AITOPAI热报
PapersMulti-sourceSelected73°

Thousand-Graph Hypothesis for Code Reasoning

The Thousand-Graph Hypothesis: A Testable Hypothesis of Task-Conditioned Relation Materialization in Repository-Level Code Reasoning

arXiv: DeepSeek@Fei Ding
Paper· 5 sources
Read the original
Why we picked it

This method improves code reasoning accuracy in large repositories without expensive pre-training or maintenance burden.

The Thousand-Graph Hypothesis proposes an entity-only external interface with task-conditioned relation materialization for repository-level code reasoning. The approach was evaluated on DeepSeek-V4-Flash and SWE-bench Verified benchmarks. Base, one-layer, and two-layer conditions achieved 92.1%, 94.2%, and 95.6% success respectively without pre-built entity-relation edges.

Original · arXiv: DeepSeek

The Thousand-Graph Hypothesis: A Testable Hypothesis of Task-Conditioned Relation Materialization in Repository-Level Code Reasoning

Large software repositories are often beyond model context limits. Training repository knowledge into models is costly and quickly stale, while local retrieval can miss scattered requirements, and explicit relation graphs add ongoing maintenance burden. We propose an entity-only external interface with task-conditioned relation materialization during inference. A two-layer index separates global routing from local entity focus and is evaluated on DeepSeek-V4-Flash and SWE-bench Verified. The base, one-layer, and two-layer conditions achieve 92.1%, 94.2%, and 95.6% success, respectively, under zero pre-built entity-relation edges.

#DeepSeek-V4-Flash#SWE-bench#code reasoning#repository-level#RAG

Keep reading

Recent stories with the same tag or section
PapersBuildersMulti-source13:00
Chinese content研究拆解 LLM 网络威胁情报实体抽取失败模式：主要问题不是幻觉

DeepSeek-V4-Flash 和 GPT-6.1-Sol 抽实体都不靠谱，但坏的原因不是幻觉，而是把背景信息当成攻击实体。用置信度分层排审核优先级，别当真值。

#DeepSeek-V4-Flash#GLM-5.3-Flash#GPT-6.1-Sol#网络威胁情报
Event
arXiv: DeepSeek@Jiahong Qian5 sourcesSource
PapersBuilders12:44
Chinese contentarXiv 论文：证伪驱动的模型无关智能体协议，输出纪律决定决策下限

用中医辨证基准证明了件事：很多模型翻车不是能力不够，是输出纪律不行。他们设计的证伪协议能把 Kimi K3 的方差从 0.218 压到 0.007，写智能体的值得看看。

#TCMEval-SDT#DeepSeek-V4-Flash#Kimi K3#智能体
Event
arXiv: DeepSeek@Jing Liu et al. (4)2 sourcesSource
PapersUS vs ChinaSelected04:29
Tsinghua, Oxford, Stanford Paper Finds LLMs Reason Even When Told Not To

The finding that a missing think tag or short reply does not mean a model skipped reasoning undermines a common evaluation shortcut. Teams benchmarking reasoning models or building answer-only constraints should re-check how compliance is measured, since suppressing visible reasoning costs roughly 15 accuracy points. The result spans 5 models, making it a broad behavioral pattern rather than a single-vendor quirk.

#DeepSeek#DeepSeek-V4-Flash#reasoning models#AI safety
Event
rohanpaul_ai@rohanpaul_ai2 sourcesSource
PapersBuildersMulti-source10:33
Chinese content测量优先的智能体排行榜审计框架：污染风险与评分器验证

一群研究者专门审计智能体排行榜的基准污染问题，用 GPT-4.1 和 DeepSeek-V4-Flash 在 SWE-bench 上做了对照实验，结论是很多污染指控证据不足，做评测的人值得看看这套框架。

#SWE-bench Verified#HAL#GPT-4.1#DeepSeek-V4-Flash
Event
arXiv: DeepSeek@Dishu Yang et al. (4)3 sourcesSource