Skip to content
AI Price Spy
AI Price Spy
Advertise

Choose your AI. Find your way.

Track resets, compare AI prices, find useful tools and learn from real projects.

RESETS & LIMITS

Know when you can use AI again.

Check Codex reset events and product limits, or open your reset radar.

Events include sources and check times. Radar contains snapshots and records saved on this device.

Find good articles and people

Explore original posts and creators by topic.

View all

The latest in AI

Official feed data, with original sources and clear publication times.

All AI news
OpenAI:Alignment 研究博客(RSS)

OpenAI 训练并部署激活分类器作为前沿模型安全监控层

OpenAI 自 GPT-5.6 起在其最强模型的内部和外部流量上部署激活分类器(probes),利用模型内部激活逐 token 实时评分,可在生成中途阻断违规回复和工具调用并交由推理模型复审。实验显示 probes 比更大的文本分类器对测试的越狱攻击更鲁棒,准确率随模型能力提升,且在后训练各检查点间迁移良好,但需为新基础模型重新训练,且只能检测可显式训练的已知违规类型。

公众号:腾讯混元

腾讯混元发布 Hyra 2.0:让通用 agent 研发可复用的科学工具与模型

腾讯混元发布 Hyra 2.0,将自动化科研重点从直接解题转向由 agent 研发和升级科学模型、工作流与软件工具。系统采用 Overall、Mission、Worker 三层分工,接入企业微信、计算平台和代码仓库的实际研发环境,成果已公开在 GitHub 仓库。

公众号:数字生命卡兹克

数字生命卡兹克分享折腾一年的Agent自动化系统架构

自媒体作者卡兹克分享自己搭建近一年的Agent自动化系统,核心是把任务起点从人身上移开,由事件触发Agent工作。系统分为自动化中台、Agent执行、CI中台和飞书决策四层:中台用rules.json统一管理事件触发和Agent调度,本地MacMini跑Agent任务,自建CI中台完成审查、检查、合并、发布、核实五步,只在删数据、花钱、改方向等关键决策时通过飞书卡片找作者拍板。

See what you can do here

A product walkthrough will be added after the design is complete.

Put AI to work, together

Share experiences and discover new tools and ideas with the community.

AI subscription shop