🚀 AI 成長日報

2026-08-05 | 每日 AI 知識更新

🇺🇸 歐美 AI 熱門新聞

Spotify expands AI remix and covers project with Merlin partnership

Spotify says Merlin, which represents more than 30,000 independent labels and distributors, has joined Universal Music Group in backing its upcoming AI-powered remix and covers product. The paid tool will let fans create AI-generated covers and remixes of participating artists’ music while ensuring

techcrunch.com ·

Texas halts new data centers as governor calls for audits

Tech companies and developers have been scouring the U.S. for places to build data centers, and they’ve been drawn to Texas’ loose regulations and seemingly abundant power supply. But even Texas can be pushed to the brink.

techcrunch.com ·

🇨🇳 中國 AI 新聞

从实验到产线——AI 工作流的规模化挑战与协作生态 | 2026 ChinaJoy AI未来生态大会

AI工作流如何从实验室的“惊艳一刻”走向产线的“日常运转”？当生成能力不再是门槛，规模化落地的瓶颈在哪里，人又该扮演什么角色？

内容生产正在经历从“技术验证”到“产业落地”的关键跨越。单点突破易，系统协同难，真正的挑战不在于模型能否生成，而在于工作流能否稳定运转、协作生态能否有效构建。技术红利终会趋平，越过规模化这道坎之后，行业最后比拼的是对场景的理解深度与工程化的落地能力。2026 ChinaJoy AI未来生态大会上，36氪游戏与Funloom AI、阿里云、VAST、珀乐互动等行业先行者，共同带来圆桌探讨。

圆桌嘉宾

36kr.com ·

由三星SDS牵头的韩国国家人工智能算力中心正式开工

为增强韩国人工智能技术竞争力、拓展AI服务，被称作“人工智能高速通道”的人工智能算力中心建设工作正式启动。8月3日，韩国科学技术信息通信部在全罗南道光州市海南郡索拉西多数据中心园区，举办人工智能算力中心奠基仪式。该项目总投资2.5万亿韩元，占地面积约4.8万平方米，规划建设一栋两层建筑，预计2028年竣工投用。去年，三星SDS联合NAVER云、三星物产、三星电子、Kakao、KT、光州市政府以及西南海岸企业城市开发公司组建联合体，通过公开招标拿下该项目。（财联社）

36kr.com ·

📄 國際新發表 AI 論文

OpenClaw and Ollama in Agentic AI: Toward Fully Autonomous and Scalable AI Agent Systems

Authors: · Published:

標題: OpenClaw and Ollama in Agentic AI: Toward Fully Autonomous and Scalable AI Agent Systems | 內容: arXiv:2607.28629v1 Announce Type: new 摘要: 從 reactive LLMs 快速過渡至 persistent, action-capable systems，已暴露出 Agentic AI 在 architectural understanding 上的關鍵缺口，特別是 autonomous AI agents 的 inference, orchestration, 與 execution layers 分離。儘管 recent advances，designing and evaluating full-stack agentic systems 的 unified frameworks 仍顯不足。本文提出 comprehensive, layered architecture for Agentic AI，outlining the evolut

Can AI Evaluate AI Scientists? A Benchmarking Study of Autonomous Research Generation Systems Using Automated Multi-Model Review

Authors: · Published:

標題: Can AI Evaluate AI Scientists? A Benchmarking Study of Autonomous Research Generation Systems Using Automated Multi-Model Review | 內容: arXiv:2607.28631v1 Announce Type: new 摘要: 具備 autonomous research 能力的 AI Scientist systems 有潛力大幅加速 scientific discovery。然而，evaluating and comparing the quality of AI-generated papers 仍是一大挑戰。我們提出並實作 rigorous benchmarking protocol，採用 automated peer-review system，運用 frontier large language models 從四個 core dimensions 評估 scientific papers：originality, scientific r

LLM Framework for Discovering Major Mathematical Conjectures: AI's Quest for the Next Riemann Hypothesis

Authors: · Published:

標題: LLM Framework for Discovering Major Mathematical Conjectures: AI's Quest for the Next Riemann Hypothesis | 內容: arXiv:2607.28632v1 Announce Type: new 摘要: Major mathematical conjectures 仍高度依賴 expert intuition，因此缺乏 unified method 進行 systematic generation and validation of conjectures with substantial mathematical potential。我們提出 three stage pipeline for major conjecture discovery：透過 explicit local evidence modules 進行 region search、針對 foundationality, novelty, 與 potential significance 進行 reflect

ThinkReset: Learnable Intermediate Interface Construction for Bounded-Context Long-Horizon Reasoning

Authors: · Published:

標題: ThinkReset: Learnable Intermediate Interface Construction for Bounded-Context Long-Horizon Reasoning | 內容: arXiv:2607.28642v1 Announce Type: new 摘要: Long chain-of-thought reasoning 能提升 complex problems 的 performance，但也會引入 redundancy accumulation, context overflow, 與 error anchoring。我們認為，在 bounded context windows 下，core bottleneck 並非 trajectory compression 或 test-time control，而是 absence of a reusable intermediate interface 以 replace discarded history 並 support continued solving。我們進一步指出 outco

TAPR: Enhancing LLM Performance with a Task-Aware Prompt Rewriter

Authors: · Published:

標題: TAPR: Enhancing LLM Performance with a Task-Aware Prompt Rewriter | 內容: arXiv:2607.28657v1 Announce Type: new 摘要: Large Language Models (LLMs) 通常需要 carefully crafted prompts 才能 unlock their full potential，這對 non-expert users 可能構成 barrier。本研究透過 introducing a Task-Aware Prompt Rewriter (TAPR) 解決此 challenge。TAPR 是一種 model，能 reformulate user prompts 為 task-optimized prompts，explicit goal 為 improving downstream LLM performance。我們使用 reinforcement learning with Group Relative Policy Optimization (

🔓 開源論壇熱門 Top 10

🤗 Hugging Face 竄起模型

🎬 YouTube AI 熱門影片

📺 wow.insight 頻道新影片

📺 Greyson Zhang 頻道新影片

📺 Best Partners 頻道新影片

🎤 AI 教父/大神訪談

📺 中國 Bilibili AI 熱門

🌐 數位網站與設計趨勢

📝 writing-mode

writing-mode

📅 Tue, 21 Jul 2026 09:39:10 +0000 · CSS-Tricks (CSS/設計)

📝 pointer-events

pointer-events

📅 Wed, 15 Jul 2026 19:57:51 +0000 · CSS-Tricks (CSS/設計)