一项新的基准测试 ARC-AGI-3 揭示了 GPT-5.5 和 Opus 4.7 等先进 AI 模型存在严重的推理错误。这些模型在该基准测试上的成功率仅为 0.8%，凸显了在抽象推理能力方面持续存在的差距。研究结果表明，尽管技术取得了进步，但当前的 AI 系统在基本的人类水平任务方面仍面临困难。 AI
影响 揭示了前沿模型中持续存在的推理差距，表明当前架构可能无法扩展到人类水平的抽象思维。
排序理由 该集群报告了对现有 AI 模型的新基准测试评估，属于研究范畴。
在 Mastodon — mastodon.social 阅读 →
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
The generalizable LLM failure mode isn't "can't reason". It's that outcome reward cements whatever theory was active when a level happened to clear. ARC Prize's analysis of GPT-5.5 and Opus 4.7 on ARC-AGI-3 (0.43%/0.18%) names this alongside two cousins. Self-improvement loops th…
📰 Systematic Reasoning Errors in GPT-5.5 and Opus 4.7: ARC-AGI-3 Reveals 0.8% Success Rate in 2026 The ARC-AGI-3 benchmark exposes three systematic reasoning errors in GPT-5.5 and Opus 4.7, revealing why even the most advanced AI models fail basic human-level tasks. These flaws h…
📰 Yapay Zeka Modelleri 2026'da Neden 3 Sistemsel Hata Yapıyor? GPT-4 ve Gemini 1.5 ARC-AGI-3 Testin... Yeni nesil yapay zeka modelleri bile üç temel akıl yürütme hatası yapıyor. ARC-AGI-3 testi, bu hataların teknolojik ilerlemenin ardında gizli bir zayıflık olduğunu gösteriyor...…