分享内容简介

分享内容大纲

历史回顾

神经语言模型

词向量

预训练语言模型

规模法则（Scaling laws）与涌现能力

从训练到测试

搜索与推断

自我学习

自我评估

思维树

AlphaGo

类AlphaZero树搜索

新的Scaling Law

Self-x AI

自我改进

自我一致性

自我对齐

自我精炼

自我反思

通向自我意识

主讲人介绍

张江，北京师范大学系统科学学院教授，集智俱乐部、集智学园创始人，集智科学研究中心理事长，曾任腾讯研究院、华为战略研究院等特聘顾问。主要研究领域包括因果涌现、复杂系统分析与建模、规模理论等。

主要涉及到的参考文献

 • F.Sun et al.: Learning Word Representations by Jointly Modeling Syntagmatic and Paradigmatic Relations (slides)

http://www.bigdatalab.ac.cn/~lanyanyan/slides/2015/ACL2015-sun.pdf

 • Mikolov, T., Chen, K., Corrado, G., & Dean, J. Efficient Estimation of Word Representations in Vector Space[C]//International Conference on Learning Representations. 2013.

https://arxiv.org/abs/1301.3781

 • Qiu, R., Zhou, D., Qian, W., et al. Ask, and it shall be given: Turing completeness of prompting[R]. 2024.

https://arxiv.org/pdf/2411.01992

 • Zhou, D., Zhang, S., Gheini, M., et al. Least-to-Most Prompting Enables Complex Reasoning in Large Language Models[J]. ArXiv, abs/2205.10625, 2022: n. pag.

https://arxiv.org/abs/2205.10625

 • Pérez, J., Martinez, F., & Barcelo, P. On the Turing Completeness of Modern Neural Network Architectures[J]. ArXiv, abs/1901.03429, 2019: n. pag.

 • Siegelmann, H. T., & Sontag, E. D. On the Computational Power of Neural Nets[J]. Journal of Computer and System Sciences, 1995, 50(1): 132–150.

http://binds.cs.umass.edu/papers/1992_Siegelmann_COLT.pdf

https://arxiv.org/abs/1901.03429

 • Wei, J., Wang, X., Schuurmans, D., et al. Chain of Thought Prompting Elicits Reasoning in Large Language Models[J]. ArXiv, abs/2201.11903, 2022: n. pag.

https://arxiv.org/abs/2201.11903

 • Chen, Q., Wu, X., Wang, Z., et al. Unlocking the Capabilities of Thought: A Reasoning Boundary Framework to Quantify and Optimize Chain-of-Thought[J]. ArXiv, abs/2410.05695, 2024: n. pag.

https://arxiv.org/abs/2410.05695

 • Kumar, T., Zhang, Y., & He, C. Scaling Laws for Precision[R]. 2024.

https://arxiv.org/abs/2411.04330

 • Wu, Y., Ma, Z., & Li, B. Inference Scaling Laws: An Empirical Analysis of Compute-Optimal Inference for Problem-Solving with Language Models[R]. 2024.

https://arxiv.org/abs/2408.00724

 • Huang, J., Wang, X., Wei, J., et al. Large Language Models Can Self-Improve[J]. ArXiv, abs/2210.11610, 2022: n. pag.

https://arxiv.org/abs/2210.11610

 • Wang, X., Wei, J., Schuurmans, D., et al. Self-Consistency Improves Chain of Thought Reasoning in Language Models[J]. ArXiv, abs/2203.11171, 2022: n. pag.

https://arxiv.org/abs/2203.11171

 • Li, X., Wang, X., Gao, J., et al. Self-Alignment with Instruction Backtranslation[J]. ArXiv, abs/2308.06259, 2023: n. pag.

https://arxiv.org/abs/2308.06259

 • Madaan, A., Touvron, H., Lample, G., et al. Self-Refine: Iterative Refinement with Self-Feedback[J]. ArXiv, abs/2303.17651, 2023: n. pag.

https://arxiv.org/pdf/2303.17651

 • Shinn, N., Labash, A., & Ahn, S. Reflexion: language agents with verbal reinforcement learning[C]//Neural Information Processing Systems. 2023.

https://arxiv.org/pdf/2303.11366

 • Tao, Z., Wang, X., & Wei, J. A Survey on Self-Evolution of Large Language Models[J]. ArXiv, abs/2404.14387, 2024: n. pag.

https://arxiv.org/pdf/2404.14387

直播信息

时间：

2024年12月7日（本周六）晚上19:30-21:30

参与方式：

o1模型代表大语言模型融合学习与推理的新范式。集智俱乐部联合北京师范大学系统科学学院教授张江、Google DeepMind研究科学家冯熙栋、阿里巴巴强化学习研究员王维埙和中科院信工所张杰共同发起「大模型II：融合学习与推理的大模型新范式 」读书会，本次读书会将关注大模型推理范式的演进、基于搜索与蒙特卡洛树的推理优化、基于强化学习的大模型优化、思维链方法与内化机制、自我改进与推理验证。希望通过读书会探索o1具体实现的技术路径，帮助我们更好的理解机器推理和人工智能的本质。

从2024年12月7日开始，预计每周六进行一次，持续时间预计 6-8 周左右。欢迎感兴趣的朋友报名参加，激发更多的思维火花！

详情请见：大模型2.0读书会：融合学习与推理的大模型新范式！