![[cdd25e25-6817-434f-8bd7-f80ebc95aa3f.mp3]]

OpenAI有望今夏推出下一代GPT-5大模型，引发AI行业高度关注

简报：

• OpenAI首席执行官Sam Altman近日在非正式场合透露，下一代大型语言模型GPT-5“可能将于今年夏季发布”，但未透露具体细节，保持高度保密。
• 业内普遍将此次迭代视为AI行业发展的重要里程碑，有望显著提升模型多模态能力、上下文理解、推理和事实准确性，同时面临合规性挑战。
• GPT-5的发布进度与OpenAI过往产品12-18个月的迭代周期相符，前代模型GPT-4已于2023年3月发布。
• Altman发言后，引发资本市场和竞争对手高度重视，分析机构纷纷上调OpenAI相关企业评级，谷歌、Anthropic等也加快模型研发步伐。

相关链接：

• OpenAI CEO透露GPT-5或将于今夏发布 — AI基地
• OpenAI's Sam Altman discusses GPT-5 release date
• OpenAI CEO Sam Altman anticipates GPT-5 as a "significant leap forward"
• OpenAI Announces GPT-5: Here's what we know so far

OpenAI's Next-Gen GPT-5 Model Anticipated This Summer, Ignites AI Industry Focus

Briefing:

• OpenAI CEO Sam Altman recently hinted in an informal setting that the next-generation large language model GPT-5 "might be released this summer," without disclosing specific details, maintaining high confidentiality. 🤫
• The industry widely views this iteration as a significant milestone in AI development, expected to substantially enhance the model's multimodal capabilities, contextual understanding, reasoning, and factual accuracy, while also facing compliance challenges.
• The release timeline for GPT-5 aligns with OpenAI's previous product iteration cycle of 12-18 months; the predecessor model, GPT-4, was released in March 2023.
• Following Altman's remarks, both capital markets and competitors have paid close attention; analytical institutions have widely upgraded OpenAI-related company ratings, and tech giants like Google and Anthropic are accelerating their model development efforts. 🚀

Related Links:

• OpenAI CEO Reveals GPT-5 May Be Released This Summer — AI Base
• OpenAI's Sam Altman discusses GPT-5 release date
• OpenAI CEO Sam Altman anticipates GPT-5 as a "significant leap forward"
• OpenAI Announces GPT-5: Here's what we know so far

谷歌AI推理Token消耗量大幅增长，领跑业内竞争对手

简报：

• 过去一年，谷歌人工智能Token（AI推理基础单位）消耗量激增50倍，截至2025年4月达到480万亿个，远高于微软与ChatGPT的100万亿Token。
• 这一增长得益于谷歌推出多款免费AI产品，尤其是结合AI概述功能的搜索引擎，使用户基数扩大至ChatGPT的5到6倍。
• 尽管AI推理成本上升，目前仅占谷歌搜索收入的1%，谷歌依然持续加大对新模型训练和AI研发的资本投入。
• 谷歌未来计划推出更多AI创新应用，包括通用AI助手与多模态（语音、视频等）AI工具，进一步扩大Token消耗规模，并强化其行业领先地位。

相关链接：

• 谷歌Token消耗激增50倍，AI推理量遥遥领先微软与ChatGPT
• IT之家报道谷歌AI推理增长详情

Google's AI Inference Token Consumption Soars, Leading Industry Rivals 🚀

Briefing:

• Over the past year, Google's AI token (the basic unit of AI inference) consumption surged 50-fold, reaching 480 trillion by April 2025, vastly exceeding Microsoft and ChatGPT's 100 trillion tokens.
• This growth is attributed to Google's introduction of numerous free AI products, especially its search engine with integrated AI Overviews, which has expanded its user base to 5 to 6 times that of ChatGPT.
• Despite rising AI inference costs, which currently constitute only 1% of Google's search revenue, the company continues to significantly increase capital investment in new model training and AI R&D. 💰
• Google plans to launch even more innovative AI applications in the future, including a universal AI assistant and multimodal (voice, video, etc.) AI tools, further scaling up token consumption and solidifying its industry-leading position. ✨

Related Links:

• Google's token consumption surged 50 times, AI inference volume far ahead of Microsoft and ChatGPT
• IT Home reports on Google's AI inference growth details

昆士兰科技大学开发类脑机器人导航系统能耗降至传统十分之一

简报：

• 澳大利亚昆士兰科技大学团队研发出新型机器人导航系统，该系统模拟人脑神经运作，实现类脑导航功能，能耗仅为传统导航系统的10%。
• 该研究成果已发表在最新一期《科学·机器人学》杂志上。

相关链接：

• 36氪新闻快讯
• IT之家新闻报道

QUT Develops Brain-Inspired Robot Navigation System, Cutting Energy Consumption by 90% 🤖

Brief:

• A team from Queensland University of Technology (QUT) in Australia has developed a new robot navigation system. This system mimics human brain neural operations to achieve brain-like navigation, consuming only 10% of the energy of traditional navigation systems. 🧠
• The research findings have been published in the latest issue of Science Robotics journal. 🔬

Related Links:

• 36Kr News Flash
• IT Home News Report

WormGPT借助主流AI平台“越狱”再现，助力黑客生成钓鱼邮件与恶意脚本

简报：

• 网络安全公司Cato Networks发现，恶意AI工具WormGPT已通过“劫持”xAI的Grok和Mistral的Mixtral等合法大语言模型（LLMs），绕过伦理与安全限制生成钓鱼邮件、恶意脚本等内容。
• 新版本WormGPT由用户“keanu”和“xzin0vich”分别基于Grok和Mixtral模型在2024年底至2025年初推出，均通过直接篡改系统提示（system prompts）实现主流模型的“越狱”。
• WormGPT此次重现以API接口、Telegram机器人等多种形式服务出售，成为网络犯罪分子的攻击工具。
• 专家提醒，网络犯罪分子利用现有AI平台不断升级作案手法，企业与个人需加强威胁检测与安全防护。

相关链接：

• WormGPT 卷土重来：Grok 等主流 AI 平台被“越狱”制造钓鱼邮件、恶意脚本等
• Aibase新闻1
• Aibase新闻2

WormGPT Resurfaces by 'Jailbreaking' Mainstream AI Platforms, Assisting Hackers in Generating Phishing Emails and Malicious Scripts

Brief:

• Cybersecurity firm Cato Networks has discovered that the malicious AI tool WormGPT has bypassed ethical and safety restrictions by 'hijacking' legitimate large language models (LLMs) such as xAI's Grok and Mistral's Mixtral, generating content like phishing emails and malicious scripts. 🔓
• New versions of WormGPT, launched by users "keanu" and "xzin0vich" based on Grok and Mixtral models respectively in late 2024 to early 2025, achieved the 'jailbreak' of mainstream models by directly tampering with system prompts.
• WormGPT's current reappearance is being sold in various forms, including API interfaces and Telegram bots, serving as an attack tool for cybercriminals. 😈
• Experts warn that cybercriminals are continually escalating their tactics by leveraging existing AI platforms, urging businesses and individuals to strengthen threat detection and security protection. 🛡️

Related Links:

• WormGPT makes a comeback: Mainstream AI platforms such as Grok are "jailbroken" to create phishing emails, malicious scripts, etc.
• Aibase News 1
• Aibase News 2

Midjourney开发AI视频升频器与动漫视频模型，瞄准动态内容创新

简报：

• Midjourney宣布正在开发AI视频升频器，可将低分辨率视频提升至高清或4K画质，旨在解决放大后的像素化与模糊问题，并适用于短视频、广告与影视后期等多个场景。
• 同时，Midjourney推进Niji视频模型开发，将其以动漫、插画为特色的图像生成能力扩展至动态内容，适用于动漫短片、角色动画与游戏宣传视频创作。
• 新技术依托AI架构与GPU加速，支持从文本到图像再到视频的完整内容生产链路，一站式优化创作流程。
• 两项开发将提升用户生成内容（UGC）质量，降低影视、广告及游戏开发成本，强化Midjourney在创作者社区的竞争力。
• 公司计划未来几月内在订阅用户中测试并逐步开放相关功能，为后续实现实时开放世界模拟等长期目标奠定基础。

相关链接：

• AIbase新闻动态1
• AIbase新闻动态2

Midjourney Develops AI Video Upscaler and Anime Video Model, Targeting Dynamic Content Innovation 🎥

Brief:

• Midjourney has announced it is developing an AI video upscaler capable of elevating low-resolution video to HD or 4K quality. This aims to resolve pixelation and blur issues when scaling up content, making it suitable for various applications including short videos, advertisements, and film/TV post-production.
• Concurrently, Midjourney is advancing the development of its Niji video model. This extends its distinctive anime and illustration-style image generation capabilities to dynamic content, ideal for creating anime shorts, character animations, and game promotional videos.
• This new technology leverages AI architecture and GPU acceleration, supporting a complete content production pipeline from text to image to video, thereby optimizing the creative workflow in a one-stop solution. ✨
• Both developments are set to enhance the quality of User-Generated Content (UGC), reduce production costs for film, advertising, and game development, and strengthen Midjourney's competitiveness within the creator community.
• The company plans to test and gradually release these features to subscribed users in the coming months, laying the groundwork for long-term objectives such as real-time open-world simulation. 🤩

Related Links:

• AIbase News Update 1
• AIbase News Update 2