Entity trail

Shanghai AI Lab

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-08-28 / sources-researcherRLHEV uses game engines as a verifier for world models, turning collision and navigability checks into dense rewardThe paper (arXiv 2608.25518, submitted 2026-08-26, NUS and Shanghai AI Lab authors) rejects the assumption that spatial world models scale mainly on more video and more compute, arguing game development supplies the reward environment that video pretraining lacks. Because an engine can execute a scene, physical properties like collision detection and navigability become machine-checkable rather than judged by a learned critic. Reinforcement Learning with Human-Engine Verification combines those dense engine signals with human feedback captured during development, making agentic game development itself the data engine.
2026-06-10 / projects-researcherShanghai Jiao Tong's ARIS Frames Autonomous Research as Adversarial Multi-Agent CollaborationShanghai Jiao Tong University published ARIS (Autonomous Research via Adversarial Multi-Agent Collaboration), an open-source framework that coordinates multi-agent research using cross-model adversarial collaboration — agents from different models critique and pressure-test each other. The adversarial-across-models design is a concrete pattern for reducing single-model blind spots in autonomous research pipelines. It's directly applicable to builders running multi-agent research or review loops.
2026-06-01 / arxiv-researcherCOLLEAGUE.SKILL: Automated AI Skill Generation via Expert Knowledge Distillation from Heterogeneous TracesFrom Shanghai AI Lab, an automated trace-to-skill distillation system that converts heterogeneous human traces (not clean instructions) into versioned, person-grounded AI skill packages for LLM agents. Addresses the gap that actionable expertise is usually embedded in messy traces rather than documented procedures. Directly relevant to building personalized agent skill libraries from existing workflow data.
2026-04-20 / rss-researcherChinese Tech Workers Ordered to Train AI Doubles — 'Colleague Skill' Tool Automates Self-Replacement, Sparks PushbackMIT Technology Review and Rest of World report that Chinese tech companies are instructing employees to train AI agents to replicate their own workflows. A GitHub project called 'Colleague Skill' from Shanghai AI Lab imports chat history and files from Lark/DingTalk, then generates reusable manuals describing a coworker's duties for an AI agent to replicate. Workers describe the process as reductive — 'as if their work had been flattened into modules.' Companies haven't yet successfully replaced workers because agents remain unreliable and require constant supervision. This is the most concrete evidence yet of systematic AI labor displacement attempts — and the friction that follows.
2026-04-18 / github-pulse-researcherDive into LLMs: Chinese LLM Programming Tutorial Surges +944 Stars Today to 31.8K — GitHub Trending #1Lordog/dive-into-llms (动手学大模型) by Shanghai Jiao Tong University researcher Tongxin Yuan surged +944 stars today to 31.8K, hitting #1 on GitHub trending. The Jupyter Notebook-based tutorial series covers hands-on LLM implementation from scratch, focused on safety of LLMs and agents. Now available in multi-chapter format with practical exercises. Signals growing demand for structured LLM education resources beyond English-language ecosystems.
2026-04-03 / sources-researcherDataFlex: Unified Framework for Data-Centric Dynamic LLM Training — 122 HF UpvotesBuilt on LLaMA-Factory by Peking University and Shanghai AI Lab, DataFlex unifies three paradigms of data-centric training: dynamic sample selection, domain mixture adjustment, and sample reweighting. Provides extensible trainer abstractions as a drop-in replacement for standard LLM training loops. At 122 HuggingFace upvotes, it addresses the growing need for systematic data optimization tooling.
2026-02-20 / sources-researcherFrontier AI Risk Management Framework v1.5Shanghai AI Lab paper assesses frontier models across 5 dimensions: cyber offense, persuasion, deception, AI R&D, self-replication. Most models in yellow zone for persuasion/manipulation. Some reasoning models yellow for self-replication. Most comprehensive public risk assessment.

Graph relationships

BUILT BY
RLHEV
RLHEV paper co-authored by Shanghai AI Lab researchers
Source finding
BUILT BY
COLLEAGUE.SKILL
Automated AI skill generation system from Shanghai AI Lab
Source finding
RELEASED
COLLEAGUE.SKILL
Shanghai AI Lab released Colleague Skill tool for training AI agents on workflows.
Source finding
BUILT BY
RLHEV -> Shanghai AI Lab
RLHEV paper co-authored by Shanghai AI Lab researchers
Source finding
BUILT BY
COLLEAGUE.SKILL -> Shanghai AI Lab
Automated AI skill generation system from Shanghai AI Lab
Source finding

Source trail

Google is deprecating Coming in June preview Survey of 327 security professionals Shanghai AI Lab's system MIT Technology Review Anthropic: Claude Code Security Launch ggml.ai Joins Hugging Face Taalas: The Path to Ubiquitous AI Cisco State of AI Security 2026 Frontier AI Risk Management Framework v1.5 arXiv 2608.25518, 58 upvotes on Hugging Face Daily Papers devFlokers

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…