AAI Search
AI 资讯日报 · 2026-08-20

2026-08-20 · AI 资讯日报

今日新收录 119 条公开资讯，按模型 / 产品 / 行业 / 论文 / 观点 自动归类汇编（非 AI 生成，点击可溯源原文）。

今日精选6 条

  1. 1.
    论文研究ADEPT: Accelerating Dexterity via Pre-Training and Post-Training using Reinforcement Learning— arXiv

    We introduce Accelerating Dexterity via Pre-Training (ADEPT), a large-scale reinforcement learning (RL) framework for learning sim-to-real transferable dexterity across high degree-of-freedom (DoF) ro…

  2. 2.
    论文研究Beyond Teacher Likelihood: Group-Calibrated On-Policy Distillation for Long-Context Reasoning— arXiv

    On-policy distillation (OPD) trains a student on its own responses using dense token-level guidance from a stronger teacher. In long-context tasks, however, token-level teacher support can favor local…

  3. 3.
    论文研究Finetuning Strategies for Querying Sounds by Vocal Imitation— arXiv

    This technical report describes our winning submission to the AES AIMLA 2025 Challenge on querying sound effects by vocal imitation. We investigate two complementary fine-tuning strategies: contrastiv…

  4. 4.
    论文研究Lévy Attention: Single-Pass Predictive Uncertainty for Continuous-Time Attention— arXiv

    Deep models for irregularly-sampled time series answer queries at arbitrary continuous timestamps, yet report nothing about how far each answer should be trusted. We show the attention layer itself ca…

  5. 5.
    论文研究Learned, Then Lost: A Measured Single-Example Counterfactual in Pre-training— arXiv

    A single training example's contribution to a finished model is normally estimated rather than measured, because measuring it takes two expensive full pre-training runs that differ in one row of one b…

  6. 6.
    论文研究ChildSafeAds Shared Task 2026: Commercial Content in Child-Facing YouTube Videos— arXiv

    ChildSafeAds is a shared task on commercial content in YouTube videos likely to reach children and teenagers. It contains 3,360 videos from 939 channels. Each instance begins with a segment submitted…

模型发布/更新6 条

  1. 1.
    Introducing Intelligence Age— OpenAI

    Introducing Intelligence Age, a new OpenAI blog exploring how transformative AI could reshape power, governance, the economy, and individual freedom.

  2. 2.
    Stampli cuts launch hours by 68% using ChatGPT Work— OpenAI

    With a fixed deadline and design resources committed elsewhere, Stampli used Codex and ChatGPT Work to compress weeks of launch production into days.

  3. 3.
    A third of web pages published since ChatGPT launched were written by AI, study finds— TechCrunch

    AI 导读 · AI渗透网络内容创作成常态，揭示生成式AI对信息生态的深层影响。

  4. 4.
    Agentic Search. More accurate and efficient results from your AI systems.— Mistral AI

    The retrieval layer that helps AI systems navigate, read, and verify information inside even the most complex documents

  5. 5.
    Introducing cross-Region inference for OpenAI GPT-5.6 models on Amazon Bedrock— AWS ML

    AI 导读 · 跨区域推理打破地域限制，全球可用性大幅提升，企业部署AI模型更灵活高效。

  6. 6.
    Jobs - Anthropic— Anthropic

    Jobs Anthropic

产品发布/更新5 条

  1. 7.
    Ramp launches its own AI model router, called Router— TechCrunch

    AI 导读 · 模型路由成新战场，Ramp自研方案降低切换成本，或重塑企业AI应用生态。

  2. 8.
    AmazingAng/old-coder— GitHub

    An old coder's strategy for the agent era: don't read the code — make it run the gauntlet. Evidence-first development skill for coding agents, inspired by Uncle…

  3. 9.
    zorost/AI-Engineering-Lab— GitHub

    A free, self-paced 24-week AI engineering course: Python, machine learning, LLMs, RAG, fine-tuning, agents and MCP, Azure and Vertex and Bedrock, and Databricks…

  4. 10.
    six-nut/PocketMen-with-you— GitHub

    Create high-fidelity Codex companions from 2+ photos with local open-weight neural editing and no OpenAI API key.

  5. 11.
    AlgoNoRhythm/Flare— GitHub

    The graph based agentic IDE

行业动态6 条

  1. 12.
    Show HN: I trained a 125M model to autocomplete piano on-device— Hacker News

    I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except ins…

  2. 13.
    Don't paste the AI, please— Hacker News
  3. 14.
    Adobe Firefly adds AI audio tools and Google's Gemini Omni Flash— The Decoder

    AI 导读 · AI音频创作门槛骤降，Adobe生态整合或重塑配音行业。

  4. 15.
    Show HN: Huzzah – a novel approach to coding with AI— Hacker News

    AI 导读 · AI编程工具新范式，直击代码代理协作痛点，值得开发者实测。

  5. 16.
    LLMs could write like humans but post-training guardrails make their text detectable— The Decoder

    AI 导读 · 安全护栏压缩表达空间，揭示大模型文本可检测性的深层技术博弈。

  6. 17.
    Anti-AI fonts are useless and harmful— Hacker News

论文研究6 条

  1. 18.
    ADEPT: Accelerating Dexterity via Pre-Training and Post-Training using Reinforcement Learning— arXiv

    We introduce Accelerating Dexterity via Pre-Training (ADEPT), a large-scale reinforcement learning (RL) framework for learning sim-to-real transferable dexterity across high degree-of-freedom (DoF) ro…

  2. 19.
    Beyond Teacher Likelihood: Group-Calibrated On-Policy Distillation for Long-Context Reasoning— arXiv

    On-policy distillation (OPD) trains a student on its own responses using dense token-level guidance from a stronger teacher. In long-context tasks, however, token-level teacher support can favor local…

  3. 20.
    Finetuning Strategies for Querying Sounds by Vocal Imitation— arXiv

    This technical report describes our winning submission to the AES AIMLA 2025 Challenge on querying sound effects by vocal imitation. We investigate two complementary fine-tuning strategies: contrastiv…

  4. 21.
    Lévy Attention: Single-Pass Predictive Uncertainty for Continuous-Time Attention— arXiv

    Deep models for irregularly-sampled time series answer queries at arbitrary continuous timestamps, yet report nothing about how far each answer should be trusted. We show the attention layer itself ca…

  5. 22.
    Learned, Then Lost: A Measured Single-Example Counterfactual in Pre-training— arXiv

    A single training example's contribution to a finished model is normally estimated rather than measured, because measuring it takes two expensive full pre-training runs that differ in one row of one b…

  6. 23.
    ChildSafeAds Shared Task 2026: Commercial Content in Child-Facing YouTube Videos— arXiv

    ChildSafeAds is a shared task on commercial content in YouTube videos likely to reach children and teenagers. It contains 3,360 videos from 939 channels. Each instance begins with a segment submitted…

技巧与观点6 条

  1. 24.
    not much happened today— AI News

    **Ornith-1.5** launches as a new open-weight model family with **9B dense, 35B MoE, and 397B MoE** variants under **MIT license**, featuring quantized formats like **FP8, GGUF, MLX…

  2. 25.
    How to control costs and show ROI for Claude Code on Google Cloud— Anthropic

    How to control costs and show ROI for Claude Code on Google Cloud Anthropic

  3. 26.
    Up to 3.2x Faster Inference with LFM2.5-DSpark— HuggingFace Blog

    AI 导读 · 推理速度提升3.2倍，直击大模型部署成本痛点，性能与效率兼得。

  4. 27.
    Build a no-code ML workflow with Snowflake, Amazon SageMaker Canvas and Amazon Quick – Part 2: Data preparation and model building with Amazon SageMaker Canvas— AWS ML

    AI 导读 · 无代码ML再进一步，打通Snowflake与SageMaker，让数据准备和建模更亲民。

  5. 28.
    Build a no-code ML workflow with Snowflake, Amazon SageMaker Canvas and Amazon Quick – Part 3: Visualizing insights with Amazon Quick Sight— AWS ML

    AI 导读 · 低代码ML全链路落地，从预测到可视化无缝衔接，企业分析效率倍增。

  6. 29.
    Authoring Dogwood policies from natural language in Amazon Bedrock AgentCore— AWS ML

    AI 导读 · 自然语言生成策略，简化AI代理治理，安全管控更灵活高效。

快讯

日报生成时间：2026-09-06
← 2026-08-19最新日报2026-08-21 →