文本选自：The Economist（经济学人）
作者：Unknown
原文标题：The bigger-is-better approach to AI is running out of road
原文发布时间：25 Jun. 2023
The bigger-is-better approach to AI is running out of road
When it comes to "large language models" (LLMs) such as GPT—which powers ChatGPT, a popular chatbot made by OpenAI, an American research lab—the clue is in the name. Modern AI systems are powered by vast artificial neural networks, bits of software modelled, very loosely, on biological brains. GPT-3, an LLM released in 2020, was a behemoth. It had 175bn "parameters", as the simulated connections between those neurons are called. It was trained by having thousands of GPUs (specialised chips that excel at AI work) crunch through hundreds of billions of words of text over the course of several weeks. All that is thought to have cost at least $4.6m.
But the most consistent result from modern AI research is that, while big is good, bigger is better. Models have therefore been growing at a blistering pace. GPT-4, released in March, is thought to have around 1trn parameters—nearly six times as many as its predecessor. Sam Altman, the firm's boss, put its development costs at more than $100m. Similar trends exist across the industry. Epoch AI, a research firm, estimated in 2022 that the computing power necessary to train a cutting-edge model was doubling every six to ten months.
This gigantism is becoming a problem. If Epoch AI's ten-monthly doubling figure is right, then training costs could exceed a billion dollars by 2026—assuming, that is, models do not run out of data first. An analysis published in October 2022 forecast that the stock of high-quality text for training may well be exhausted around the same time. And even once the training is complete, actually using the resulting model can be expensive as well. The bigger the model, the more it costs to run.
人工智能 “大即是好” 的方法快要走不通了
精听党背景导读
OpenAI于今年5月发布了ChatGPT iOS应用，该应用已经更新了多项新功能，其中包括对Siri快捷方式的支持。今天，该应用又推出了一个新功能，即集成微软的Bing搜索引擎，但这个功能只对付费用户开放。
据更新说明介绍，ChatGPT Plus订阅用户可以使用一个名为“浏览”的新功能。该功能结合了GPT-4和Bing搜索，当启用时，ChatGPT可以提供“全面的答案和对事件和信息的最新见解，这些内容超出了模型最初的训练数据范围。”换言之，这意味着ChatGPT能够联网使用Bing搜索，并根据最新数据提供更准确的答案。
文本选自：The Economist（经济学人）作者：Unknown原文标题：The bigger-is-better approach to AI is running out of road原文发布时间：25 Jun. 2023关键词：人工智能 模型 成本
精听党带着问题听
1.“走投无路”用英语可以怎么表达？2. 如何理解段一中的“excel at”？3. 大型语言模型GPT-3的训练成本大概是多少？
精听党选段赏析
标题解读
The bigger-is-better approach to AI is running out of road
人工智能 “大即是好” 的方法快要走不通了
approach n. 方法；1. The company’s approach to customer service is excellent. 这家公司对客户服务的态度非常好。
run out of road 走投无路；快到尽头了；1. The company’s current business model is running out of road and they need to come up with a new strategy. 公司目前的商业模式快到尽头了，他们需要想出新的策略。
段一
When it comes to “large language models” (LLMs) such as GPT—which powers ChatGPT, a popular chatbot made by OpenAI, an American research lab—the clue is in the name. Modern AI systems are powered by vast artificial neural networks, bits of software modelled, very loosely, on biological brains. GPT-3, an LLM released in 2020, was a behemoth. It had 175bn “parameters”, as the simulated connections between those neurons are called. It was trained by having thousands of GPUs (specialised chips that excel at AI work) crunch through hundreds of billions of words of text over the course of several weeks. All that is thought to have cost at least $4.6m.
when it comes to 当谈到…；当提到…；1. When it comes to music, I'm a layman. 说到音乐，我是个外行。
chatbot n. 一种聊天程序；聊天机器人；
clue n. 线索；1. The detective followed the clue to solve the case. 侦探循着线索破案了。
vast adj. 巨大的；1. The vast ocean stretched out before us. 浩瀚的海洋展现在我们面前。
neural adj. 神经的，神经系统的；1. neuron n. [解剖] 神经元，神经单位；
loosely adv. 宽松地；不精确地；
behemoth n. 巨兽；庞然大物；1. The new skyscraper is a behemoth compared to the surrounding buildings. 与周围的建筑物相比，这座新摩天大楼简直是个庞然大物。
parameter n. 参数；1. The scientist adjusted the parameters of the experiment. 科学家调整了实验的参数。
simulated adj. 模拟的；1. The flight simulator provides a simulated experience of flying a plane. 飞行模拟器提供了驾驶飞机的模拟体验。
excel at 擅长于（某项活动）；在…方面表现杰出；
crunch vi. 嘎吱作响地咬碎；1. She crunched on an apple while she worked. 她工作时嘎吱嘎吱地吃着一个苹果。
course n. 过程；1. The course of true love never did run smooth. 真爱之路永不会是平坦的。（莎士比亚）2. over the course of 在…期间；
参考译文
说到“大型语言模型”(LLMs)，比如GPT——它为ChatGPT提供动力，ChatGPT是美国研究实验室OpenAI制造的一款流行的聊天机器人——名字里就有线索。现代人工智能系统由庞大的人工神经网络驱动，这是一种非常不精确地模仿生物大脑的软件。2020年发布的大型语言模型GPT-3是一个庞然大物。它有1750亿个“参数”，这些神经元之间的模拟连接被称为“参数”。它是通过数千个GPU（擅长人工智能工作的专用芯片）在几周内处理数千亿字的文本来训练的。所有这些被认为至少花费了460万美元。
段二
But the most consistent result from modern AI research is that, while big is good, bigger is better. Models have therefore been growing at a blistering pace. GPT-4, released in March, is thought to have around 1trn parameters—nearly six times as many as its predecessor. Sam Altman, the firm’s boss, put its development costs at more than $100m. Similar trends exist across the industry. Epoch AI, a research firm, estimated in 2022 that the computing power necessary to train a cutting-edge model was doubling every six to ten months.
consistent adj. 一致的；1. His work has been consistent throughout his career. 他的工作在他的职业生涯中始终如一。
blistering adj. 非常快的；1. The athlete ran at a blistering pace. 运动员以极快的速度奔跑。
predecessor n. 前任；
put vt. 说，表达；
trend n. 趋势；1. Fashion trends come and go. 时尚潮流总是变幻无常。
cutting-edge adj. 尖端的；1. The company is known for its cutting-edge technology. 公司以其尖端技术而闻名。
double vi. 加倍；1. The population of the city is doubling every year. 这座城市的人口每年都在加倍。
参考译文
但现代人工智能研究最一致的结果是，虽然大是好，但越大越好。因此，模型一直在以惊人的速度增长。3月份发布的GPT-4被认为有大约1万亿参数，几乎是其前身的6倍。公司老板萨姆·阿尔特曼（Sam Altman）认为其开发成本超过1亿美元。整个行业都存在类似的趋势。研究公司Epoch AI在2022年估计，训练一个尖端模型所需的计算能力每六到十个月就会翻一番。
段三
This gigantism is becoming a problem. If Epoch AI’s ten-monthly doubling figure is right, then training costs could exceed a billion dollars by 2026—assuming, that is, models do not run out of data first. An analysis published in October 2022 forecast that the stock of high-quality text for training may well be exhausted around the same time. And even once the training is complete, actually using the resulting model can be expensive as well. The bigger the model, the more it costs to run.
gigantism n. 巨人症；
exceed vt. 超过；1. His performance exceeded our expectations. 他的表现超出了我们的预期。
forecast vt. 预测；1. The weatherman forecasted rain for tomorrow. 天气预报员预报明天有雨。
stock n.（商店的）现货，存货，库存；
exhausted adj. 筋疲力尽的；1. After running the marathon, I was completely exhausted. 跑完马拉松后，我已经筋疲力尽了。
参考译文
这种巨人症正在成为一个问题。如果Epoch AI的10个月翻一番的数据是正确的，那么到2026年，训练成本可能会超过10亿美元——假设模型没有首先耗尽数据。2022年10月发表的一项分析预测，用于培训的高质量文本库存可能会在同一时间耗尽。即使训练完成了，实际使用得到的模型也可能是昂贵的。模型越大，运行成本就越高。
精听党每日单词
approach
/əˈproʊtʃ/ n. 方法；
run out of road
走投无路；快到尽头了；
when it comes to
当谈到…；当提到…；
chatbot
/ˈtʃætˌbɒt/ n. 一种聊天程序；聊天机器人；
clue
/kluː/ n. 线索；
vast
/væst/ adj. 巨大的；
neural
/ˈnʊrəl/ adj. 神经的，神经系统的；
loosely
/ˈluːsli/ adv. 宽松地；不精确地；
behemoth
/bɪˈhiːməθ/ n. 巨兽；庞然大物；
parameter
/pəˈræmɪtər/ n. 参数；
simulated
/ˈsɪmjuleɪtɪd/ adj. 模拟的；
excel at
擅长于（某项活动）；在…方面表现杰出；
crunch
/krʌntʃ/ vi. 嘎吱作响地咬碎；
course
/kɔːrs/ n. 过程；
consistent
/kənˈsɪstənt/ adj. 一致的；
blistering
/ˈblɪstərɪŋ/ adj. 非常快的；
predecessor
/ˈpredəsesər/ n. 前任；
put
/pʊt/ vt. 说，表达；
trend
/trend/ n. 趋势；
cutting-edge
/ˌkʌtɪŋ ˈedʒ/ adj. 尖端的；
double
/ˈdʌb(ə)l/ vi. 加倍；
gigantism
/dʒaɪˈɡæntɪzəmˌˈdʒaɪɡæntɪzəm/ n. 巨人症；
exceed
/ɪkˈsiːd/ vt. 超过；
forecast
/ˈfɔːrkæst/ vt. 预测；
stock
/stɑːk/ n.（商店的）现货，存货，库存；
exhausted
/ɪɡˈzɔːstɪd/ adj. 筋疲力尽的；
精听党文化拓展
众所周知，像 ChatGPT 这样的能够根据提示生成相关内容的 AI 工具已经在科技行业掀起了一股热潮，包括微软和 Alphabet 公司在内的许多企业都将它们集成到了搜索引擎中。但就目前来看，许多商业用户仍对生成式人工智能技术持谨慎态度，因为它可能会编造瞎话。
甲骨文表示，其人力资源软件正在被大型企业用于招聘新员工和提供绩效评估等任务，而甲骨文将在该软件的许多字段上添加一个按钮，该按钮将自动为职位列表或绩效目标等事项生成草稿文本。
甲骨文自适应智能应用程序产品管理副总裁理查德・布赫海姆 (Rich Buchheim) 说，将人工智能助手以按钮的形式呈现，而不是直接告诉你答案，是为了确保这些按钮“产生良好和安全的结果”。布赫海姆表示，生成的文本仍然需要人工审核。“我们不认为生成式 AI 能帮你找到目标。它只会给你一个起点，并提供有用的信息供你参考。”
甲骨文的数据分析和人力资源管理技术与创新副总裁盖伊・沃特曼 (Guy Waterman) 表示，在未来，甲骨文正在研究如何使用人工智能进行更复杂的人力资源管理任务，例如如何在不同市场遵守当地法规的情况下编写符合要求的职位要求清单。“可能需要一周或两周的时间才能做出决定并实施。如果我们能将其从几周缩短到几小时甚至几分钟，那么这就是生成式人工智能真正发挥作用的地方，”沃特曼说。
精听党每日美句
One who has seen the ocean thinks nothing of mere rivers.
曾经沧海难为水。