Official platform of Tsinghua Big Data Research Center, sharing the team's latest research, teaching updates, and big data news.
Stanford, Berkeley and Nvidia researchers introduce LLM-as-a-Verifier, a universal verification framework that enhances agent performance, safety and stability on long‑horizon tasks, and outperforms Claude Mythos and GPT‑5.5 on the Terminal‑Bench and SWE‑Bench benchmarks.
Graphify is a Python tool and Claude Code skill that creates a persistent, queryable knowledge graph of code, documentation, and media, cutting token usage by up to 71.5× compared with raw file reads, and it does so through a three‑pass pipeline that combines deterministic AST extraction, optional local audio transcription, and AI‑driven semantic extraction.
The ICLR 2026 analysis shows Chinese institutions, led by Tsinghua with 331 papers, dominate total submissions, while U.S. institutions retain the lead in high‑impact Oral papers, highlighting a shift in volume but not in top‑tier research influence.
ProteinOPD introduces a multi‑teacher, on‑policy preference‑distillation framework that aligns protein language models with multiple design objectives—foldability, solubility and thermostability—while preserving generation quality, achieving up to 54% stability gains and an eight‑fold training speedup.
The article surveys recent research on Agent Harness engineering, showing that real‑world agent instability stems from system‑level factors beyond model capability, introduces the seven‑layer ETCLOVG architecture, presents benchmark gains from harness tweaks, maps open‑source projects to the framework, and outlines five key open research directions.
MedScope proposes a "Think with Videos" paradigm that lets AI models actively locate and verify evidence in long clinical videos, using coarse‑to‑fine tool calling, evidence‑centric training data (ClinVideoSuite) and a grounding‑aware reinforcement learning objective, achieving superior performance on multiple video‑understanding benchmarks.
The article examines how AI is silently reshaping everyday life by enabling one‑person companies, redefining job roles, disrupting traditional education, and driving a shift toward networked, AI‑augmented organizations, supported by concrete industry examples and recent policy moves.
Anthropic's new paper introduces lightweight LoRA‑based introspection adapters that let large language models translate their internal activations into natural‑language reports of learned behaviors, achieving a 59% success rate on the AuditBench benchmark and exposing previously undetectable encrypted fine‑tuning attacks.
This article walks through the entire feature‑engineering pipeline—data cleaning, missing‑value imputation, encoding, outlier handling, scaling, feature construction, and selection—using Pandas and Scikit‑learn, and shows how to wrap the steps into a reproducible Scikit‑learn Pipeline.
Anthropic’s Code w/ Claude developer conference revealed three major upgrades—a stronger foundation model, the Claude Platform’s multi‑agent orchestration, and the Claude Code desktop client—showcasing real‑world cases where 50 k lines of Scala were rewritten in four days and a 20‑day approval process was halved, while API usage jumped 17‑fold and weekly developer time on Claude rose to 20 hours.
