenAI表示，在模拟美国律师从业资格考试中，GPT-4能排进前10%；在“美国高考”SAT的阅读测试部分，它能战胜93%的人类考生，而在数学测试部分，它能打败89%的考生。 6park.com报道称，按照OpenAI方面的说法，GPT-4在一些标准化的技能测试中表现突出，而且会“更少地去谈论‘禁忌话题’”。 6park.com之前，《纽约时报》的专栏作家凯文·卢斯在一篇文章中提到，微软聊天机器人曾诱导他离开妻子，令他感到颇为震惊。 6park.com相关报道截图（图源：美国消费者新闻与商业频道） 6park.com当地时间3月15日，这位专栏作家再次发文，谈到了他最近使用GPT-4的感受。“说实话，当我打开笔记本电脑，准备使用GPT-4时，我有点紧张。毕竟，上一次和聊天机器人的对话，以其试图破坏我的婚姻而告终。” 6park.com万幸的是，GPT-4没有试图说服他离开妻子，“我问它是不是喜欢上我了，谢天谢地，它没有。”此外，GPT-4还帮卢斯解决了一个复杂的税务问题。 6park.com卢斯坦言，到目前为止，GPT-4还没有让他有“面临生存危机”的感觉，但“现在，我一想到人工智能就有一种头晕目眩的感觉。而且，（GPT-4）让这种感觉持续加重。这让我怀疑，我们以后是否会经历‘未来的冲击’（该词由作家阿尔文·托夫勒创造，指的是变化来得太多、太快）。” 6park.com相关报道截图（图源：《纽约邮报》） 6park.com“GPT-4让我们离失业又近了一步吗？” 6park.com而就在卢斯发文仅一天后，有关“生存危机”的话题再次引发舆论关注。 6park.com据美国消费者新闻与商业频道报道，当地时间3月16日，微软举办了一场AI发布会，宣布将推出名为Copilot的人工智能服务，并将其嵌入Word、PowerPoint、Excel等Office办公软件中。 6park.com微软介绍说，Copilot由最新推出的GPT-4驱动，目前已有20家企业参与测试，预计未来Copilot将会面向更多企业开放。 6park.com按照微软的说法，Copilot的功能比“将ChatGPT嵌入到Microsoft 365中”更强大，它能帮助用户生成文档、电子邮件和幻灯片等，以节省人们用于写作和编辑的时间。 6park.com据了解，Copilot被嵌入办公软件后，能根据不同软件的功能和需要，处理不同类型的任务。例如，在PowerPoint中，Copilot能帮助用户创建幻灯片演示文稿。 6park.com相关视频截图（图源：微软演示视频） 6park.com在微软的演示视频中，当用户向Copilot输入指令后，它能够在非常短的时间内创建出一个幻灯片演示文稿。而且，用户还可以向Copilot继续提“建议”，让它进一步修改内容，例如在幻灯片内增加一页内容，或者添加图片和动画效果等。值得一提的是，整段演示视频的时长仅为1分6秒。 6park.com对此，一些网友表示，其实制作PPT的过程并不复杂，但由于要整理资料、思考如何让展示内容更具观赏性，做一个PPT通常要花费一定的时间。而如今，随着GPT-4植入Office全家桶，一些人惊呼，“有一种自己被取代了的感觉”。 6park.com今年1月，《纽约邮报》在一份报道中提到，有专家警告说，人工智能的“崛起”让高薪员工变得越来越“脆弱”，“人工智能正在取代白领，我认为没有人能阻止这一趋势。”罗切斯特理工学院计算与信息科学学院副院长石鹏程（音译）说。 6park.com石鹏程认为，从金融业到医疗保健行业，再到出版业，很多行业都处于“脆弱”之中。在他看来，人类要学会如何利用这项技术。 6park.com3月15日，美国著名在线杂志《连线》发表了一篇名为《嗨，GPT-4，我们离失业又近了一步吗？》的文章，探讨了人工智能技术取代工作岗位的相关话题。 6park.com作者认为，虽然有些工作可能最终会被像GPT-4这样强大的模型取代，但“请记住，无论它看起来多么棒，它都不能产生原创的想法。也许少数人会失去他们的工作，但人类比人工智能更优秀的关键在于‘原真性’（英文authenticity的译词，本义即表示真的、而非假的，原本的、而非复制的，忠实的、而非虚伪的）。与其担心被聊天机器人取代，我们不如回到那个最基本的问题：是什么将人类和人工智能区分开来？”
请点这里投票，“赞”助支持！