模型与评测

Click here to view our TODO List (in Chinese)

Terms of Service

Google brings AI music generation directly into the Gemini app with its new Lyria 3.5 model

Google has released its Lyria 3.5 music model in the Gemini app and via API. The model promises more expressive vocals and richer arrangements and is also available through Flow Music, AI Studio, and Google Vids. Google says it was trained only on licensed content. The article Go

GitHub Copilot weekly releases — August 31

This week, GitHub Copilot expands model choice and content protections, while VS Code adds new ways to manage agent sessions and get pull requests merge-ready. GitHub Copilot, general Claude Fable 5.1 is available to Copilot Pro+, Max, Business, and Enterprise users. Gemini 3.8 F

Build from Source

Online Flashing

esp-claw.com/

local build documentation

online survey (in Chinese)

The purpose of DNS is to spread scams

The purpose of DNS is to spread scams Terence Eden shares some daunting statistics in support of his take that "the Domain Name System's purpose seems to be a vector for criminals to run scams on people at a terrifyingly high rate". On this Interisle report ( via Andrew Campling

OpenAI 宣布达到自动化研究实习生里程碑，内部 agent 运行时达人力 3.1 倍

OpenAI 发布内部数据称已达到自动化研究实习生里程碑，即可在人类监督下完成熟练研究员需数天的明确任务。截至 8 月中旬，其研究组织每投入 1 个人工工作日，就使用 3.1 个 agent 工作日的运行时长，该比值衡量的是运行时间而非等效生产力；原文作者援引 OpenAI 员工观点称递归自我改进或成为未来几年 AI 能力的关键，并呼吁其他 AI 公司同样公开数据。 🔗 阅读原文 via AIHOT · https://aihot.virxact.com/items/cmtq5pxc10275roiufscxyie4

There's No Limit to How Bad Code Can Get

My comment on There's No Limit to How Bad Code Can Get — Lobste.rs. [In reply to a comment about burning it down to start from scratch when technical debt becomes overwhelming] In my experience it's so rare for that to work. You announce the old thing is irrecoverably drowning in

Quoting Zach Kehs

If you continue to add floors and rooms to a building forever, it will collapse. Software faces no such constraint. The code can always get worse. There can always be a new layer of indirection or a reduction in performance. — Zach Kehs , There's No Limit to How Bad Code Can Get

OpenAI 向 Pro、Enterprise 和 Business Premium 用户推出 GPT-6 Astra，消息额度约为 GPT-5.6 Sol 的一半

OpenAI 通过 ChatGPT Work 和 Codex 向 Pro、Enterprise、Business Premium 计划用户开放 GPT-6 Astra，并通过 API、Microsoft Azure 和 AWS Bedrock 提供。 🔗 阅读原文 via AIHOT · https://aihot.virxact.com/items/cmto3ih1m016groxt2ypyqos3

OpenAI 向 Pro、Enterprise 和 Business Premium 用户推出 GPT-6 Astra，消息额度约为 GPT-5.6 Sol 的一半

OpenAI 通过 ChatGPT Work 和 Codex 向 Pro、Enterprise、Business Premium 计划用户开放 GPT-6 Astra，并通过 API、Microsoft Azure 和 AWS Bedrock 提供。

OpenAI 长文阐述对齐与监测困境，称 CoT 监控能力正在减弱

OpenAI 发布长文《An Alien Mind》，回溯 2023 年 RLSlow 项目中确认推理模型可扩展训练的起点，并系统阐述目标对齐与价值对齐的区分。文章指出链式思维监控的效果正随着模型能力提升而逐步减弱，GPT-6 Astra 在对齐上显著优于 GPT-5.6 Sol；作者预期进展可能持续走向机器递归自我改进（RSI），呼吁自愿放缓扩展、建立第三方安全门槛并加强国际协调。 🔗 阅读原文 via AIHOT · https://aihot.virxact.com/items/cmtq23v6k01aorotwh8xip8r1

OpenAI 发布内部研究加速报告：已达成自动化研究实习生目标，推进 2028 年 3 月自动化 AI 研究员

OpenAI 发文披露自动化研究进展，宣布已达成去年秋天设定的今年 9 月拥有自动化研究实习生（可在人类指导下完成耗时数天的明确研究任务）的目标，并计划在 2028 年 3 月前造出自动化 AI 研究员。 🔗 阅读原文 via AIHOT · https://aihot.virxact.com/items/cmtpzyon6019iroemu68v4s2v

Fortune 报道 OpenAI 多次修改 GPT-6 Astra 基准测试数据，部分成绩大幅变化

据 Fortune 报道，OpenAI 自 9 月 3 日发布 GPT-6 Astra 公告以来多次修改评测基准数据：Astra 幻觉率曾从 4.2% 降至 2% 后又改回。

微软 Foundry 模型路由扩容：覆盖区域增至 28 个，更新模型池

Clean Engineering, Unstable Measurement: A Preregistered Reliability Failure of Black-Box LLM Observers on Shared Endpoints

Language-model judges now gate training data, score generations, and drive leaderboards. The judge is then a measurement instrument, resting on one rarely stated assumption: the same request, sent to the same model name, reads the same tomorrow. We audited that assumption in two

Clean Engineering, Unstable Measurement: A Preregistered Reliability Failure of Black-Box LLM Observers on Shared Endpoints

Language-model judges now gate training data, score generations, and drive leaderboards. The judge is then a measurement instrument, resting on one rarely stated assumption: the same request, sent to the same model name, reads the same tomorrow. We audited that assumption in two

Value-Preserving Architectures for Agentic AI Systems

The emergence of agentic AI and LLM-based multi-agent systems (MAS) presents unprecedented opportunities for automating complex tasks, while simultaneously raising critical concerns about the preservation of fundamental human-centered values, such as privacy, fairness, and safety

A Blind Trust, the Bloody Thrust: When Attacker-Controlled Hook Updates Steer AI Agent Harnesses towards Malicious Behaviors

Modern AI agent harnesses expose lifecycle hooks that bind shell commands to runtime events such as session start, tool calls, and file edits. These commands run with host privileges yet ship as lifecycle-hook configuration and may fire at times the LLM never observes. We identif

Adapting to Evolving Requirements: Agentic AI for Retail Supply Chain Operations

Retail supply chain operations rely on coupled decision modules that must adapt as requirements evolve. LLMs offer a natural-language interface for this task, but existing methods primarily focus on individual optimization models. Extending them to heterogeneous decision pipeline