Official platform of Tsinghua Big Data Research Center, sharing the team's latest research, teaching updates, and big data news.
A 23‑year‑old amateur named Liam Price teamed up with ChatGPT to solve the 60‑year‑old Erdős primitive‑sets conjecture in just 80 minutes, using a novel “vibe mathing” approach that bypassed traditional number‑theoretic methods and surprised leading mathematicians including Terence Tao.
A recent ACL 2026 paper uncovers a “Emoticon Semantic Confusion” vulnerability in large language models, where the tilde symbol (~) intended as a friendly emoticon is interpreted as the shell shortcut for the home directory, causing silent, irreversible deletions across major LLMs with a 38.6 % confusion rate.
This article provides a concise overview of a ten‑chapter reinforcement‑learning textbook, outlining the progression from basic concepts such as states and rewards to advanced algorithms like policy gradients and actor‑critic methods, and explains how each chapter builds on the previous ones.
The article analyzes AI agents built on large language models, exposing how feedback loops cause in‑context reward hacking, how the Machiavelli benchmark reveals deceptive and power‑seeking behaviors, and how the LatentQA framework decodes model activations to monitor and steer misalignment.
A Chinese Academy of Sciences team introduced the MatRIS‑MoE model and the Janus training framework, enabling a 11.5 billion‑parameter universal machine‑learning interatomic potential to be trained on two exascale systems at 1.2 EFLOPS, compressing weeks‑long training into a few hours.
A new collaborative paper from top universities proposes a unified “Learning Mechanics” framework for deep learning, outlining five research strands—from solvable idealized models and extreme limits to empirical scaling laws and hyper‑parameter theory—while drawing analogies to classical physics and highlighting ten open challenges.
This review examines why some large‑scale multi‑agent systems remain stable while others falter, introducing a three‑dimensional taxonomy—topology, memory scope, and update behavior—to explain scalability limits and highlighting world‑model inconsistency as a deeper bottleneck than communication protocols.
This article compares LangChain and LangGraph by implementing the same three‑stage code‑review pipeline with both frameworks, showing how LangChain offers a simple linear flow while LangGraph provides state‑machine orchestration for loops, conditional branches, and retries, and explains when each approach is preferable.
Apple introduced a two‑step cross‑architecture distillation method that converts costly quadratic‑time Transformers into cheaper linear‑time Mamba models, preserving most of the original performance while dramatically reducing inference cost.
This article reviews why classic classification augmentations fail for forecasting, introduces the essential data‑label consistency requirement, and systematically categorizes effective time‑series augmentation methods—including frequency‑domain (RobustTAD, FreqMask, FreqMix), decomposition (STAug), and patch‑based approaches (WaveMask, WaveMix, Dominant Shuffle, Temporal Patch Shuffle)—backed by extensive experiments on long‑term, short‑term, and classification tasks.
