2025-09-26 12:00:00
苹果在WWDC 2025期间介绍了其基础模型框架，使开发人员能够将公司的本地AI模型整合到其应用程序中，而不会产生推理成本。这些模型包括引导生成和工具调用等功能。随着iOS 26的推出，Lil Artist，Daylish，MoneyCoach，查找，任务，第一天，Crouton，Signeasy，Dark Noise，Lights Out，Capture，Lumy，CardPointers和Guitar Wiz都将Apple的AI模型融合在一起，以增强具有本地功能的用户体验，从而使iOS 26推出。

2025-09-26 11:36:16
大主教保罗·理查德·加拉格尔（Paul Richard Gallagher）在联合国安理会致辞，呼吁暂停致命自治武器（法律），因为他们缺乏道德决策能力。他强调需要采用以人为中心的人工智能发展方法，强调尊重人类尊严和共同利益。加拉格尔（Gallagher）警告不要在军事系统中涉及AI的军备竞赛，包括核指挥机构，强调了和平与安全的潜在风险。他敦促安全理事会密切监视技术进步，并确保新兴技术服务于人类，而不是造成分裂或破坏。

2025-09-26 11:30:01
超级富裕的家庭办事处，例如杰夫·贝佐斯（Jeff Bezos）和埃里克·施密特（Eric Sc​​hmidt）的家庭办事处，正在通过公共股票而不是直接投资在AI初创公司投资。高盛的一项调查显示，52％的全球家庭办事处通过原始公共股票或ETF暴露于AI，而只有四分之一的投资直接投资于初创企业。与私人市场相比，由于更合理的估值，家庭办事处越来越喜欢AI股票。他们还表现出对使用AI的公司的生产力和效率提高的公司以及能源提供者等次要受益人的兴趣。值得注意的是，有89％的被调查家庭办事处具有某种形式的AI投资或计划，这强调了AI在财富管理策略中的重要性日益增长。

2025-09-26 11:03:18
It’s time to push back against the AI internet The Dead Internet Theory,虽为虚假阴谋论，但在实际操作中却有其道理。据研究显示，2025年中期，约74%的新在线内容由AI或机器人生成，预计年底这一比例将超过90%。随着AI技术的发展，包括Podcast、视频和书籍在内的各种类型的内容越来越多地被机器生成，引发了对人类创造者角色的质疑。为了应对这一趋势，用户应要求平台提供工具来优先展示人类创作的内容，并选择避开由AI生成的信息，以保障网络内容的质量与真实性。 Everyone should be clamoring for tools and features that block AI content. And for content companies, blocking AI slop is your new Job One. The Dead Internet Theory is a false conspiracy theory. But in practical terms, it might as well be true.  Emerging from the deranged muck of 4chan and Wizardchan in the late 2010s, the Dead Internet Theory holds that secret cabals of all-powerful government or corporate conspirators use bots and AI-generated content to replace humans on the internet. The goal: to manipulate public perception, control narratives, and influence the public’s behavior.  A central tenet of the theory is that most online content is generated by bots and AI, not people.  The false part is the conspiracy. The true part is that most internet content is indeed bots and AI.  Midway through 2025, roughly 74% of newly created content online was generated with the help of AI or bots, according to several large-scale studies. Only about one-quarter of online content is created by people without AI assistance. And the rate of change is rising fast. By the end of the year, more than 90% of all content will be AI-generated, according to some predictions.  The Dead Internet Theory isn’t true. But it might as well be true. To quote the late comedian George Carlin: “You don’t need a formal conspiracy when interests converge.” And in Silicon Valley, the interests are definitely converging.  GenAI content, now on an epic scale A company called Inception Point AI is churning out AI-generated podcasts on an industrial scale, powered by custom AI agents that leverage OpenAI, Perplexity, Claude, Gemini, and other chatbots to build the content. The company’s Quiet Please Podcast Network has created more than 5,000 podcast shows — not episodes, shows! — hosted by more than 50 AI “personalities.” The company intends to create thousands more “personalities” in the future.  It costs them $1 per episode to produce. So, if an episode sells a $2 ad, they make a profit.  Based on what the company’s CEO says in public, Inception Point AI seems captured by the delusion that AI-generated personas are human beings. It’s unknown how many listeners have been deluded into believing that the AI podcast hosts are real people. (The AI hosts identify themselves as such at the top of each episode.)  “We believe that in the near future half the people on the planet will be AI, and we are the company that’s bringing those people to life,” said CEO Jeanine Wright, ignoring the advice of Microsoft AI CEO Mustafa Suleyman, who said in an essay last month that “We must build AI for people; not to be a digital person.” If they believe they’re “bringing people to life,” then they by definition believe themselves to be gods.  She added that the “people who are still referring to all AI-generated content as AI slop are probably lazy Luddites.”(People who create a podcast start to finish by pushing a button are calling others “lazy.”) Of course, Inception Point AI isn’t alone in the industry. Others, including PodcastAI, Wondercraft AI, and Jellypod,are flooding the zone with fake-people podcasts, too.  The AI podcast startups have a lot of catching up to do. The video startups are way ahead in terms of volume.  Companies like T-Series, Sony SAB, SET India, and Zee TV have produced between 20,000 and 234,000 videos each. T-Series leads with nearly 24,000 videos, while some channels like Zee TV have exceeded 215,000 uploads. It’s everywhere Google is helping them. Creator AI Studios — which is built into YouTube’s ecosystem — enables small teams and solo creators to publish hundreds of videos per day through auto-editing, thumbnail generation, scene detection, and AI-generated scripts. Platforms like Argil AI, RightBlogger, Team-GPT, and  Designs.ai  let creators generate scripted TikTok videos without cameras using AI models for ideas, editing, and even synthetic voiceovers. AI-generated books are on the rise, too. Two years ago, Amazon had to cap book uploads to three books per daybecause people were uploading far more than that.  Estimates suggest that more than 70% of new self-published Kindle books are partially or fully AI-generated. On Amazon.com alone, people may be collectively uploading as many as 1,000 AI-generated books per day.  While the state of the art in text, audio, and video fakery ranges now from “very good” to “perfect,” the ability to make or alter photos using AI took a huge leap forward with Google Gemini 2.5 Flash Image (a.k.a. “Nano Banana”). Perfect pictures are a banality now.  More than 15 billion images have been created using text-to-image algorithms since 2022. Roughly 34 million new such images are churned out every day now, using models based on Stable Diffusion and others.  They’re used for business objectives in advertising, media and entertainment, e-commerce, fashion, and architecture, as well as for digital art and the lucrative field of online influencing. The technology is also used maliciously to create fake news and propaganda, scams, non-consensual deepfake pornography, and AI-generated child sexual abuse material. The key benefit of AI-generated images — whether for benign and malignant uses — is low cost. AI pictures are super cheap to make.  Studies of platforms like Facebook show that users often fail to recognize images as synthetic, even when they’re badly made and look ridiculous. A 2025 Microsoft report, for instance, found that 73% of survey respondents found it hard to spot AI-generated images, and correctly identified them only 38% of the time.  We are months away from an internet where more than 99% of online content is AI-generated. As a result, the human race is being rapidly frog-marched into a world where we interact primarily, or even exclusively, with machines.  Whose idea was this? (Probably the same people who decided to replace “writers” and “readers” with “content creators” and “content consumers.”)  Has it occurred to anyone that the purpose of “content” — articles, books, photographs, videos and the recorded voice — exists for human beings to communicate with each other, rather than for machines to shovel data at people?  And why do we accept this? We need to start holding the content delivery companies to account.  Demand prioritization for people-created content It’s time users demand, pay for, or exclusively use services that give you the choice to prioritize human-generated content.  The paid-search engine Kagi Search, for example, enables users to avoid AI-generated content primarily through its image search filtering and labeling system, which allows users to choose whether to include, exclude, or exclusively display AI-generated images. Personalization features enable users to block or downrank specific domains if unwanted AI-generated or low-quality imagery slips past initial filters, and these controls are accessible both in image and web search. (Full disclosure: My son works at Kagi.) DuckDuckGo also allows users to filter out AI-generated images from search results by providing a dedicated dropdown menu in its image search interface, where users can select to hide all AI images.  Some stock photo platforms, such as Freepik, have added tools to exclude AI-generated results from search queries.  But this list is pathetically short. While Google offers hard-to-find and ineffectual “NOPE” buttons to turn off some AI content, Microsoft Bing, Facebook, Instagram, Reddit, X, LinkedIn, Pinterest, other sites do not offer users the ability to opt out of AI at all. If the industry is going to provide the tools for replacing nearly all online content with AI slop, then they must also provide the tools to opt out.  We must demand the option to see content created by people either primarily or exclusively. It’s time for the living to rise up against the dead internet.  SUBSCRIBE TO OUR NEWSLETTER From our editors straight to your inbox Get started by entering your email address below.

2025-09-26 10:45:59
本周的情节与内特（Nate）的讨论有关他的新书“如果有人建造它，每个人都死了：为什么超人AI会杀死我们所有人”，与Eliezer Yudkowsky共同撰写。该书通过寓言探索了超人AI的风险，这些寓言突出了智能众生如何在没有恶意意图人类意图的情况下追求资源。谈话还深入研究了媒体中AI的更好的讲故事方法，尤其是赞扬“ Ex Machina”，因为它对AI的动机的细微刻画。

2025-09-26 10:31:00
Glassboro公立学区已成为第一个使用AI来检测枪支并立即发出警报的新系统的客户，旨在加强针对学校枪支暴力的安全措施。

2025-09-26 10:30:54
评论讨论了人工智能（AI）均化人类行为和创造力的风险。旨在最大化用户参与度和满意度的AI系统通常建议流行或熟悉的选择，而不是新的或独特的选择。随着时间的流逝，这可以使个人在偏好和利益方面变得越来越少。作者认为，尽管个性化的AI代理可以学习个人偏好，但他们仍然倾向于在这些参数中安全地发挥作用，可能导致个人复杂性和独特性的丧失。为了应对这一趋势，评论提出激励AI系统鼓励探索和冒险，从而维护了生活的各个领域的人类多样性和创造力。评论

2025-09-26 10:30:00
由于AI行业的增长，Palantir和Nvidia的股价大幅上涨。Palantir报告了其有史以来的首个季度收入10亿美元，同比增长48％，同时将全年指导提高到41亿美元。NVIDIA的销售额同比增长56％，达到467亿美元的第2季度，尽管美国对中国的出口限制，但仍被预计将达到540亿美元。两家公司都利用AI的进步来增长。Palantir通过其人工智能平台，并通过AI系统中使用的高级半导体芯片扩展到联邦合规服务，NVIDIA。

2025-09-26 10:17:49
坦帕湾时报（Tampa Bay Times）与健康的圣彼得堡基金会合作就AI进行了免费的社区讨论。在活动期间，小组成员的要点集中在人工智能上。

2025-09-26 09:50:00
由于零售投资者的受欢迎程度，过去一年中有两个小型AI股票（BBAI）和Soundhound AI（Soun）在过去一年中的增长超过200％。BigBear.ai拥有26亿美元的市值，确保了政府合同并拥有强劲的积压，而Soundhound AI则使其客户群多样化，并在Amelia AI的收购后看到了巨大的收入增长。尽管有风险，BigBear.ai因其财务稳定性和政府部门的长期机会而受到青睐。