Tencent
Posted 2w ago

元宝—面向开放域信息获取与推理的大模型智能体研究

Tencent
Beijing, Beijing, China
OnsiteInternship
Responsibilities
  • designing agents
  • optimizing strategies
  • building datasets
Requirements
  • PhD or outstanding Master\u0002s in CS-related fields
  • Strong deep learning and engineering skills
  • Experience with large models
  • Reinforcement learning, and evaluation
  • Proficient in C/C++ and at least one DL framework
Technical tools mentioned
CaffeTensorFlowPyTorchC++ShellPythonMATLAB

Job description

课题详情

课题背景:

随着大语言模型在问答、知识分析与任务执行等场景中的应用不断深入,模型能力正从“基于参数知识作答”逐步走向“面向开放环境主动获取信息并完成推理决策”。在真实用户场景中,模型往往需要针对开放域问题自主完成搜索意图理解、查询规划、外部信息获取、网页内容甄别、多源证据整合、冲突信息消解与最终答案生成。这类能力已超出传统静态知识问答或狭义检索增强生成范畴,更接近于具备搜索、分析与推理闭环能力的信息智能体。

然而,当前具备联网搜索能力的大模型系统,在复杂开放域任务中仍存在明显不足,例如搜索策略单一、对搜索结果利用不充分、跨网页信息整合能力弱、长链路推理不稳定,以及在复杂问题下容易出现“能搜到但不会分析”“能引用但难以判断可靠性”等问题。因此,本课题聚焦于开放域信息获取与推理场景,研究具备联网能力的大模型智能体系统,结合奖励模型(RM)与强化学习(RL)方法,提升模型在复杂信息任务中的搜索、分析、推理与回答一体化能力。

课题挑战:

1、开放域信息智能体流程设计:如何设计适用于互联网开放环境的 Agent 系统流程,使模型能够自主完成问题理解、查询规划、分步搜索、证据筛选、归纳推理与答案生成;

2、复杂搜索轨迹的奖励建模:如何针对查询改写质量、搜索结果相关性、信息利用效率、证据可靠性和最终回答质量,构建细粒度奖励模型;

3、RM/RL 驱动的策略优化:如何结合奖励模型与强化学习方法,持续优化智能体的搜索决策、信息整合与长链路推理能力;

4、开放环境下的鲁棒性与真实性问题:如何应对互联网信息噪声大、内容冗余、来源冲突、时效变化快等问题,提升系统的稳定性、事实一致性与可追溯性;

5、自动化评测体系建设:如何面向开放域信息获取与推理任务构建自动评测系统,覆盖搜索效果、证据归因、答案正确性、时效性与用户体验等关键维度。

具体工作:

你将参与具备联网能力的大模型智能体系统设计与优化,围绕开放域复杂信息任务,研究搜索规划、多轮信息获取、证据选择、跨网页内容整合与引用生成等关键能力;参与构建训练数据、偏好数据与评测数据,探索奖励模型训练、强化学习优化和自动化评估方法;推动混元大模型在开放域信息获取、分析推理与智能体执行场景下的能力持续提升。

课题要求

1、包含但不限于计算机、信息工程、模式识别、人工智能、自动化、软件工程、电子工程、统计学、应用数学、物理学/量子计算、信息安全、信号与信息处理等专业的博士和优秀硕士;

2、熟练掌握大模型、多模态相关的基本算法及应用,熟悉caffe、tensorflow,pytorch等至少一个深度学习框架;

3、较强的工程实现能力,熟练掌握 C/C++ 编程,熟悉 Shell/Python/Matlab 至少一种编程语言。

About Tencent

Provides integrated internet services, digital entertainment, and cloud technology.

Year founded
1998
Employees
112771
Organization type
Public
Headquarters
CN

Similar jobs

Research Intern roles near Beijing, Beijing
5d
Save
Mark Applied
Hide
研究型实习生-学习型查询优化的研究和应用
Beijing, Beijing, China
RemoteInternship
Alibaba Group
Alibaba GroupNYSE: BABA: Global technology specializing in e-commerce and cloud computing.
PhD in computer science, artificial intelligence, mathematics, or related field; strong AI-for-database and systems expertise; Java/C++ development; distributed computing and query optimization knowledge; high-level research publications.
Java, C++, Snowflake, BigQuery, Spark, StarRocks
5d
Save
Mark Applied
Hide
研究型实习生 - Agentic Search算法
Beijing or Hangzhou
RemoteInternship
Alibaba
AlibabaNYSE: BABA: Provides online marketplaces, cloud computing, and digital payment services.
PhD student in computer science, artificial intelligence, statistics, or related field; strong algorithms; Python and PyTorch; experience with agents, tool use, RAG, multi-turn retrieval, and research experimentation.
Python, PyTorch, RAG, ReAct
5d
Save
Mark Applied
Hide
研究型实习生-学习型查询优化的研究和应用
Beijing, Beijing, China
HybridInternship
Alibaba Cloud
Alibaba CloudNYSE: BABA: Global cloud computing infrastructure and services provider
PhD in computer science, AI, mathematics, or related field; strong AI-for-systems theory, algorithms, Java/C++, distributed computing, query optimization, and research publication experience.
Java, C++, Snowflake, BigQuery, Apache Spark, StarRocks
1w
Save
Mark Applied
Hide
PhD Research Intern - Music-Conditioned Video
Beijing, Beijing, China
OnsiteInternship
Dolby Laboratories
Dolby LaboratoriesNYSE: DLB: Develops audio and video signal processing and compression technologies.
PhD candidates need deep learning research experience in music or video generation, strong Python and PyTorch skills, experience with Transformer or Diffusion Models, analytical ability, and written Chinese and English.
PyTorch, Python, Transformer, Diffusion Models
1w
Save
Mark Applied
Hide
PhD Research Intern - Music-Conditioned Video
Beijing, Beijing, China
OnsiteInternship
Dolby
DolbyNYSE: DLB: Develops and licenses audio and visual entertainment technologies.
PhD candidate in deep learning for music and video generation with experience in multimodal models, Transformers, diffusion models, PyTorch, and Python; strong analytical and Chinese and English communication skills.
PyTorch, Python, Transformer, Diffusion Models
2w
Save
Mark Applied
Hide
微信小微-记忆系统的构建与下游应用
Beijing, Beijing, China
OnsiteInternship
Tencent
TencentHong Kong Stock Exchange: 0700: Multinational technology conglomerate providing internet and entertainment services.
Seeking MS or PhD students in computer science, AI, statistics, information engineering or data mining with skills in user modeling, memory retrieval, NLP and LLM/RAG techniques; strong data analysis and problem‑solving.
NLP, RAG, LLM, knowledge graph, graph learning
2w
Save
Mark Applied
Hide
2027 Off-cycle Internship - Global Research - Mainland China
Beijing or Shanghai
OnsiteInternship
UBS
UBSSIX Swiss Exchange: UBSG: Provides global wealth management and investment banking services.
Student able to defer studies 6 months, graduating Dec 2027–Jun 2028; legal right to work locally; interest in finance; strong communication, analytical and strategic thinking; familiarity with AI use cases.
1mo
Save
Mark Applied
Hide
研究型实习生-Agent算法方向
Beijing or Hangzhou
OnsiteInternship
Xiaohongshu
Xiaohongshu: Social media and e-commerce platform for lifestyle sharing.
Research-focused intern in agent/ML and multimodal content understanding; strong CV/multimodal background, PyTorch and Linux experience, strong programming and research skills; long internship (>=4 months) and 4 days/week in Beijing.
PyTorch, Linux, LLM, VLLM