AGI前沿 on AI 实战派 · 从技术到赚钱https://guijiagi.com/categories/agi%E5%89%8D%E6%B2%BF/Recent content in AGI前沿 on AI 实战派 · 从技术到赚钱Hugozh-cn本站内容采用 CC BY-NC-SA 4.0 国际许可协议授权Wed, 28 Oct 2026 06:49:00 +0800智能体的记忆架构：从每会话重开到终身学习还差什么https://guijiagi.com/posts/2026-10-28-08-agent-long-term-memory-architecture/Wed, 28 Oct 2026 06:49:00 +0800https://guijiagi.com/posts/2026-10-28-08-agent-long-term-memory-architecture/当前Agent每次对话几乎都是失忆重来，解读长期记忆、经验沉淀与终身学习架构的技术现状与门槛。文本、图像、机器人进同一架构：多模态统一走到哪一步了https://guijiagi.com/posts/2026-10-28-07-multimodal-unified-transformer/Wed, 28 Oct 2026 06:42:00 +0800https://guijiagi.com/posts/2026-10-28-07-multimodal-unified-transformer/从视觉语言模型到世界模型再到机器人控制，不同模态正在被塞进同一个Transformer，解读统一架构的进展与瓶颈。当算力翻倍换不来能力翻倍：Scaling Law的拐点来了吗https://guijiagi.com/posts/2026-10-28-06-scaling-law-knee-point/Wed, 28 Oct 2026 06:35:00 +0800https://guijiagi.com/posts/2026-10-28-06-scaling-law-knee-point/近一两年前沿模型的能力提升曲线出现放缓迹象，解读scaling law是否遭遇拐点，以及逼近AGI还能靠什么。OpenAI停止用实时互联网训练：合成数据与自愿安全协议成新底座https://guijiagi.com/posts/2026-10-28-05-openai-live-internet-training-stop/Wed, 28 Oct 2026 06:28:00 +0800https://guijiagi.com/posts/2026-10-28-05-openai-live-internet-training-stop/OpenAI宣布停止在实时互联网上训练模型，同期白宫推动自愿超级智能安全协议，解读数据来源与安全治理双双转向。OpenAI宣布模型在数学物理取得三项新成果：从解题到立理https://guijiagi.com/posts/2026-10-28-04-openai-model-proves-new-theorems/Wed, 28 Oct 2026 06:21:00 +0800https://guijiagi.com/posts/2026-10-28-04-openai-model-proves-new-theorems/OpenAI披露其内部模型在计算理论与量子场论取得三项新结果，含指数时间假说被证伪等，解读AI从辅助解题走向独立产出新知识。