每周科技速递 - 2026-06-08

Posted on Edited on Word count in article: 2k Reading time ≈ 7 mins.

每周科技速递 - 2026-06-08

本周科技圈焦点：系统提示词大规模泄露项目引爆 GitHub，突破 4.1 万星标；自进化 Agent 框架 GenericAgent 以 6 倍 Token 效率降低持续高热；64M 参数小模型 minimind 实现 2 小时从零训练；Gary Marcus 尖锐批评 AI slop 侵蚀生产力；比特币逼近 $60,000 关口，机构情绪显著转向谨慎。

🧠 LLM / 大模型

  • 🔥12 | jingyaogong/minimind — 🧠「大模型」2小时完全从0训练64M的小参数LLM！Train a 64M-parameter LLM from scratch in just 2h!
  • 🔥8 | sgl-project/sglang — SGLang is a high-performance serving framework for large language models and multimodal models.
  • 🔥8 | ScrapeGraphAI/Scrapegraph-ai — Python scraper based on AI
  • 🔥8 | yamadashy/repomix — 📦 Repomix is a powerful tool that packs your entire repository into a single, AI-friendly file. Perfect for when you nee
  • 🔥6 | NVIDIA/TensorRT-LLM — TensorRT LLM provides users with an easy-to-use Python API to define Large Language Models (LLMs) and supports state-of-
  • 🔥6 | GreyDGL/PentestGPT — Automated Penetration Testing Agentic Framework Powered by Large Language Models
  • 🔥6 | 0xPlaygrounds/rig — ⚙️🦀 Build modular and scalable LLM Applications in Rust
  • 🔥6 | skyzh/tiny-llm — A course of learning LLM inference serving on Apple Silicon for systems engineers: build a tiny vLLM + Qwen.
  • 🔥5 | ollama v0.30.7-rc1 — What's Changed docs: update Zod example to use native toJSONSchema by @fuleinist in https://github.com/ollama/ollama/pul
  • 🔥5 | bitsandbytes-foundation/bitsandbytes — Accessible large language models via k-bit quantization for PyTorch.
  • 🔥5 | open-compass/opencompass — OpenCompass is an LLM evaluation platform, supporting a wide range of models (Llama3, Mistral, InternLM2,GPT-4,LLaMa2, Q
  • 🔥5 | guardrails-ai/guardrails — Adding guardrails to large language models.
  • 🔥5 | ModelTC/LightLLM — LightLLM is a Python-based LLM (Large Language Model) inference and serving framework, notable for its lightweight desig
  • 🔥5 | thu-pacman/chitu — High-performance inference framework for large language models, focusing on efficiency, flexibility, and availability.
  • 🔥5 | stanford-crfm/helm — Holistic Evaluation of Language Models (HELM) is an open source Python framework created by the Center for Research on F

🤖 AI Agent

  • 🔥13 | lsdefine/GenericAgent — Self-evolving agent: grows skill tree from 3.3K-line seed, achieving full system control with 6x less token consumption
  • 🔥11 | CherryHQ/cherry-studio — AI productivity studio with smart chat, autonomous agents, and 300+ assistants. Unified access to frontier LLMs
  • 🔥11 | EvoMap/evolver — The GEP-powered self-evolving engine for AI agents. Auditable evolution with Genes, Capsules, and Events. | evomap.ai
  • 🔥10 | openclaw v2026.6.5-beta.2 — Highlights - QQBot now strips model reasoning/thinking scaffolding before native delivery, preventing raw con
  • 🔥8 | openclaw v2026.6.1 — Highlights - Agents and CLI-backed runtimes recover more cleanly from interrupted tool calls, stale session bindings, co
  • 🔥7 | Xiangyue-Zhang/auto-deep-researcher-24x7 — 🔥 An autonomous AI agent that runs your deep learning experiments 24/7 while you sleep. Zero-cost monitoring, Leader-Wor
  • 🔥6 | Upsonic/Upsonic — Build autonomous AI agents in Python.
  • 🔥6 | skalesapp/skales — Local-first AI desktop agent for Windows, macOS, Linux & Android. Set a goal, it works on its own. Teams (pair two deskt
  • 🔥5 | Build tools, to build more —
  • 🔥5 | Opus 4.8 —
  • 🔥5 | stakpak/agent — Ship your code, on autopilot. An open source agent that lives on your machines 24/7 and keeps your apps running. 🦀
  • 🔥5 | grinev/opencode-telegram-bot — OpenCode mobile client via Telegram: run and monitor AI coding tasks from your phone while everything runs locally on yo
  • 🔥5 | BlockRunAI/Franklin — The AI agent with a wallet — spends USDC autonomously to get real work done. Apache-2.0, TypeScript.
  • 🔥5 | gmickel/flow-next — Spec-driven AI workflow plugin for Claude Code, OpenAI Codex, and Factory Droid. Zero-dep task tracking, worker subagent
  • 🔥5 | swarmclawai/swarmclaw — Open-source self-hosted AI agent runtime and multi-agent framework for autonomous agent swarms. Agent memory, MCP tools,

💰 加密货币

  • 🔥7 | A quick review of the Ways and Means tax bills: State of Crypto —
  • 🔥7 | Michael Saylor revives bitcoin-buy speculation as scrutiny over Strategy grows —
  • 🔥7 | Bitcoin near $60,000 today vs February: Institutional sentiment has flipped —
  • 🔥7 | foundry nightly-43ed279745c5decd51dfc60eaaba4b0911a30e9c — Other - fix: redact private keys from Debug (#15091) by @stevencartavia - refactor(script): isolate Tempo session signer
  • 🔥5 | OffcierCia/DeFi-Developer-Road-Map — DeFi Developer roadmap is a curated Developer handbook which includes a list of the best tools for DApps development, re
  • 🔥5 | lifinance/contracts — LI.FI Smart Contracts
  • 🔥5 | exactly/protocol — exactly is a decentralized, non-custodial and autonomous marketplace for depositing and borrowing crypto assets at a var

🔬 前沿科技

  • 🔥15 | asgeirtj/system_prompts_leaks — Extracted system prompts from Anthropic - Claude Code, Claude Design, Opus 4.8, Sonnet 4.6. OpenAI - ChatGPT 5.5 Thinkin
  • 🔥14 | rasbt/LLMs-from-scratch — Implement a ChatGPT-like LLM in PyTorch from scratch, step by step
  • 🔥10 | tensorflow/tensorflow — An Open Source Machine Learning Framework for Everyone
  • 🔥10 | huggingface/transformers — 🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and
  • 🔥9 | ultralytics/ultralytics — Ultralytics YOLO 🚀
  • 🔥8 | roboflow/supervision — We write your reusable computer vision tools. 💜
  • 🔥7 | Firefox Merges Support for Vulkan Video Decoding —
  • 🔥7 | VibeOS: First ever AI-native operating system —
  • 🔥7 | Show HN: Nightwatch, The open-source, read-only AI SRE —
  • 🔥7 | How's Linear so fast? A technical breakdown —
  • 🔥7 | If LLMs Have Human-Like Attributes, Then So Does Age of Empires II —

📢 KOL Updates

（本周无 Twitter/X KOL 数据更新）

📊 Weekly Trend Summary

本周宏观趋势：AI 基础设施与开发工具持续占据 GitHub 趋势榜主导地位，系统提示词泄露、从零训练 LLM、自进化 Agent 等开源项目展现社区对 AI 透明性与可审计性的强烈需求。与此同时，Gary Marcus 等批评声音对 AI 炒作周期提出警示，机构投资者对加密市场的态度从年初狂热转向谨慎，标志着一个更加理性的技术周期可能正在形成。

📦 GitHub Releases

  • picoclaw nightly — Nightly build for v0.2.9-nightly.20260607.7d2b0c2a This is an automated build and may be unstable. Use with caution. Ful https://github.com/sipeed/picoclaw/releases/tag/nightly
  • openclaw v2026.6.5-beta.2 — Highlights - QQBot now strips model reasoning/thinking scaffolding before native delivery, preventing raw con https://github.com/openclaw/openclaw/releases/tag/v2026.6.5-beta.2
  • openclaw v2026.6.1 — Highlights - Agents and CLI-backed runtimes recover more cleanly from interrupted tool calls, stale session bindings, co https://github.com/openclaw/openclaw/releases/tag/v2026.6.1
  • openclaw v2026.6.2-beta.1 — 2026.6.2 Highlights - Plugin and skill installs now use an operator install policy instead of the old dangerous-code sca https://github.com/openclaw/openclaw/releases/tag/v2026.6.2-beta.1
  • openclaw v2026.5.31-beta.4 — 2026.5.31 Highlights - Agents and CLI-backed runtimes recover more cleanly from interrupted tool calls, stale session bi https://github.com/openclaw/openclaw/releases/tag/v2026.5.31-beta.4
  • ollama v0.30.7-rc1 — What's Changed docs: update Zod example to use native toJSONSchema by @fuleinist in https://github.com/ollama/ollama/pul https://github.com/ollama/ollama/releases/tag/v0.30.7-rc1
  • ollama v0.30.6 — New models - Gemma 4 QAT weights: the Gemma 4 family is now optimized with Quantization-Aware Training (QAT) to dramatic https://github.com/ollama/ollama/releases/tag/v0.30.6
  • ollama v0.30.5 — What's Changed Fixed the gemma4:12b floating point exception crash on x86, CUDA, Linux, and Windows systems. ollama laun https://github.com/ollama/ollama/releases/tag/v0.30.5
  • ollama v0.30.4 — New models - Nemotron-3-Ultra: NVIDIA Nemotron 3 Ultra is built for high-throughput reasoning and long-running agent wor https://github.com/ollama/ollama/releases/tag/v0.30.4
  • ollama v0.30.3 — New models - Gemma 4 12B: high-performance multimodal intelligence that runs directly on laptops, combining efficiency w https://github.com/ollama/ollama/releases/tag/v0.30.3
  • ollama v0.30.2 — What's Changed ollama launch now supports Qwen Code and can guide users through installing the Cline CLI when it is miss https://github.com/ollama/ollama/releases/tag/v0.30.2
  • vllm v0.22.1 — Highlights This release features 8 commits from 6 contributors (1 new)! v0.22.1 is a patch release on top of v0.22.0 wit https://github.com/vllm-project/vllm/releases/tag/v0.22.1
  • foundry nightly-43ed279745c5decd51dfc60eaaba4b0911a30e9c — Other - fix: redact private keys from Debug (#15091) by @stevencartavia - refactor(script): isolate Tempo session signer https://github.com/foundry-rs/foundry/releases/tag/nightly-43ed279745c5decd51dfc60eaaba4b0911a30e9c
  • zeroclaw v0.8.0-beta-2 — ZeroClaw v0.8.0-beta-2 This is the second beta of the v0.8.0 line, and the largest release since v0.7.5. Its headline is https://github.com/zeroclaw-labs/zeroclaw/releases/tag/v0.8.0-beta-2
  • asgeirtj/system_prompts_leaks ⭐ 41.4K (+103/天) | JavaScript — Extracted system prompts from Anthropic - Claude Code, Claude Design, Opus 4.8, Sonnet 4.6. OpenAI - https://github.com/asgeirtj/system_prompts_leaks
  • rasbt/LLMs-from-scratch ⭐ 96.8K (+92/天) | Jupyter Notebook — Implement a ChatGPT-like LLM in PyTorch from scratch, step by step https://github.com/rasbt/LLMs-from-scratch
  • lsdefine/GenericAgent ⭐ 12.7K (+89/天) | Python — Self-evolving agent: grows skill tree from 3.3K-line seed, achieving full system control with 6x les https://github.com/lsdefine/GenericAgent
  • jingyaogong/minimind ⭐ 51.3K (+75/天) | Python — 🧠「大模型」2小时完全从0训练64M的小参数LLM！Train a 64M-parameter LLM from scratch in just 2h! https://github.com/jingyaogong/minimind
  • EvoMap/evolver ⭐ 8.2K (+65/天) | JavaScript — The GEP-powered self-evolving engine for AI agents. Auditable evolution with Genes, Capsules, and Ev https://github.com/EvoMap/evolver
  • CherryHQ/cherry-studio ⭐ 47.0K (+63/天) | TypeScript — AI productivity studio with smart chat, autonomous agents, and 300+ assistants. Unified access to fr https://github.com/CherryHQ/cherry-studio
  • huggingface/transformers ⭐ 161.4K (+58/天) | Python — 🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, https://github.com/huggingface/transformers
  • tensorflow/tensorflow ⭐ 195.6K (+51/天) | C++ — An Open Source Machine Learning Framework for Everyone https://github.com/tensorflow/tensorflow

📝 Blog Picks

  • Slop, productivity, and why the AI-fueled world is going nowhere mighty fast — Gary Marcus | Gary Marcus 对 AI 生成内容泛滥（slop）的尖锐批评，指出当前 AI 炒作与实际生产力提升之间存在巨大鸿沟，呼吁行业回归理性 https://garymarcus.substack.com/p/slop-productivity-and-why-the-ai
  • How Endava is redesigning software delivery around AI agents — OpenAI Blog | OpenAI 案例研究：Endava 如何围绕 AI Agent 重新设计软件交付流程，探索企业级 AI 自动化的落地路径 https://openai.com/index/endava-frontiers
  • Dreaming: Better memory for a more helpful ChatGPT — OpenAI Blog | OpenAI 发布 ChatGPT 记忆功能升级，通过梦境式学习机制提升长期对话连贯性与个性化体验 https://openai.com/index/chatgpt-memory-dreaming
  • Build tools, to build more — Ben's Bites | Ben's Bites 分析 AI 工具链建设趋势：更好的工具是构建更多创新应用的基础，而非替代品 https://www.bensbites.com/p/build-tools-to-build-more

📊 Data Sources: RSS 284 | Twitter 0 | Reddit 0 | Web 0 | GitHub 25 releases + 48 trending | Dedup: 208 articles 🤖 Generated by tech-news-digest v3.14.0 | https://github.com/draco-agent/tech-news-digest | Powered by OpenClaw