MNIST-PRO: MNIST is Back as a Partially Observable World for AI Agents
- 类型：arxiv
- 标识：2608.31022
- 链接：http://arxiv.org/abs/2608.31022v1
- 主分类：agent
- 形态：benchmark
- 被引：0
- 被引来源：Semantic Scholar + OpenAlex
- S2被引：0
- OpenAlex被引：0
- 影响力被引：0
- TLDR：MnIST-PRO is addressed, a benchmark that isolates agentic perception by converting MNIST digit recognition into a sequential, glimpse-based search task with lookback constraints, and ten multimodal models are evaluated, showing that simply acquiring visual evidence is not enough and agents must also be able to build and update a reliable perceptual state.
- OpenAlex ID：W7204885105
- OpenAlex DOI：10.48550/arxiv.2608.31022
- DOI：10.48550/arxiv.2608.31022
- DOI来源：OpenAlex
- 开放获取：green
- 开放获取链接：https://arxiv.org/pdf/2608.31022
- OpenAlex更新：2026-09-04
- 副分类：evaluation
- 待LLM分类：否
- 标题中文：MNIST-PRO：MNIST 作为部分可观察世界回归，用于 AI Agents
- TLDR中文：提出 MnIST-PRO 基准，将 Agent 感知能力隔离测试：把 MNIST 数字识别转换为带回看约束的序列式 glimpse 搜索任务，并评估了十个多模态模型，结果表明仅获取视觉证据是不够的，Agent 还必须能够构建并更新可靠的感知状态。
- 来源文件：
- /inbox/tom/_candidates/2026-09-01-agent-rag-longcontext-candidates.json
- /inbox/tom/_candidates/2026-09-02-agent-rag-longcontext-candidates.json
- [S2 enrich]
- [OpenAlex backfill]