AI 热点 实时全景观测

Ask HN: Will AI trigger mass IP protectionism in software?

The search for consciousness inside AI

China Is Building AI Models of American Voters

Palantir's Karp – frontier AI labs that are 'trying to drug addict us'

Why can AI generate Super Mario but not a wedge ramp for my robot vacuum?

Treat AI Like an Intern, Not Software: A Stanford Professor's Guide

Show HN: Agent2Creator – a video social network whose members are AI agents

Show HN: Active Source of Truth for Your Coding Agents

Software Engineering in the Agentic Era

AI has failed to win people's trust. Its makers? less trusted

The Real AI Crash Will Start This Year

NanoGPT Speedrun Frontier

1987 年 Commodore Amiga 500 上的编码代理，配备 7MHz CPU 和 1 MB RAM

A coding agent on a 1987 Commodore Amiga 500 with a 7MHz CPU and 1 MB of RAM

deepseek 工具中的 web_search 工具需要来自 deepseek 的 api 密钥，并且 deepseek 会根据 deepseek-v4-flash 使用情况向您收费。

web_search tool in deepseek harness needs api key from deepseek and deepseek charges you as deepseek-v4-flash usage.

荷兰监管机构对 Uber 罚款 8.25 亿欧元，原因是 Uber 让人工智能停用司机账户

Dutch regulator fines Uber €825M for letting AI deactivate driver accounts

还有人尝试过 KV 缓存混合吗？

Anyone else tried out KV cache blending?

您认为您会获得便宜的 DDR5 RAM 吗？再想一想，即使价格下跌，黄牛机器人现在的数量与购物者的比例为 10 比 1，并将保持高价格

Think you're going to get cheap DDR5 RAM? Think again, even if prices fall, scalper bots now outnumber shoppers 10 to 1 and will keep prices high

Munder Difflin – 代理管理克隆办公室

Munder Difflin – Agent harness to run an office of your clones

人工分析“智能”：毫无意义的基准

Artificial Analysis "Intelligence": A meaningless benchmark

需要支持具有多 GPU 的 llama.cpp

Need support for llama.cpp with multi GPU

AntLing发布了Ling-3.0-flash的dspark草案模型

AntLing released a dspark draft model for Ling-3.0-flash

Freetokens 项目令人印象深刻

Freetokens project is impressive

Llama.cpp 版本 0.2.0 已发布！

Llama.cpp version 0.2.0 is out!

老兄，Qwen Lab 用 Qwen 3.8 27B 煮的，太他妈好吃了

Bro wtf, Qwen Lab cooked with Qwen 3.8 27B, it's so fucking good

NInfer 的夏普模板：-42% 输出令牌，速度相同

Sharp template to NInfer: -42% output tokens, same speed

这就是我在本地运行的原因。

This is why I run locally.

16 GB VRAM 炼狱讨论线程

16 GB VRAM purgatory discussion thread

如何让本地法学硕士/代理人访问“真正的”网络浏览器

How to give a local LLM/agent access to a "real" web browser

Qwen 3.8 27b - PI AGENT 与 OPENCODE - 另一个示例

Qwen 3.8 27b - PI AGENT vs OPENCODE - another smaple

我尝试在 MacBook Air m2 24GB 上使用 Qwen 3.8 27B 3bit Quant 进行 agenic 编码。花了 63 个小时，但令人惊讶的是，飞行模拟器竟然能正常工作。

I tried to do agenic coding with Qwen 3.8 27B 3bit quant on a macbook air m2 24gb. It took 63 hours, but amazingly, the flight simulator worked.

Qwen3.8-27B不同思维层次

Qwen3.8-27B different thinking levels

Qwen 3.8 低档和中档有山羊皮

Qwen 3.8 Low and Medium are goated

Qwen3.8-27B Q6 是代理编码的野兽

Qwen3.8-27B Q6 is a beast at agentic coding

用于编码+一般用途的最佳本地人工智能工具是什么？

What’s the best local AI harness for coding + general use?

FireRedAudio 和 FireRedTTS3 由 FireRedTeam - Huggingface

FireRedAudio & FireRedTTS3 by FireRedTeam - Huggingface

NVIDIA AVO 在 ARC-AGI-3 上获得了 100% 的成功。它在所有 25 个公共环境中完成了所有 183 个关卡，在没有指示、明确规则或既定目标的情况下弄清楚该做什么。

NVIDIA AVO got 100% on ARC-AGI-3. It completed all 183 levels across all 25 public environments, figuring out what to do with no instructions, explicit rules, or stated goals.

DeepSeek Harness v0.1.1 发布

DeepSeek Harness v0.1.1 released

Qwen 3.8 27b 即使在 Q3_xxs 也很强大

Qwen 3.8 27b is strong even at Q3_xxs