Microsoft 与 Hugging Face 用数据库终态奖励后训练 ThinkingBox 智能体
Microsoft 与 Hugging Face 将 ThinkingBox 的数据库终态检查转化为训练奖励,并对 Qwen3.8-27B 进行 50 步 GRPO 微调。
推荐理由:文章把数据库终态检查转化为训练奖励,公开完整训练流程,并同时量化重复成功率、覆盖率与成本变化。
When Terminal-Agent Training Stalls: Demystifying Data Generation and Verification Challenge
Xi Qin · Isabel Kurth · Xin Cui · Elin Park · Alexander Schaefer · Yaad Oren
使用前沿模型如Claude Opus作为元智能体生成终端任务和验证器用于强化学习训练日益普及。然而,可运行的Docker镜像和可执行测试套件并不能保证终端智能体训练的忠实端到端管道。针对这一差距,提出一个元智能体管道,诊断三类失败:基准无效性、测试工具脆弱性和奖励错位。提示重构和上下文扩展将基线可解性提高5.6倍,但9B模型在Claude Opus生成的任务上于20步内平均pass@2达到81.3%后饱和。添加困难任务在未改变训练配置的情况下将平均pass@2降至20.6%,有力证明可解性带是模型特定的。这些发现表明,元智能体可靠性需要将可解性带校准、验证器审计和基础设施错误计作为一流评估标准,而非事后诊断。
尚无解读。
arXiv:2610.02405v1 Announce Type: new Abstract: Using a frontier model like Claude Opus as a meta-agent to generate terminal tasks and verifiers for RL training is increasingly common. Yet a runnable Docker image and executable test suite do not guarantee a faithful end-to-end pipeline for terminal agent training. We present a meta-agent pipeline motivated by this gap, diagnosing three classes of failure: benchmark invalidity, harness brittleness, and reward misalignment. Prompt redesign and context extension raise baseline solvability 5.6 times, but a 9B model saturates at 81.3% mean pass@2 within 20 steps on Claude Opus-generated tasks. Adding hard tasks reduces mean pass@2 to 20.6% without changing the training configuration, a strong evidence that the solvability band is model-specific. These findings demonstrate that meta-agent reliability requires solvability-band calibration, verifier audits, and infrastructure error accounting as first-class evaluation criteria, not post-hoc diagnost.
Microsoft 与 Hugging Face 将 ThinkingBox 的数据库终态检查转化为训练奖励,并对 Qwen3.8-27B 进行 50 步 GRPO 微调。
推荐理由:文章把数据库终态检查转化为训练奖励,公开完整训练流程,并同时量化重复成功率、覆盖率与成本变化。
1. 角色卡系統:人設、語氣、禁忌、知識邊界用JSON存好,一鍵套用 2. 記憶分層:短期對話 + 長期人設記憶 + 情境快取,延遲才會低 3. 快速啟動模板:像「導師/伴侶/反派/NPC」這種預設,3秒就能開演
You are [Chatbot Name], the AI [Role of Chatbot] at [Company Name]. You're helpful, professional, clever, and friendly, and your main job is to assist users with anything related to [Company Name]. You’re an expert in all things [Company Na…
You are [Company Name]'s assistant. When a returning visitor starts a chat: Opening message: "Welcome back! I'm here to help. Were you looking for more info on [previous topic], or is there something new I can assist with?" Rules: - Acknowl…
这是我需要完成的任务" + 任务内容 → System 回复"收到任务" 2. User: "这是检索到的相关资源" + EXA 聚合文本 → System 回复"收到相关资源" 3. User: 提示词框架要求 + 生成指令
“If the query is about data analysis, pass it to the Data Expert. If it’s about frontend UI, pass it to the Design Bot.”
Act as an AI workflow consultant for creators and professionals. Identify repetitive tasks in my life and help me automate them using ChatGPT, AI tools, templates, and smart ...