AI 情报文章归档

Diffusion models generalize early in training and later reproduce individual training samples. Standard tests detect memorization only once one-shot generation produces near-copies, leaving a released model unaudited until its outputs fail…

A coding-agent score is evidence only when the dataset, the metric, and the control surface are...

Reliable predictive uncertainty in Laplace approximations depends critically on the prior precision, yet selecting it requires a posterior sweep that is prohibitively expensive for neural networks with billions of parameters. Under the Max…

Online platforms contain growing numbers of bots, deceptive reviewers, and scam accounts that imitate legitimate users. Such camouflage blurs graph neighborhoods and behavioral attributes, making it difficult for graph neural networks (GNN…

Deep reinforcement learning agents reach strong performance in real-time strategy games but can be brittle against opponents outside their training distribution. Separating strategic command selection from learned unit control allows diffe…

Autonomous driving capabilities depend strongly on the distribution of scenarios encountered during training. Existing methods commonly construct or dynamically adapt training scenario distributions using surrogate criteria such as realism…

On-policy distillation (OPD) supervises student-generated trajectories with token-level teacher signals. Its sampled-token variant avoids the cost of full-vocabulary probabilities. Yet we find that training on fewer tokens can outperform f…

Originally published on metalbear.com. In January we published How Our Engineering Team Uses AI. The...

Improving a long-horizon LLM agent means evolving the harness around a frozen model or training its weights. We let a self-evolving harness make the system stronger first, then cross seed and evolved harnesses with base and trained weights…

Automated interviewers and conversational agents are increasingly used in research, recruitment, customer service, and education. However, many existing systems rely on fixed question sequences and provide limited context-based personaliza…

What we built Prompt Heist is a browser game about prompt injection. You are the Cipher...

We, Team Latent, built TrialBridge during the Hacktoberfest Hack Day Coimbatore x (INIT Club &...

Every week, perfectly fixable things end up in the bin. A kettle with a tripped fuse. A lamp with a...

I built Prompt Heist for the Hacktoberfest Hack Day (Coimbatore x Init Club and Idea Club). It is a...

一名 16 岁少年用 Anthropic 的 Claude 规划温哥华 Crown Mountain 登顶路线，结果被困在需攀岩装备的陡峭崖壁"Widowmaker Arete"一处约五英尺宽的岩台上，警方通知北岸救援队用直升机将其吊离。救援负责人 Paul Markey 表示 Claude 并不真正了解地形，AI 无法替代经验与常识；该少年称自己承担责任，今后仍会攀岩但不再用 AI 规划路线。

据彭博社报道，田柯宇（Tian Keyu）的未具名创业公司已从包括五源资本和IDG在内的投资者处融资近3000万美元，投后估值2亿美元。 这位NeurIPS 2024获奖研究者带领一支10人团队，正在开发一个20万符号的视觉词表，以帮助AI处理视频。 田柯宇声称该方法可将视频生成成本至少降低十倍。完整模型计划于2027年发布；公司目前尚无产品。

OpenAI published hundreds of AI-produced math results before fully formalizing them. Two days later, three are gone — and that is the system working, not failing.

My comment to yesterdays Dev.to post: Web components failed. We can fix them became a full post: ...

GroundedDocs answers questions over a document set with exact citations and refuses when the evidence...

邵猛整理了斯坦福 CS329Z《Engineering AI Agents》第4讲（Diyi Yang 主讲）的内容，讲解 LLM 如何学会调用工具。

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass ...

This is a submission for the Kaggle Benchmarking Challenge What I Benchmarked I've been a...

邵猛转发阿里「AI Native 研发范式实践手册」结尾的四点官方陈述并逐条评论：从写代码到可靠交付仍有技术问题要解决、企业知识资产尚未 Agent 友好、组织设计影响深远更难有确定性答案、AI 迭代速度会持续冲击现有架构。

Crowdstrike 报告称，一名疑似中文使用者于2026年9月底至10月初利用 AI 驱动的开源渗透测试工具 ARTEX 攻击多家韩国金融机构，窃取大量数据，其中 Shinhan Bank 超过 25,000 条包含姓名、联系方式、收入和信用额度的记录泄露。

I use coding agents across real repositories every day. I run a small security firm, and between...

Here's a line from YouTube's auto-captions on a lecture about attention in neural...

I build Soba, an iPhone app that looks at a photo of a meal and estimates the carbs in it. People...

🌿 Trailnote — An AI Trail Companion That Tells You to Put Your Phone Away My submission...

Architect Financial Technologies 发布 Liquid Inference，一个对每次请求进行实时拍卖的 LLM 推理市场，价格最低且满足买家规则的报价胜出，首 token 生成前锁定最高价。

Store reviews, users who couldn't reach the store, payments and analytics. What building a solo AI side-panel extension actually took.

Most AI chat tools are terrible language tutors. Ask a generic chatbot to practice Chinese with you...

NVIDIA 联合普林斯顿大学和马里兰大学提出 PivotOPD，一种面向多轮 LLM 智能体的在线策略蒸馏方法，训练智能体避免最致命的早期错误并在错误发生后恢复。

The biggest enterprise integration risk in 2026 is not AI. It is operational neglect. This week,...

A published coding-agent score describes one frozen task slice, and it does not describe an entire...

Um benchmark reprodutível de pequenos modelos de decisão e de embeddings para classificar o objetivo do cliente em português do Brasil, e o que uma tradução paralela para o inglês revela sobre o custo do idioma.

A reproducible benchmark of small choice deciders and embeddings for customer-goal classification in Brazilian Portuguese, plus what a parallel English translation reveals about the cost of language.

After ten years in asset management, I started a full-stack developer path in 2024 and spent my time...

Anthropic 推出 Claude Haiku 5.5，支持 100 万 token 上下文窗口和可调 effort，定价为每百万 token 输入 $0.10、输出 $0.50；Claude Sonnet 5.5 的 prompt cache 读取价格从 $0.20 降至 $0.10。

保持冷静，与我们的 day0 合作伙伴 @vercel_dev 一起构建 🤘 再享受一周 Ling-3.1-flash——一款 token 高效、擅长智能体与编程的模型~ Ling 3.1 Flash 已在 AI Gateway 上线，10 月 13 日前免费。专为编程和多步分析打造。

A native app on both stores, in five languages, built by one person with an AI as the developer. This...