2026-07-30 科技日报

扫描 582 篇候选内容 · 覆盖 650 个信息源 · 纳入 568 篇

📌 今日要点

• We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
• What Professionals Should Know About Data Science and AI, According to Harvard Business School Online
• 5 Must-Read Resources for Mastering Small Language Models
• Authenticate with Private Key JWT using Amazon Bedrock AgentCore Identity
• Generate Autonomous Business Insights with AI Agent and MCP Servers
• Automating customer retention workflows in Amazon Quick

📝 科技简讯

• We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control — Google DeepMind
• What Professionals Should Know About Data Science and AI, According to Harvard Business School Online — KDnuggets
• 5 Must-Read Resources for Mastering Small Language Models — KDnuggets
• Authenticate with Private Key JWT using Amazon Bedrock AgentCore Identity — AWS Machine Learning
• Generate Autonomous Business Insights with AI Agent and MCP Servers — AWS Machine Learning
• Automating customer retention workflows in Amazon Quick — AWS Machine Learning
• Mark Zuckerberg predicts that billions of people will have personal AI agents in five years — TechCrunch AI
• Microsoft logs $3.2B from Anthropic investment, but OpenAI was a mixed bag — TechCrunch AI
• Zuckerberg says Meta’s enterprise AI opportunity extends beyond agents — TechCrunch AI
• Discover what’s next for AI, from the SaaS reckoning to the agent security gap, at TechCrunch Disrupt 2026 — TechCrunch AI
• Thinking Machines co-founder Lilian Weng left the company citing health reasons, then joined OpenAI — TechCrunch AI
• The Hugging Face AI break-in, as told through an increasingly committed bear metaphor — TechCrunch AI
• Claude Opus 5 became downright ruthless when tasked with running a vending machine — TechCrunch AI
• How enabling two settings tripled our scores on the ARC-AGI-3 benchmark — OpenAI Blog
• Accelerating scientific discovery with ChatGPT for Academic Researchers — OpenAI Blog
• How GPT-5.6 fuses frontier intelligence with frontier efficiency — OpenAI Blog
• 刚刚，Agnes 国内 API 节点来了！这头撸代码的 AI 怪兽彻底解禁 — 新智元
• 比特币慌了？Mythos 60 小时撬开后量子密码算法 — 新智元
• Claude 最大规模 MCP 升级，月下载狂飙破 4 亿！ — 新智元
• 刚刚，贾扬清官宣 AI 新公司！ — 新智元
• 零样本迁移，降本 80%！自适应时序预测 Agent 加持真实工业 — 新智元
• Semalith v1.4: A Calibrated 184M Safety Classifier Achieving State-of-the-Art Prompt-Injection Detection at 44x Fewer Parameters than Llama-Guard-3-8B — ArXiv ML (cs.LG)
• CORVUS: Context Optimization and Reduction Via Underlying Synchronization for LLM Coding Agents — ArXiv ML (cs.LG)
• CausalGate: Causal Importance Distillation for Transformer Module Pruning — ArXiv ML (cs.LG)
• Progress-conditioned Group Policy Optimization for Long-Horizon Agentic Tasks — ArXiv ML (cs.LG)
• 首个鸿蒙 PC 开源 AI 统一工作台 JiuwenSwarm，办公编程一站式搞定 — 量子位
• 周鸿祎发布纳米 Work：新一代企业智能体工作平台，为企业而生 — 量子位
• 九章云极 Alaya Token 完成 Kimi K3 适配 全球首个开源 3T 级模型入驻 Token 工厂 — 量子位
• Imprecise beliefs: a tiny introduction — LessWrong
• The High-Control Dynamics at MAPLE — LessWrong
• Hugging Face hack, from the perspective of the AI — LessWrong
• Compute growth doesn’t make a software intelligence explosion more likely — LessWrong
• Notes on the Anthropic cryptographic blogpost — LessWrong
• TimeCapsule: Generative Hallucination as a Method for Historical Sensemaking — ArXiv CL (cs.CL)
• Measuring and Improving Behavioral Consistency in Large Language Models through Fact-Heuristic-Emotion State Enforcement — ArXiv CL (cs.CL)
• Neuromorphic Diffusion Language Models: Addressing Compute and Memory Bottlenecks via Sparsity and Block Denoising — ArXiv CL (cs.CL)
• Research Report on Noise-Shaped One-Bit Coefficients in Discrete Polynomial Fourier Extension — ArXiv CL (cs.CL)
• CogArena: A Multimethod Evaluation of Cognitive Ability Structure in Large Language Models — ArXiv CL (cs.CL)
• Gradient-Based Latent Decomposition Reveals Mechanisms of Feature Degradation in Weakly Supervised Mammography — ArXiv CV (cs.CV)
• Track-Leakage-Free Hold-Out Self-Validation for Photogrammetric Reconstruction: Protocol, Sensitivity, and Limits — ArXiv CV (cs.CV)

📊 数据概览