2024年真是生成式人工智能研究大放异彩的一年！最让我们惊讶的是，整个领域的焦点发生了翻天覆地的变化。尤其是在 2023 年和 2024 年，情况开始变得截然不同，由于大模型模型已经能够做很多事情，因此也更加关注应用层面的研究。
论文集合地址： https://github.com/aishwaryanr/awesome-generative-ai-guide
论文合集的分类框架如上图所示，把AI研究想象成一个从输入到输出的系统，就像实际部署的场景一样。这个框架分为几层，每层都有其独特的关注点：
输入层： 这是大模型应用的起点，聚焦于输入处理和提示工程的研究。通过巧妙调整输入数据的方式，我们可以让大型语言模型（LLM）输出更优质的结果。
数据/模型层： 这一层关注的是模型的“燃料”和“引擎”。研究内容包括提升数据质量、生成合成数据，确保模型在丰富多样的数据集上训练。此外，还有基础架构的创新，比如新模型架构、多模态能力（融合文本、图像等）、成本与尺寸优化、模型对齐以及扩展上下文长度等。
应用层： 研究如何将LLM应用于现实世界。无论是特定领域的模型（如代码生成、文本转SQL或医疗应用），还是微调、检索增强生成（RAG）和多智能体系统等技术，这一层都是将理论转化为实用工具的关键。
输出层： 如何确保模型的输出靠谱？这一层的研究集中在评估方法上，从人机交互系统到基准测试和LLM评委，提供了多种有效评估AI输出的手段。
挑战： 生成式AI的局限性：对抗性攻击、模型可解释性、幻觉问题等，这些都是我们需要克服的现实挑战，以确保AI更安全、更可靠。
输入层
提示工程
- DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines
- The Prompt Report: A Systematic Survey of Prompting Techniques
数据模型层
1. 数据质量/合成数据生成
- On LLMs-Driven Synthetic Data Generation, Curation, and Evaluation: A Survey
- Detecting Pretraining Data from Large Language Models
- A Survey on Data Synthesis and Augmentation for Large Language Models
- Scaling Synthetic Data Creation with 1,000,000,000 Personas
- Are AI Detectors Good Enough? A Survey on Quality of Datasets With Machine-Generated Texts
2. 基座大模型
- DeepSeek-V3 Technical Report
- xLSTM: Extended Long Short-Term Memory
- Sparks of Artificial General Intelligence: Early Experiments with GPT-4
- A Survey of Large Language Models
- SAM 2: Segment Anything in Images and Videos
- Qwen Technical Report
- RWKV: Reinventing RNNs for the Transformer Era
- KAN: Kolmogorov-Arnold Networks
- The Llama 3 Herd of Models
- Segment Anything
- Differential Transformer
- Foundation Models for Music: A Survey
3. 模型优化 (大小, 成本)
- A Survey of Small Language Models
- The Era of 1-bit LLMs: All Large Language Models are in 1.58 Bits
- TokenFormer: Rethinking Transformer Scaling with Tokenized Model Parameters
- A Survey on LLM Inference-Time Self-Improvement
- FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning
- LLM Pruning and Distillation in Practice: The Minitron Approach
- What is the Role of Small Models in the LLM Era: A Survey
4. 多模态
- Towards Generalist Biomedical AI
- MusicLM: Generating Music From Text
- The Dawn of LMMs: Preliminary Explorations with GPT-4V(ision)
- Multimodal Chain-of-Thought Reasoning in Language Models
5. 大模型对齐
- Meta-Rewarding Language Models: Self-Improving Alignment with LLM-as-a-Meta-Judge
- RLAIF vs. RLHF: Scaling Reinforcement Learning from Human Feedback with AI Feedback
- The Capacity for Moral Self-Correction in Large Language Models
- sDPO: Don't Use Your Data All at Once
6. 长上下文
- LongLoRA: Efficient Fine-tuning of Long-Context Large Language Models
- Evaluating Language Model Context Windows: A "Working Memory" Test and Inference-time Correction
- YaRN: Efficient Context Window Extension of Large Language Models
- LazyLLM: Dynamic Token Pruning for Efficient Long Context LLM Inference
- LongNet: Scaling Transformers to 1,000,000,000 Tokens
应用层
1.领域模型
- Qwen2.5-Coder Technical Report
- A Survey of Large Language Models for Healthcare: From Data, Technology, and Applications to Accountability and Ethics
- ChemCrow: Augmenting Large-Language Models with Chemistry Tools
- MentaLLaMA: Interpretable Mental Health Analysis on Social Media with Large Language Models
- A Preliminary Study of o1 in Medicine: Are We Closer to an AI Doctor?
- A Survey on Language Models for Code
- PMC-LLaMA: Towards Building Open-Source Language Models for Medicine
- ChemLLM: A Chemical Large Language Model
- A Survey of Large Language Models in Medicine: Progress, Application, and Challenge
- Can Large Language Models Unlock Novel Scientific Research Ideas?
2. RAG
- Corrective Retrieval Augmented Generation
- HybridRAG: Integrating Knowledge Graphs and Vector Retrieval Augmented Generation for Efficient Information Extraction
- Active Retrieval Augmented Generation
- GNN-RAG: Graph Neural Retrieval for Large Language Model Reasoning
- Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection
- Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
- Retrieval-Augmented Generation for Large Language Models: A Survey
- Text2SQL is Not Enough: Unifying AI and Databases with TAG
- Searching for Best Practices in Retrieval-Augmented Generation
- Seven Failure Points When Engineering a Retrieval Augmented Generation System
3. 智能体
- The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery
- Large Language Model-Brained GUI Agents: A Survey
- A Survey on Large Language Model based Autonomous Agents
- Augmented Language Models: a Survey
- A Taxonomy of AgentOps for Enabling Observability of Foundation Model based Agents
- Toolformer: Language Models Can Teach Themselves to Use Tools
4. 多智能体
- Emergent Autonomous Scientific Research Capabilities of Large Language Models
- OpenDevin: An Open Platform for AI Software Developers as Generalist Agents
- AutoGen Studio: A No-Code Developer Tool for Building and Debugging Multi-Agent Systems
- Mora: Enabling Generalist Video Generation via A Multi-Agent Framework
- AIOS: LLM Agent Operating System
- AnyTool: Self-Reflective, Hierarchical Agents for Large-Scale API Calls
- Large Language Model-Based Agents for Software Engineering: A Survey
5. 大模型微调
- Instruction Tuning with GPT-4
- LLMs + Persona-Plug = Personalized LLMs
- Pretraining Data Mixtures Enable Narrow Model Selection Capabilities in Transformer Models
- QLoRA: Efficient Finetuning of Quantized LLMs
- LoraHub: Efficient Cross-Task Generalization via Dynamic LoRA Composition
- LoRA+: Efficient Low Rank Adaptation of Large Models
- SQL-PaLM: Improved Large Language Model Adaptation for Text-to-SQL
- A Survey on Employing Large Language Models for Text-to-SQL Tasks
- Can Small Language Models Help Large Language Models Reason Better?: LM-Guided Chain-of-Thought
输出层
大模型评估
- A Multitask, Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and Interactivity
- Let Me Speak Freely? A Study on the Impact of Format Restrictions on Performance of Large Language Models
- RAGEval: Scenario-Specific RAG Evaluation Dataset Generation Framework
- Replacing Judges with Juries: Evaluating LLM Generations with a Panel of Diverse Models
- A Survey on LLM-as-a-Judge
- AgentBench: Evaluating LLMs as Agents
- A Survey on Evaluation of Large Language Models
- Self-Taught Evaluators
- PromptBench: A Unified Library for Evaluation of Large Language Models
- A Comprehensive Evaluation of Quantized Instruction-Tuned Large Language Models: An Experimental Analysis up to 405B
- Evaluating Large Language Models: A Comprehensive Survey
- Mathematical Capabilities of ChatGPT
挑战
生成式AI的局限性
- LLMs Know More Than They Show: On the Intrinsic Representation of LLM Hallucinations
- A Survey on Hallucination in Large Vision-Language Models
- A Survey of Hallucination in Large Foundation Models
- Chain-of-Verification Reduces Hallucination in Large Language Models
- A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions
- One Small Step for Generative AI, One Giant Leap for AGI: A Complete Survey on ChatGPT in AIGC Era
- Knowledge Conflicts for LLMs: A Survey
点击阅读原文，可以访问带有论文链接的文章
添加微信，回复”RAG“进入交流群