OC

前谷歌人工智能研究员表示，ChatGPT 的成功本来可以来得更早

我如何将 Jakub Uszkoreit 采访中的问题和答案翻译成捷克语？这是翻译：采访：Jakub Uszkoreit 关于 Google 和生物软件Jakub Uszkoreit 在 Google 从事 Google Translate 项目工作多年后，创立了 Inceptive 公司，致力于利用人工智能开发生化软件。你在干什么2021年，我们创立了Inceptive，利用深度学习和快速生化实验来制造可编程的药物。我们认为这是迈向所谓的生物软件的第一步。什么是生物软件？生物软件与计算机软件类似，都具有行为规范，然后由人工智能编译器将其转换为分子，当将其输入生物系统时，这些分子会表现出特定的功能。这是否类似于针对冠状病毒的 mRNA 疫苗？mRNA 疫苗就是一个简单的例子。编译器说“制造改良的病毒抗原”，我们的细胞制造这种蛋白质。但你可以想象具有更复杂行为的分子，例如 RNA 病毒。如何不产生有害的 RNA 序列？医学实践长期以来一直存在于科学之外，因此已经开发了许多保护系统。我们还将使用这些来开发我们的分子，从单细胞的小型系统开始，沿着既定的控制程序进行。感谢您抽出时间。谢谢。

使用 AI 将您的教练业务扩大 10 倍：借助 ChatGPT 和其他工具

顶级教练正在利用人工智能，特别是 ChatGPT，通过加快研究、分解复杂的想法、强化品牌形象、构建可扩展的项目以及增加课程的情感深度来增强他们的实践。这些工具可实现更个性化的客户体验、更广泛的覆盖范围并提高管理工作负载的效率，同时加深客户关系。

Will Spotify's AI DJ make you listen longer? Spotify is testing an AI DJ feature that personalizes song selections and introduces them with a warm, realistic voice for English-speaking users (referred to as "X") and Spanish-speaking users ("Livi"). The AI uses data on user preferences to create playlists and explain why certain songs are recommended. Spotify claims this increases listener engagement and reduces song skipping. However, some质疑这种背景信息是否必要，而Spotify的测试结果显示，这些语音提示鼓励用户发现新音乐或尘封已久的歌曲，从而延长了用户的听歌时间。尽管AI DJ功能受到一些人的欢迎，但也有人认为它可能令人分心或困惑。 Will Spotify's AI DJ make you listen longer? Spotify AI DJ plays songs based on data it gathers about users' personal preferences and introduces some of them with a warm voice, similar to a traditional radio DJ. Spotify hide caption toggle caption Spotify Humans want fundamentally incompatible things from music services. "We want to listen to all the stuff that everyone likes, but also only the stuff that we like," said New Yorker staff writer Kelefa Sanneh, the author of Major Labels: A History of Popular Music in Seven Genres . "And we want it to be broad and social, but also narrow and personal."  Sanneh said this problem is very hard to solve. But that's not stopping Spotify from trying — with its AI DJ feature. The AI DJ, which the music streamer began rolling out nearly two years ago and is still in beta, builds on Spotify's long track record of providing listeners with personalized playlists based on individual data. Spotify has millions of listeners, and a staff of real human beings couldn't voice contextual recommendations for each one of them. But AI can. Personalized DJ "This is your personalized DJ just for you," said Ziad Sultan, Spotify's vice president of personalization. Sultan said the DJ feature goes further than creating personalized playlists: For English-speaking listeners, it uses an AI voice known as "X" to tell the listener why it is making the selections it is making. (The streamer also offers a Spanish language AI DJ, "Livi.") Spotify employee Xavier Jernigan is the voice behind X, the streamer's English-language AI DJ. Spotify hide caption toggle caption Spotify "X is this incredibly realistic but also warm voice that makes people want to engage and listen to the DJ and follow their recommendations," Sultan said. The company's research suggests these occasional voice prompts encourage users to discover new or long-forgotten music and so listen longer, said Xavier Jernigan, the Spotify employee who is the talent behind X. Spotify declined to share its research. "We know through our data that when people hear stories about the artists, about the songs and know the why as to why we recommended that song, they're less likely to skip it," said Jernigan. "It's about storytelling. It's about human connection."  But — do listeners really want it? Others question whether this kind of context is necessary. "Our goal is to understand the user well enough, and understand the music well enough, that there aren't those sorts of necessities to have to contextualize it to you," said Joe Inzerillo, the chief product and technology officer at SiriusXM, which owns Spotify competitor Pandora. That's why Pandora doesn't have one. The New Yorker 's Sanneh said he sometimes finds the Spotify AI DJ to be distracting or baffling — like when it introduced a block of songs by describing them as "music that a lot of people seem to think is good for cleaning their house to," and then proceeded to play, among other titles, Shakira's "Si Te Vas" and Celine Dion's cover of "River Deep Mountain High." " And so now I'm thinking about Celine, and I'm like, 'Do people clean their house to Celine Dion?'" Yet some people quite like the Spotify AI DJ — including DJ Umami, a real, human DJ who spins on the radio and in clubs, and is a resident DJ for the San Francisco Giants. "It plays my likes on shuffle based on when I actually listen to this type of music and tempo and genre and all of those data points," Umami said. "Then they'll throw in a couple songs that they think that I would like. And I love that." Jennifer Vanasco edited the broadcast and digital versions of this story. Chloee Weiner mixed the audio.

2025 年：“隐形”人工智能代理将融入企业层级

麦肯锡表示，未来的企业将把人类员工与复杂的人工智能代理整合起来，使员工的工作实现 60-70% 的自动化。目前，大约三分之一的美国工人在雇主不知情的情况下使用人工智能。到2025年，这些“隐形”的人工智能代理预计将在运营中发挥更积极的作用。AI 代理可以跨多个系统执行复杂的任务，并与其他代理或 API 集成，以根据上下文自主执行操作。虽然基本的人工智能代理作为 RAG plus 运行，但更先进的人工智能代理将形成多代理生态系统，尽管由于安全和监管问题，完全自主的情况很少见。使用案例包括简化员工入职流程、协助销售团队以及通过信息收集和起草回复来增强客户服务。

Anthropic 的新 AI 工具有望简化提示写作并将准确性提高 30%

Anthropic 在其开发者控制台中引入了新工具，旨在自动化和增强 Claude 等人工智能模型的即时工程。主要功能包括使用最佳实践改进现有提示的“提示改进器”，以及确保输出格式一致的示例管理系统。这些更新旨在提高企业人工智能开发的效率和可靠性，据报道准确性提高高达 30%。这些工具还有助于更轻松地从其他人工智能系统迁移到 Claude，将 Anthropic 定位为负责任且实用的企业人工智能解决方案的领导者。

Fitbit 可能会在 2025 年推出人工智能驱动的睡眠日记功能

Fitbit 正在为 Fitbit Premium 订阅者开发睡眠日记功能，该功能可能由人工智能提供支持，根据用户输入提供个性化睡眠提示。该功能允许用户通过文本或语音记录记录他们的睡眠问题，并且该应用程序使用这些数据来提供量身定制的建议。在版本 4.30.fitbit-mobile-110146981-694155636 的 APK 拆解过程中发现，该功能仍处于开发的早期阶段，具有基本的 UI。使用该功能可能需要用户提交每日条目以获得更深入的见解和个性化建议。

世界上第一款人工智能生成的游戏可供任何人在线玩，而且它是超现实的

一个新的 AI 生成的 Minecraft 克隆名为 Oasis，允许玩家探索 3D 世界并实时改变他们的环境。该游戏使用经过数百万游戏时间训练的人工智能来生成每一帧，并根据上传的图像提供可定制的起始环境。尽管采用了创新的方法，但由于实时生成，该游戏仍会出现意想不到的故障，例如随机出现的物品以及在未观察到时景观会发生变化。开发人员认为这是迈向人工智能驱动的交互式世界模拟的一步，但承认现阶段这更多的是一种好奇心，而不是一个有趣的游戏。

人工智能的不公平决定可能会让我们对人类的不良行为漠不关心

人工智能在影响教育、就业、医疗保健和政府援助等生活各个方面的决策中发挥着至关重要的作用。然而，对人工智能的日益依赖引发了人们对公平和透明度的担忧，并有可能加剧社会不平等。发表在《认知》上的研究表明，经历人工智能带来的不公平可能会导致“人工智能引起的冷漠”，由于对不道德行为的不敏感，个人在事后不太可能解决人类的不当行为。无论是否熟悉人工智能技术，这种效应都会持续存在。政策制定者和开发人员必须专注于减少人工智能系统中的偏见并提高透明度，以减轻这些社会影响并维护道德标准。

从问题到发现：NASA 的新 Earth Copilot 带来了 Microsoft AI 功能，使复杂数据的访问民主化 - 微软官方博客

NASA 与 Microsoft 合作开发了 Earth Copilot，这是一种利用 Azure OpenAI 服务的人工智能驱动工具，旨在简化对 NASA 广泛的地球科学数据的访问。该计划旨在使用户能够通过自然语言查询与复杂的地理空间数据集进行交互，从而实现数据访问的民主化，从而促进在农业、城市规划、灾害应对和气候研究等各个领域更轻松的分析和更广泛的应用。此次合作利用 Azure 的云基础设施和机器学习功能来提高数据发现和分析的效率，与 NASA 的开放科学计划保持一致，以实现更具包容性的科学探索。

Diablo Canyon 的生成式人工智能部署是美国核电行业的首次：PG&E

Pacific Gas & Electric (PG&E) 将在其 Diablo Canyon 核电站部署 Atomic Canyon 的 Neutron Enterprise 解决方案，这标志着生成式人工智能首次在美国核设施中进行现场使用。Neutron Enterprise 基于 NVIDIA 的 AI 平台构建，旨在提高文档搜索和检索效率，从而有可能减少员工搜索记录所花费的时间。初步部署定于 2025 年初进行，并计划扩大其在工厂维护、运营、设计和工程功能中的作用。Atomic Canyon 与橡树岭国家实验室合作，专门针对核能用例开发了 Neutron Enterprise。