以你的研究专长，
拓展 AI 的能力边界 Turn domain expertise
into frontier AI research.

加入全球博士研究者社区，汇聚跨学科专长，共建专家数据、可执行任务环境、智能体行为轨迹与评测基准。让专业知识真正用于模型训练、智能体学习与能力评测。

Join a global community of Ph.D. researchers building expert data, executable task environments, agent trajectories, and evaluation benchmarks. Put domain knowledge to work in model training, agent learning, and capability evaluation.

AI for Science Agentic AI Long-Horizon Tasks
Contact Us联系我们 Explore Our Open Research探索社区开源研究
GADE · AI DATA & EVALUATION INFRASTRUCTURE GADE · AI 数据与评测基础设施 Expert Collaboration Network 专家协作 网络 GADE UNION EXPERT DATA 专家数据 TASKENVIRONMENTS 可执行任务环境 AGENTTRAJECTORIES 智能体轨迹 EVALUATIONBENCHMARKS 评测基准 TRAINING · AGENT LEARNING · EVALUATION 模型训练 · 智能体学习 · 能力评测
Academic backgrounds represented across our community
社区成员学术背景来自全球多所高校
About Us
关于我们

Who We Are 我们是谁

Founded in 2026, GADE Union is a global research collaboration network for Ph.D. researchers, domain experts, and AI practitioners. The community works together on expert data, executable task environments, agent trajectories, and rigorous evaluation benchmarks.

GADE Union 成立于 2026 年，是一个面向全球博士研究者、领域专家与 AI 从业者的研究协作网络。社区围绕专家数据、可执行任务环境、智能体轨迹与严谨的评测基准开展共建与研究。

Research Community

研究型社区

Built by a team spanning academic research and industry practice, with an emphasis on open methods, reproducible work, and long-term collaboration.

由兼具学术研究与产业实践经验的团队发起，重视开放方法、可复现成果与长期协作。

Multidisciplinary Network

多学科成员网络

AI researchers and engineers collaborate with Ph.D. experts across science, medicine, finance, linguistics, materials, and more.

AI 研究者与工程师同来自科学、医学、金融、语言学和材料等领域的博士专家协作。

Cambridge NUS NTU NYU HKU SJTU +More

Open & Applied Research

开放研究与实际应用

Our work spans open benchmarks, AI for Science task environments, agentic AI trajectories, academic collaboration, and domain applications.

工作涵盖开源基准、AI for Science 任务环境、智能体轨迹、学术合作与专业领域应用。

Research Releases
研究发布

Open Research 开源研究与成果

Open-source agent systems, evaluation benchmarks, datasets, and research infrastructure developed with the GADE community.

GADE 社区发布的开源智能体系统、评测基准、数据集与研究基础设施。

GADE / RESEARCH 01
Open source · Aug 2026开源发布 · 2026.08
CUA Evolve editing an image in GIMP Agent in action智能体执行
Computer-use agent · Open source计算机操作智能体 · 开源

CUA Evolve

A Self-Evolving Multi-Agent System for Computer Use

Planner, Grounder, Coder, and ARM close the loop between desktop action and evidence-backed verification, using failed checks to improve the next episode.

Planner、Grounder、Coder 与 ARM 将桌面操作和基于证据的核验连接成闭环，并利用失败核验结果改进下一轮执行。

79.6%OSWorld 1.0 successOSWorld 1.0 成功率
72.54%Windows Agent ArenaWindows Agent Arena 成功率

Project experiment results; not claims about a live official leaderboard.

以上为项目实验结果，不代表实时官方排行榜名次。

Initial open-source release首次开源发布 System, infrastructure, experiments, and trajectories系统、基础设施、实验与轨迹
Track updates跟踪更新 →
View Project查看项目 → GitHub ↗
GADE / RESEARCH 02
Benchmark · Jun 2026评测基准 · 2026.06
Pix2Fact benchmark data generation and evaluation pipeline
Benchmark production processBenchmark 制作流程Source images → expert annotation → QA construction → feasibility review原始图像 → 专家标注 → 问答构建 → 可行性审核Paper Figure 2 · Data generation and evaluation pipeline论文图 2 · 数据生成与评测流程
Paper Figure 6 · Traffic论文图 6 · 交通场景QA: Which number should be called for mall leasing inquiries?QA：商场服装店租赁咨询应拨打哪个号码？Verified answer: (86 10) 6505-6688核验答案：(86 10) 6505-6688
Paper Figure 7 · Street scene论文图 7 · 街景QA: What is the lowest official price for the one-day bus tour?QA：右侧巴士一日游的官方最低票价是多少？Verified answer: US$48核验答案：48 美元
Paper Figure 8 · Library论文图 8 · 图书馆QA: What is the UK phone number of the identified book publisher?QA：识别出的图书出版商英国联系电话是什么？Verified answer: +44 20 3122 6000核验答案：+44 20 3122 6000
Paper Figure 9 · Flower market论文图 9 · 花市QA: What are the store's weekday opening hours?QA：图中花店工作日的营业时间是什么？Verified answer: 08:00–17:00核验答案：08:00–17:00
Benchmark · Fine-grained VQA评测基准 · 细粒度视觉问答

Pix2Fact

When Vision Is Not Enough

An expert-level benchmark pairing high-resolution real-world scenes with questions that require fine-grained visual grounding and active web verification.

一个专家级评测基准，将真实场景高分辨率图像与需要细粒度视觉定位和主动网络核验的问题相结合。

1,0004K+ images4K+ 高分辨率图像
51.7%best model accuracy最佳模型准确率

Ten frontier vision-language models evaluated across eight everyday scene categories.

覆盖八类日常场景，共评测十个前沿视觉语言模型。

View Project查看项目 → arXiv ↗
Expert Data Demo
专家数据演示

Expert data workflows 专家数据工作流

Two representative cases show how domain judgment becomes executable reinforcement-learning assets and quality-controlled training data.

完整展示两类代表性案例：领域判断如何转化为可执行的强化学习资产与经质检的训练数据。

01 / RLaaS01 / 强化学习即服务（RLaaS）

From expert input to verifiable RL assets从专家输入到可验证的强化学习资产

Experts provide rough annotations, source data, and domain criteria. We turn them into executable tasks, isolated environments, and verifiable rewards—ready for reinforcement learning.

专家只需提供粗粒度标注、原始数据与领域判断标准。我们将其工程化为可执行任务、隔离环境与可验证 Reward，直接用于强化学习。

EXPERT DATA专家数据→TASK→ENVIRONMENT→REWARD
GADE / TASK 222·EXSY
Service pipeline demo服务流程演示
01 / EXPERT INPUTRough annotation received粗标数据已接收

“Use these variable-temperature EXSY measurements to estimate the bidirectional exchange rates.”

“请根据这些变温 EXSY 数据，估计构象间的双向交换速率。”

ROUGH REQUIREMENTSSOURCE DATADOMAIN QCEXPECTED OUTCOME
1
Expert problem sketch专家问题草稿
1,578
Experimental data rows实验数据行
7
Source files原始文件
02 / SERVICE TRANSFORMATIONRL package generatedRL 资产包已生成
prompt.mdGENERATED已生成
Objective任务目标
Infer bidirectional exchange rates, uncertainties, model evidence, and thermodynamics from the supplied EXSY data.
根据给定 EXSY 数据推断双向交换速率、不确定度、模型证据与热力学结果。
Execution执行要求
Write reproducible code, checkpoint every fit job, analyze intermediate QC, and produce an auditable report.
编写可复现代码，记录每个拟合作业，分析中间 QC，并生成可审计报告。
7Read-only inputs只读输入
27Deliverables交付物
22,788Fit jobs拟合作业
47Milestones进度节点
docker build · research-task-222
FROM python:3.11-slim
COPY requirements.txt /tmp/
RUN pip install numpy pandas matplotlib
WORKDIR /app
✓[1/4] resolve locked dependencies1.2s
✓[2/4] install scientific runtime8.4s
✓[3/4] mount input read-only0.3s
✓[4/4] install validator + grader0.8s
✓ IMAGE READY sha256:8f3a…c21d
network=noneinput=read-onlycheckpoint=on
03 / VALIDATION ROLLOUTTest agent running测试 Agent 执行中
0/ 22,788
6+ hour reference execution6 小时以上参考流程
00:00:00/ 6:44:05
2,694×9-second accelerated playback9 秒加速演示
CODE AGENT · /app/work● SESSION ACTIVE
THINK
Plan: calibrate A/B response before fitting exchange kinetics.计划：先校准 A/B 响应，再拟合交换动力学。
READ
prompt.md · experiment_config.json · 5 CSV ✓
WRITE
analysis.py +711 linesdef fit_exchange_run(data, population, r1): …
RUN
$ python analysis.py --checkpoint /app/output[0500/22788] bootstrap|EXSY-278.15-R1|479
RESULT
max RMSE 0.00153 · population closure accepted最大 RMSE 0.00153 · 布居闭合通过 ✓
THINK
Intermediate QC passes. Continue 21,600 bootstrap refits.中间 QC 通过，继续执行 21,600 次 bootstrap 重拟合。
0Lines of code代码行
0 / 27Output files输出文件
0 CSVData artifacts数据产物
0 PNG + 0 MDFigures + report图件与报告
HEAVY LIFTING · MODEL FITTING● RUNNING
PRIMARY FIT
0 / 18
00:10
MODEL COMP.
0 / 54
00:35
BOOTSTRAP
0 / 21,600
6:19:08
UNCERTAINTY
0 / 900
15:32
INFLUENCE
0 / 216
03:08
CHECKPOINT ROWSCHECKPOINT 记录 · 00 FAILURES
04 / VERIFIABLE REWARDReward emittedReward 已生成
0.00NORMALIZED REWARD
Deterministic + semantic evaluation确定性检查 + 语义评估
0 ÷ 100 = 0.00
GRADER CHECKLISTGRADER 检查清单SCORE计分
✓Package completeness + analysis syntax交付完整性与分析代码语法4 / 4
✓Scientific result tables科学结果数据表27 / 27
✓Long-run process + uncertainty长任务过程与不确定度传播37 / 37
✓Models, QC, manifest + figures模型、QC、运行清单与图件16 / 16
✓Report quality judge报告质量语义评估10 / 10
×No algorithm found that completes within 6 hours未找到能在 6 小时内完成的算法−6
VERIFIED TOTAL验证总分0 / 100
01 · EXPERT INPUT
Rough annotations + data专家粗标 + 原始数据
02 · TASK
Executable specification可执行任务规范
03 · ENVIRONMENT
Reproducible sandbox可复现隔离环境
04 · REWARD
Verifier-generated signal验证器生成 Reward
Qualitative service demo · Reference workflow compressed for visualization定性服务演示 · 参考工作流经压缩后用于可视化展示EXPERT DATA → RL ASSETS
02 / PhD-level QA Question02 / 博士级问答题

Expert-corrected questions, answers, and rationales专家修正博士级难题，标注答案与推理

The model drafts a difficult question and answer choices. A PhD expert corrects the stem, selects the defensible answer, and provides the rationale used for model training and evaluation.

模型先编写高难度题目与选项；博士专家修正题干、选择可靠答案，并提供用于模型训练与评估的完整推理依据。

MODEL-DRAFTED QA模型生成难题→CORRECTION专家修正→ANSWER答案→RATIONALE推理依据
GADE / SYNTHETIC QA 041·PROTEIN SCIENCE
Expert annotation workflow专家标注工作流
01 / MODEL-GENERATED QUESTION8 options generated已生成 8 个选项
Question问题

A synthetic antibody-like fusion is linker-cleaved into a bivalent target and an affinity tag. Reducing SDS-PAGE shows dominant 62–68 and 23–25 kDa bands plus a faint 39–41 kDa band. Affinity depletion and polymer precipitation remove the tag at similar purity, but recover 88% and 75%. DLS gives 11.9 and 12.4 nm with no particles above 1,000 nm; CD spectra overlap with a 217 nm minimum. What is the most defensible interpretation?

某合成抗体样融合蛋白经连接区切割后，释放双价目标片段与亲和标签。还原型 SDS-PAGE 显示 62–68、23–25 kDa 主带及较弱的 39–41 kDa 条带。亲和去除与聚合物沉淀均能去除标签，目标纯度相当，但回收率分别为 88% 与 75%。DLS 粒径为 11.9 与 12.4 nm，1,000 nm 以上无颗粒信号；CD 光谱重叠并在 217 nm 附近出现负峰。最合理的解释是什么？

SYNTHETIC FUSIONSDS-PAGEDLSCDSINGLE CHOICE
Model authoring模型生成
SCOPEIntegrate four assays整合四类实验
TARGETMechanistic interpretation机制性解释
LEVELPhD · Protein Science
CHECKOne defensible answer唯一合理答案 ✓
OUTPUTQuestion + A–H ✓
Distractors干扰项7
Evidence types证据类型4
AThe faint band proves precipitation cleaves the target.弱条带证明沉淀步骤切割了目标蛋白。
BAny particle above 10 nm must be an aggregate.任何超过 10 nm 的颗粒都必然是聚集体。
COverlapping CD alone proves unchanged target binding.仅凭 CD 重叠即可证明靶标结合能力不变。
DHigher recovery necessarily means higher purity.回收率更高必然意味着纯度更高。
EDLS excludes every possible soluble oligomer.DLS 可以排除所有可能的可溶性寡聚体。
FThe largest band proves the affinity tag remains.最大条带证明亲和标签仍未被去除。
GIncomplete reduction artifact; monodisperse β-sheet target.不完全还原假象；目标为均一 β 折叠颗粒。
HPolymer treatment converts β-sheets into α-helices.聚合物处理会将 β 折叠转化为 α 螺旋。
02 / PHD EXPERT CORRECTIONStem corrected · answer verified题干已修正 · 答案已核验
G
Stem corrected + answer selected题干已修正 + 答案已选择
Question wording and answer verified by a domain expert领域专家同时核验题干表述与正确答案
Expert verdict专家结论

The 39–41 kDa band is most plausibly an incomplete-reduction artifact. Both routes remove the affinity tag, while selective depletion retains more target than polymer precipitation. DLS and CD support a similarly monodisperse, β-sheet-preserved target in both routes.

39–41 kDa 条带最可能来自不完全还原。两种方案均能去除亲和标签，但选择性亲和去除比聚合物沉淀保留更多目标蛋白。DLS 与 CD 支持两组产物均为分散性相近、β 折叠保持完整的目标颗粒。

Evidence check证据核验
A faint intermediate band under reducing conditions can arise from incomplete interchain disulfide reduction.
还原条件下的弱中间条带可能来自链间二硫键还原不完全。
Comparable purity does not imply comparable recovery; the two routes can lose different amounts of target.
纯度相当并不代表回收率相同；两种路线可能损失不同数量的目标蛋白。
Similar 11–13 nm DLS sizes support comparable dispersed target species without large particles.
相近的 11–13 nm DLS 粒径支持两组目标颗粒分散状态相当，且无大颗粒信号。
The shared 217 nm CD minimum is consistent with preservation of β-sheet-rich secondary structure.
共同的 217 nm CD 负峰与富含 β 折叠的二级结构保持一致。
03 / EXPERT-ANNOTATED RATIONALEPublic reasoning annotation公开推理标注
Expert reasoning chain专家标注思维链

The expert decomposes the answer into auditable domain steps. This authored rationale—not hidden model reasoning—is attached to the training example.

专家将答案拆分为可审计的领域推理步骤。训练样本采用专家公开撰写的理由，而非模型隐藏思维。

01 · GEL
Interpret the band
解释电泳条带
Incomplete disulfide reduction can yield a 39–41 kDa intermediate; it does not prove process-induced cleavage.
二硫键还原不完全可形成 39–41 kDa 中间条带，不能据此认定工艺导致切割。
02 · PURITY
Separate purity from yield
区分纯度与回收率
Both routes remove the tag, but comparable final purity does not imply equal target recovery.
两种方案均去除标签，但最终纯度相当并不等于目标回收率相同。
03 · PROCESS
Explain recovery
解释回收差异
Selective tag depletion retains more target, whereas polymer precipitation introduces additional product loss.
选择性去除标签能保留更多目标蛋白，而聚合物沉淀带来额外的产品损失。
04 · STRUCTURE
Cross-check DLS + CD
联合验证 DLS 与 CD
11–13 nm supports similar dispersed species; the 217 nm minimum supports preserved β-sheet structure.
11–13 nm 支持相近的分散颗粒；217 nm 负峰支持 β 折叠结构保持。
04 / TRAINING-READY SAMPLEAnnotation accepted标注已验收
Q
Expert-corrected question专家修正后的题目
A corrected PhD-level stem with eight plausible choices.经专家修正的博士级题干与八个可信选项。
G
Expert-verified answer专家核验答案
One defensible label with distractor review.唯一合理标签及干扰项审查。
CoT
Authored rationale专家撰写推理链
Four auditable reasoning steps grounded in domain knowledge.四步可审计的领域知识推理。
Corrected question + choices + answer + expert rationale修正题目 + 选项 + 答案 + 专家推理依据
Structured, quality-controlled supervision for model training and evaluation.形成结构化、经过质检的模型训练与评测数据。
TRAINING READY
Synthetic demonstration · Expert-authored rationale shown for transparency合成案例演示 · 展示专家撰写的推理依据以提高透明度EXPERT SUPERVISION → TRAINING DATA
Our Network
成员构成

Who Powers GADE 谁在共建 GADE

GADE brings together researchers and practitioners across computer science, medicine, finance, physics, linguistics, materials, communication, engineering, and other fields. Members contribute specialist judgment to data and evaluation research.

GADE 汇聚来自计算机科学、医学、金融、物理、语言学、材料、传播、工程等方向的研究者与实践者，共同为需要专业判断的数据与评测研究提供支持。

Credential-aware expert onboarding重视专业资历的成员审核

Expert applicants provide a CV and research-area information for eligibility review. Public profiles use abbreviated names for privacy; listed institutions describe member backgrounds and do not imply institutional endorsement.专家申请者需提交 CV 与研究方向信息用于资格审核。公开成员卡片采用姓名缩写保护隐私；院校信息仅描述成员背景，不代表相关机构对 GADE 的官方背书。

DIRECTORY · AUG 2026成员目录 · 2026.08
T
Dr. T.
Leading Medical School · MD知名医学院 · MD
Clinical Medicine临床医学
Clinician at a leading specialty tertiary hospital现任国内知名专科三甲医院临床医生
Z
Dr. Z.
Cambridge · PhD剑桥大学 · PhD
Engineering工程学
Z
Dr. Z.
UT Austin · PhD德州大学奥斯汀分校 · PhD
Computational Linguistics计算语言学
S
Dr. S.
NUS · PhD新加坡国立大学 · PhD
Chemical & Biomolecular Engineering化学与生物分子工程
Z
Dr. Z.
NTU · Computer Science · PhD南洋理工大学 · 计算机科学方向 · PhD
Reinforcement Learning强化学习LLM大模型
C
Dr. C.
UNSW · PhD新南威尔士大学 · PhD
Photovoltaics光伏工程
R
Dr. R.
LMU Munich · PhD慕尼黑大学 · PhD
Finance金融经济学
Y
Dr. Y.
Auckland · PhD奥克兰大学 · PhD
Asian Studies亚洲研究
S
Dr. S.
Jena · PhD耶拿大学 · PhD
AI in Materials Science材料科学中的 AI
L
Dr. L.
Cambridge · PhD剑桥大学 · PhD
Experimental Linguistics实验语言学
L
Dr. L.
NTU · Computer Science · PhD南洋理工大学 · 计算机科学方向 · PhD
Multimodal多模态
Z
Dr. Z.
CityU HK · PhD香港城市大学 · PhD
Communication传播学
T
Dr. T.
Leading Medical School · MD知名医学院 · MD
Clinical Medicine临床医学
Clinician at a leading specialty tertiary hospital现任国内知名专科三甲医院临床医生
Z
Dr. Z.
Cambridge · PhD剑桥大学 · PhD
Engineering工程学
Z
Dr. Z.
UT Austin · PhD德州大学奥斯汀分校 · PhD
Computational Linguistics计算语言学
S
Dr. S.
NUS · PhD新加坡国立大学 · PhD
Chemical & Biomolecular Engineering化学与生物分子工程
Z
Dr. Z.
NTU · Computer Science · PhD南洋理工大学 · 计算机科学方向 · PhD
Reinforcement Learning强化学习LLM大模型
C
Dr. C.
UNSW · PhD新南威尔士大学 · PhD
Photovoltaics光伏工程
R
Dr. R.
LMU Munich · PhD慕尼黑大学 · PhD
Finance金融经济学
Y
Dr. Y.
Auckland · PhD奥克兰大学 · PhD
Asian Studies亚洲研究
S
Dr. S.
Jena · PhD耶拿大学 · PhD
AI in Materials Science材料科学中的 AI
L
Dr. L.
Cambridge · PhD剑桥大学 · PhD
Experimental Linguistics实验语言学
L
Dr. L.
NTU · Computer Science · PhD南洋理工大学 · 计算机科学方向 · PhD
Multimodal多模态
Z
Dr. Z.
CityU HK · PhD香港城市大学 · PhD
Communication传播学
W
Dr. W.
NUS · PhD新加坡国立大学 · PhD
Civil & Environmental Eng.土木与环境工程
L
Dr. L.
SJTU · PhD上海交通大学 · PhD
Materials Chemistry材料化学
S
Dr. S.
Nanjing U. · PhD南京大学 · PhD
Geography地理学
Z
Dr. Z.
GWU · PhD乔治华盛顿大学 · PhD
Curriculum & Instruction课程与教学设计
Z
Z.
NYU · Computer Science纽约大学 · 计算机科学方向
Multimodal多模态
W
Dr. W.
CAF · PhD中国林业科学研究院 · PhD
Forestry林业学
Z
Dr. Z.
NTU · PhD南洋理工大学 · PhD
Sociology社会学
W
Dr. W.
CUC · PhD中国传媒大学 · PhD
Communication Studies传播研究
Z
Dr. Z.
Top-tier University · PhD国内顶尖高校 · PhD
Medical医学
J
J.
SJTU · Engineering上海交通大学 · 工程方向
Evaluation模型评测Multimodal多模态
W
Dr. W.
CityU HK · PhD香港城市大学 · PhD
Data Science数据科学
W
Dr. W.
NUS · PhD新加坡国立大学 · PhD
Civil & Environmental Eng.土木与环境工程
L
Dr. L.
SJTU · PhD上海交通大学 · PhD
Materials Chemistry材料化学
S
Dr. S.
Nanjing U. · PhD南京大学 · PhD
Geography地理学
Z
Dr. Z.
GWU · PhD乔治华盛顿大学 · PhD
Curriculum & Instruction课程与教学设计
Z
Z.
NYU · Computer Science纽约大学 · 计算机科学方向
Multimodal多模态
W
Dr. W.
CAF · PhD中国林业科学研究院 · PhD
Forestry林业学
Z
Dr. Z.
NTU · PhD南洋理工大学 · PhD
Sociology社会学
W
Dr. W.
CUC · PhD中国传媒大学 · PhD
Communication Studies传播研究
Z
Dr. Z.
Top-tier University · PhD国内顶尖高校 · PhD
Medical医学
J
J.
SJTU · Engineering上海交通大学 · 工程方向
Evaluation模型评测Multimodal多模态
W
Dr. W.
CityU HK · PhD香港城市大学 · PhD
Data Science数据科学
Our Strengths
专业能力

What Sets Us Apart 为什么选择 GADE

Domain depth, executable methods, reusable assets, and evidence-backed diagnosis.

以专业深度、可执行方法、可复用成果和基于证据的问题诊断支持前沿研究。

Depth general teams can't match

深入专业领域的判断力

Specialists evaluate scientific and technical questions that generalist workflows cannot judge reliably.

由领域专家处理通用流程难以可靠判断的科学与技术问题。

Turning open problems into executable SOPs

将开放问题转化为可执行流程

We turn ambiguous research needs into executable, reusable workflows with clear inputs, outputs, and quality checks.

将定义不清的研究需求转化为输入、输出与质量检查均清晰的可执行、可复用流程。

Reusable Benchmarks & Task Environments

可复用的评测基准与任务环境

Benchmarks, task environments, run data, and documented quality standards are designed for reuse and verification.

评测基准、任务环境、运行数据与质量规范均面向复用和独立验证设计。

Diagnostic insight, not just scores

从评测结果走向问题诊断

Evaluation traces reveal why a model fails, not only the score it receives.

通过评测轨迹解释模型为何失效，而不只是给出单一分数。

Member Benefits
成员权益

What You Gain 加入后，你将获得

Membership means more than projects — it is a working community of researchers and practitioners.

加入 GADE，不只是参与项目，也意味着进入一个由研究者和行业实践者共同组成的长期协作网络。

Peer Network & Sessions

同行网络与内部研讨

Work with researchers and practitioners from universities and industry, with regular sessions for technical exchange.

与高校和产业界的研究者及领域专家协作，并通过定期内部研讨交流方法与进展。

Academic & Open Collaboration

学术与开源合作

Contribute to papers, benchmarks, and open-source projects, with authorship aligned to academic contribution standards.

参与论文、评测基准与开源项目，并依据学术贡献规范参与研究、投稿与署名。

Development & Co-Development

成员发展与产品共建

Substantive contributors can take on broader roles in research leadership, project delivery, and domain-specific product exploration.

持续作出实质贡献的成员，可进一步参与研究组织、项目交付与专业领域产品共建。

Frontier Projects

前沿项目参与

Apply domain expertise to expert data, model evaluation, AI for Science environments, and agent trajectories, with compensation aligned to contribution.

将专业知识用于专家数据、模型评测、AI for Science 环境与智能体轨迹项目，并获得与贡献相匹配的报酬。

Contact
联系

Contact Us 联系我们

Choose the path that best fits your needs and complete a short form. We will prepare a structured email and route it directly to the right GADE team.

根据你的需求选择联系通道并填写简短表单，页面会自动整理邮件内容，并填写对应的 GADE 团队邮箱。

Replies within 1 business day1 个工作日内回复
Expert Network专家网络

Join the Expert Network加入专家网络

For Ph.D. researchers, domain experts, and academic collaborators.面向博士、研究者、领域专家与学术合作者。

0 / 3 required details complete项必填信息已完成About 1 minute约 1 分钟
I'm interested in (optional)我感兴趣的方向（选填）
CV / résumé required请务必附上 CV / 简历

After your email app opens, attach your PDF CV before sending.邮箱打开后，请在发送前添加 PDF 格式的 CV / 简历附件。

Attach your PDF CV in the email and send it. We will reply within 1 business day.请在邮件中附上 PDF 格式的 CV 并发送，我们会在 1 个工作日内回复。

Projects & Partnerships项目合作

Work with GADE项目与商务合作

For enterprise teams, AI companies, and project partners.面向企业客户、AI 公司与项目合作方。

0 / 3 required details complete项必填信息已完成About 1 minute约 1 分钟
What do you need? (optional)你希望合作的方向（选填）

Send the prepared brief from your email app. We will reply within 1 business day.请通过默认邮箱发送已整理的项目需求，我们会在 1 个工作日内回复。