Author

Data Party THU

Official platform of Tsinghua Big Data Research Center, sharing the team's latest research, teaching updates, and big data news.

546

Articles

Likes

3.5k

Views

Comments

Latest from Data Party THU

100 recent articles max

Data Party THU

Oct 27, 2026 · Industry Insights

Tsinghua's 2026 Hotline Reports: Global Cities & China's 12345 Shift to Governance Hubs

Tsinghua University releases two 2026 reports evaluating 20 global cities' hotlines and 100 Chinese cities' 12345 hotlines, revealing four development models, a shift from service channels to governance hubs, and the critical role of organizational coordination over technology alone in urban governance effectiveness.

12345 hotlineCambridge UniversityTsinghua University

0 likes · 12 min read

Tsinghua's 2026 Hotline Reports: Global Cities & China's 12345 Shift to Governance Hubs

Data Party THU

Oct 26, 2026 · Artificial Intelligence

LabBuilder: Protocol-Driven 3D Lab Layout Generation from Text Descriptions (ICML 2026)

LabBuilder, accepted at ICML 2026, automatically generates geometrically valid, chemically safe, and robot-navigable 3D laboratory layouts from free-text experiment descriptions by integrating protocol understanding, chemical safety constraints, and navigation-aware optimization into a unified generation pipeline.

3D scene generationICML 2026LLM applications

0 likes · 12 min read

LabBuilder: Protocol-Driven 3D Lab Layout Generation from Text Descriptions (ICML 2026)

Data Party THU

Oct 25, 2026 · Artificial Intelligence

Harness-of-Harness: Continuous Development & Verification for Long-Cycle AI Coding Agents

The article introduces Harness-of-Harness (HoH), a protocol organizing AI coding agents into iterative planning, development, and independent QA cycles with persistent artifact and evidence states, showing 52.25% average improvement over vanilla agents across three benchmarks and demonstrating a 70-round FPS game development case study.

AI coding agentsHarness-of-HarnessMulti-Agent Systems

0 likes · 17 min read

Harness-of-Harness: Continuous Development & Verification for Long-Cycle AI Coding Agents

Data Party THU

Oct 24, 2026 · Artificial Intelligence

Do We Really Know What Biomedical Foundation Models Have Learned? Benchmarking Challenges Revealed

A Nature Methods paper argues current leaderboard-style benchmarks fail to capture true generalization of biomedical foundation models, advocating perturbation-based tests, ablation studies, and a community-driven ecosystem (BEACON) to evaluate mechanistic understanding beyond single-task scores.

BEACON consortiumNature Methodsbenchmarking

0 likes · 7 min read

Do We Really Know What Biomedical Foundation Models Have Learned? Benchmarking Challenges Revealed

Data Party THU

Oct 23, 2026 · Artificial Intelligence

Are We Over-Testing LLMs? Survey of Benchmark Compression Methods, Costs & Reliability

This survey reviews 119 studies on benchmark compression for efficient LLM evaluation, categorizing methods into static compression, adaptive evaluation, and benchmark analysis, while emphasizing that reliable compression requires defining evaluation goals, verifying coverage, and accounting for full lifecycle costs rather than simply reducing test items.

LLM evaluationLarge Language Modelsadaptive evaluation

0 likes · 34 min read

Are We Over-Testing LLMs? Survey of Benchmark Compression Methods, Costs & Reliability

Data Party THU

Oct 22, 2026 · Artificial Intelligence

3D Vision-Language Models: From Geometry to Open-Vocabulary & Embodied Interaction

This tutorial surveys 3D vision-language models, covering geometric representations (point clouds, meshes, Gaussian splatting), CLIP-style cross-modal alignment, taxonomy of embedding-based vs. generative architectures, and emerging applications in recognition, scene understanding, 3D generation, and embodied AI, highlighting the distinction between similarity matching and language generation capabilities.

3D Gaussian splatting3D vision-language modelsCLIP

0 likes · 29 min read

3D Vision-Language Models: From Geometry to Open-Vocabulary & Embodied Interaction

Data Party THU

Oct 21, 2026 · Artificial Intelligence

AI Predicts Borylation Sites on Drug Molecules: ETH & Roche's Active Learning Loop

ETH Zurich and Roche researchers combined active learning with geometric graph neural networks to predict C–H borylation sites on drug-like molecules, iteratively selecting 50 new substrates from 22,253 candidates and achieving MCC 0.43–0.50 on scaffold splits, outperforming XGBoost.

C–H borylationactive learningdrug discovery

0 likes · 8 min read

AI Predicts Borylation Sites on Drug Molecules: ETH & Roche's Active Learning Loop

Data Party THU

Oct 20, 2026 · Artificial Intelligence

Why Reasoning Models Hallucinate: A Lifecycle Framework for Diagnosis, Evidence, and Release Control

This survey introduces the UIPCA diagnostic framework to dissect reasoning hallucinations across five dimensions — unsupported premises, invalid inferences, dependency reuse, visible consistency, and action-policy failures — and maps detection signals to concrete interventions while exposing gaps in causal evidence for reasoning–reliability trade-offs.

UIPCA frameworkcausal inferenceevidence-based AI

0 likes · 39 min read

Why Reasoning Models Hallucinate: A Lifecycle Framework for Diagnosis, Evidence, and Release Control

Data Party THU

Oct 19, 2026 · Artificial Intelligence

OpenAI's Navier-Stokes Proof: Singularity, Scale Limits, and AI's Role in Science

OpenAI's AI-generated proof of finite-time blowup for the 3D Navier-Stokes equations reveals a mathematical singularity at molecular scales, highlighting the continuum model's limits and accelerating AI's integration into multi-scale fluid modeling where neural operators bridge scales while humans interpret physical meaning.

AI in scienceDeepMindMillennium Prize

0 likes · 7 min read

OpenAI's Navier-Stokes Proof: Singularity, Scale Limits, and AI's Role in Science

Data Party THU

Oct 18, 2026 · Artificial Intelligence

Who Improves the Robot? A Framework for Physical Recursive Self-Improvement

This paper introduces a six-level framework (L0-L5) for physical recursive self-improvement in robotics, distinguishing capability updates from mechanism updates, and finds that current literature has not yet demonstrated true recursive self-improvement with independent verification across tasks.

L0-L5 levelscapability updateindependent evaluation

0 likes · 42 min read

Who Improves the Robot? A Framework for Physical Recursive Self-Improvement