# Authoritative Sources for LLM Fact-Checking — by Topic

Generated 2026-05-20. Covers all 24 content chapters. Sources are primary papers, official docs, benchmarks, and authoritative blogs (no Medium, Substack, Twitter, YouTube, news).

---

## 1. Tokenizers, Embeddings, Semantic Space

### Foundational
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Neural Machine Translation of Rare Words with Subword Units | Sennrich, Haddow, Birch | 2016 | ACL | Introduced BPE for NLP; 4000+ citations |
| SentencePiece: A simple and language independent subword tokenizer and detokenizer | Kudo & Richardson | 2018 | EMNLP | Standard tokenizer library; 3000+ citations |
| Efficient Estimation of Word Representations in Vector Space | Mikolov et al. | 2013 | ICLR (workshop) | Word2Vec — foundational embedding work; 40000+ citations |
| GloVe: Global Vectors for Word Representation | Pennington, Socher, Manning | 2014 | EMNLP | Foundational dense embeddings; 40000+ citations |
| BERT: Pre-training of Deep Bidirectional Transformers | Devlin, Chang, Lee, Toutanova | 2019 | NAACL | Bidirectional pre-training paradigm; 70000+ citations |
| Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks | Reimers & Gurevych | 2019 | EMNLP | Sentence-level embeddings standard |

### Benchmarks & Tools
| Source | Authors/Org | Year | Venue | Why Authoritative |
|--------|-------------|------|-------|-------------------|
| MTEB: Massive Text Embedding Benchmark | Muennighoff et al. | 2023 | EACL | Standard embedding benchmark; HuggingFace leaderboard |
| tiktoken | OpenAI | 2022– | GitHub repo | Official BPE tokenizer for GPT models |
| HuggingFace Tokenizers | HuggingFace | 2019– | Library | Industry-standard tokenizer library |

### Modern Contextual Embeddings
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Matryoshka Representation Learning | Kusupati et al. | 2022 | NeurIPS | Matryoshka embeddings — variable-size representations |
| Text and Code Embeddings by Contrastive Pre-Training | Neelakantan et al. (OpenAI) | 2022 | arXiv:2201.10005 | OpenAI embedding models (text-embedding-ada-002) |

---

## 2. Transformer Architecture, Attention, MLP Layers

### Architecture Foundation
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Attention Is All You Need | Vaswani et al. | 2017 | NeurIPS | Original Transformer; 130000+ citations |
| GLU Variants Improve Transformer | Shazeer | 2020 | arXiv:2002.05202 | SwiGLU — activation used in modern LLMs |
| Transformer Feed-Forward Layers Are Key-Value Memories | Geva et al. | 2021 | EMNLP | MLP layers as associative memory; foundational interpretability |
| Gaussian Error Linear Units (GELUs) | Hendrycks & Gimpel | 2016 | arXiv:1606.08415 | GELU activation; used in GPT/BERT |
| Layer Normalization | Ba, Kiros, Hinton | 2016 | arXiv:1607.06450 | LayerNorm — foundational; 8000+ citations |

### Attention Variants
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Fast Transformer Decoding: One Write-Head is All You Need | Shazeer | 2019 | arXiv:1911.02150 | Multi-Query Attention (MQA) |
| GQA: Training Generalized Multi-Query Transformer Models | Ainslie et al. | 2023 | EMNLP | Grouped-Query Attention (GQA) — used in Llama, Gemma |
| FlashAttention: Fast and Memory-Efficient Exact Attention | Dao et al. | 2022 | NeurIPS | IO-aware exact attention; enables long context |
| FlashAttention-2: Faster Attention with Better Parallelism | Dao | 2023 | arXiv:2307.08691 | Improved parallelism + work partitioning |
| FlashAttention-3: Fast and Accurate Attention with Asynchrony and Low-precision | Shah et al. | 2024 | arXiv:2407.08608 | Hopper GPU optimizations |
| Ring Attention with Blockwise Transformers for Near-Infinite Context | Liu et al. | 2023 | arXiv:2310.01889 | Distributed attention for ultra-long context |

### Positional Encodings
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| RoFormer: Enhanced Transformer with Rotary Position Embedding | Su et al. | 2021 | arXiv:2104.09864 | RoPE — standard in Llama, Qwen, DeepSeek, Gemma |
| Train Short, Test Long: Attention with Linear Biases (ALiBi) | Press et al. | 2022 | ICLR | ALiBi — length extrapolation without positional embeddings |
| YaRN: Efficient Context Window Extension | Peng et al. | 2023 | arXiv:2309.00071 | RoPE interpolation for context extension |

### MoE (Mixture of Experts)
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Outrageously Large Neural Networks: The Sparsely-Gated MoE Layer | Shazeer et al. | 2017 | ICLR | Original MoE for language models |
| Switch Transformers: Scaling to Trillion Parameter Models | Fedus et al. | 2021 | JMLR | Simplified MoE routing; T5-based |
| Mixtral of Experts | Jiang et al. (Mistral AI) | 2024 | arXiv:2401.04088 | Production MoE: 8×7B, outperforms Llama 2 70B |
| DeepSeek-V3 Technical Report | DeepSeek-AI | 2024 | arXiv:2412.19437 | 671B MoE with multi-token prediction |

### Interpretability
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Knowledge Neurons in Pretrained Transformers | Dai et al. | 2022 | ACL | Factual knowledge localization |
| Locating and Editing Factual Associations in GPT | Meng et al. | 2022 | NeurIPS | ROME — causal tracing of facts |
| Mass-Editing Memory in a Transformer | Meng et al. | 2023 | ICLR | MEMIT — batched fact editing |
| Towards Monosemanticity: Decomposing LMs With Dictionary Learning | Bricken et al. (Anthropic) | 2023 | Transformer Circuits | Sparse autoencoders for features |
| Scaling Monosemanticity | Templeton et al. (Anthropic) | 2024 | Anthropic Research | Feature extraction from Claude 3 Sonnet |
| Transformers represent belief state geometry in their residual stream | Shai et al. | 2024 | arXiv:2405.15943 | Belief state geometry |
| The Geometry of Reasoning | Zhou et al. | 2025 | arXiv:2510.09782 | Reasoning geometry in representation space |

### State Space Models / Hybrids
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Efficiently Modeling Long Sequences with Structured State Spaces (S4) | Gu et al. | 2021 | ICLR 2022 | Original S4 — linear-time sequence modeling |
| Mamba: Linear-Time Sequence Modeling with Selective State Spaces | Gu & Dao | 2023 | arXiv:2312.00752 | Selective SSM; linear attention alternative |
| Transformers are SSMs (Mamba-2) | Dao & Gu | 2024 | ICML | Unified framework; SSD layer |
| Jamba: A Hybrid Transformer-Mamba Language Model | Lieber et al. (AI21) | 2024 | arXiv:2403.19887 | Production hybrid; Transformer + Mamba + MoE |
| Jamba 2 | AI21 Labs | 2026 | ai21.com/blog | Apache 2.0 hybrid, 256K context |

### Scaling Laws
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Scaling Laws for Neural Language Models | Kaplan et al. (OpenAI) | 2020 | arXiv:2001.08361 | Power-law scaling; 2000+ citations |
| Training Compute-Optimal Large Language Models (Chinchilla) | Hoffmann et al. (DeepMind) | 2022 | NeurIPS | Optimal compute allocation; 2000+ citations |

---

## 3. Hallucinations in LLMs

### Surveys & Taxonomies
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Survey of Hallucination in Natural Language Generation | Ji et al. | 2023 | ACM Computing Surveys | Canonical hallucination taxonomy; 1500+ citations |
| A Survey on Hallucination in Large Language Models | Huang et al. | 2023 | arXiv:2311.05232 | Comprehensive causes + mitigation review |

### Benchmarks & Metrics
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| TruthfulQA: Measuring How Models Mimic Human Falsehoods | Lin et al. | 2022 | ACL | Standard truthfulness benchmark |
| FActScore: Fine-grained Atomic Evaluation of Factual Precision | Min et al. | 2023 | EMNLP | Atomic fact verification metric |
| SimpleQA: Measuring Short-form Factuality | OpenAI | 2024 | openai.com/index/simpleqa | OpenAI factual accuracy benchmark |
| ClashEval: Quantifying the tug-of-war between an LLM's internal prior and external evidence | Wu, Wu, Zou | 2025 | arXiv:2404.10198 | Measures prior vs. evidence conflict |

### Mechanisms & Detection
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Language Models Don't Always Say What They Think | Turpin et al. | 2023 | NeurIPS | Unfaithful CoT explanations |
| Measuring Faithfulness in Chain-of-Thought Reasoning | Lanham et al. | 2023 | arXiv:2307.13702 | CoT faithfulness metrics |
| Towards Understanding Sycophancy in Language Models | Sharma et al. | 2024 | ICLR | Sycophancy — models agreeing with user bias |

### Mitigation
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Chain-of-Verification Reduces Hallucination | Dhuliawala et al. | 2023 | arXiv:2309.11495 | CoVe — self-verification loop |
| Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks | Lewis et al. | 2020 | NeurIPS | Original RAG as anti-hallucination |

---

## 4. Causal Decoding, Context Windows, Long Context

### Position & Context Mechanics
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Lost in the Middle: How Language Models Use Long Contexts | Liu et al. | 2023 | TACL | U-shaped attention curve in long context |
| Efficient Streaming Language Models with Attention Sinks | Xiao et al. | 2023 | ICLR 2024 | StreamingLLM — initial tokens as attention sinks |
| Longformer: The Long-Document Transformer | Beltagy et al. | 2020 | arXiv:2004.05150 | Sparse attention for long documents |
| Big Bird: Transformers for Longer Sequences | Zaheer et al. | 2020 | NeurIPS | Sparse + global + random attention |

### Context Extension
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| YaRN: Efficient Context Window Extension | Peng et al. | 2023 | arXiv:2309.00071 | NTK-aware RoPE interpolation; widely adopted |
| Self-Extend: Extending LLM Context Windows | Jin et al. | 2024 | arXiv:2401.01325 | Training-free extension via grouped attention |
| HELMET: How to Evaluate Long-context LMs Effectively and Thoroughly | Yen et al. | 2024 | arXiv:2410.02666 | Long-context eval framework |

### Causal Masking Fundamentals
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Language Models are Unsupervised Multitask Learners (GPT-2) | Radford et al. (OpenAI) | 2019 | OpenAI Blog | Causal (autoregressive) LM paradigm formalized |
| Language Models are Few-Shot Learners (GPT-3) | Brown et al. | 2020 | NeurIPS | Scaling causal LM to 175B; in-context learning |

### Diffusion LLM (alternatives to causal)
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Simple and Effective Masked Diffusion Language Models | Sahoo et al. | 2024 | NeurIPS | MDLM — masked diffusion for text |
| Discrete Diffusion Modeling (SEDD) | Lou et al. | 2024 | ICML | Score entropy discrete diffusion |
| Large Language Diffusion Models (LLaDA) | Nie et al. | 2025 | arXiv:2502.09992 | 8B diffusion model matching LLaMA3 |
| Mercury: The Fastest LLM | Inception Labs | 2025–2026 | inceptionlabs.ai | First production diffusion LLM |

---

## 5. Prompt Engineering, XML Markup, Chain-of-Thought

### Prompting Methods
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Chain-of-Thought Prompting Elicits Reasoning in LLMs | Wei et al. (Google) | 2022 | NeurIPS | Introduced CoT; 3000+ citations |
| Large Language Models are Zero-Shot Reasoners | Kojima et al. | 2022 | NeurIPS | "Let's think step by step" |
| Plan-and-Solve Prompting | Wang et al. | 2023 | ACL | Structured planning in zero-shot |
| Decomposed Prompting: A Modular Approach | Khot et al. | 2023 | ICLR | Task decomposition via prompting |
| Least-to-Most Prompting | Zhou et al. | 2023 | ICLR | Progressive complexity prompting |

### Structured Output & Prompt Architecture
| Source | Org | Year | Type | Why Authoritative |
|--------|-----|------|------|-------------------|
| Use XML tags to structure your prompts | Anthropic | 2024–2026 | Official docs | Primary guidance on XML markup for Claude |
| Prompt engineering best practices | OpenAI | 2024–2026 | Official docs | Prompt structure for OpenAI models |
| Structured Outputs API | OpenAI | 2024–2026 | Official docs | JSON Schema constrained decoding |
| Efficient Guided Generation for LLMs (Outlines) | Willard & Louf | 2023 | arXiv:2307.09702 | Structured generation via FSM |
| Guidance | Microsoft | 2023– | GitHub repo | DSL for controlling LLM output structure |
| LMQL | Beurer-Kellner et al. | 2023–2025 | arXiv | SQL-like language for LLM constraints |

### Prompt Optimization
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| DSPy: Compiling Declarative Language Model Calls | Khattab et al. | 2024 | ICLR/TMLR | Automated prompt optimization via signatures |
| Large Language Models as Optimizers (OPRO) | Yang et al. | 2024 | ICLR | LLMs as prompt optimizers |
| The Unreasonable Effectiveness of Eccentric Automatic Prompts | Battle & Gollapudi | 2024 | arXiv | Systematic prompt search |

### Agent-Specific Prompting
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| ReAct: Synergizing Reasoning and Acting | Yao et al. | 2023 | ICLR | Interleaved reasoning + action |
| Reflexion: Language Agents with Verbal RL | Shinn et al. | 2023 | NeurIPS | Self-reflection via verbal feedback |
| Tree of Thoughts | Yao et al. | 2023 | NeurIPS | Tree search over reasoning paths |
| Graph of Thoughts | Besta et al. | 2024 | AAAI | Graph-based reasoning framework |
| Language Agent Tree Search (LATS) | Zhou et al. | 2024 | arXiv:2310.04406 | Unified reasoning + acting + planning |

### Prompt Safety & Alignment
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Constitutional AI: Harmlessness from AI Feedback | Bai et al. (Anthropic) | 2022 | arXiv:2212.08073 | Self-critique + revision via constitution |
| Calibrate Before Use | Zhao et al. | 2021 | ICML | Few-shot calibration; format sensitivity |

---

## 6. Agents vs Chat, Tool Use, MCP Protocol

### Agent Foundations
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| ReAct: Synergizing Reasoning and Acting in Language Models | Yao et al. | 2023 | ICLR | Defines agent loop: Thought → Action → Observation |
| Cognitive Architectures for Language Agents (CoALA) | Sumers, Yao, Narasimhan, Griffiths | 2023 | TMLR 2024 | Cognitive framework for LLM agents |
| Generative Agents: Interactive Simulacra of Human Behavior | Park et al. | 2023 | UIST | Agent memory + reflection architecture |
| Building Effective Agents | Anthropic | 2024 | anthropic.com/research | Definitive agent design patterns guide |
| A Practical Guide to Building Agents | OpenAI | 2025 | platform.openai.com | OpenAI agent building patterns |

### Tool Use
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Toolformer: Language Models Can Teach Themselves to Use Tools | Schick et al. (Meta) | 2023 | NeurIPS | Self-supervised tool learning |
| PAL: Program-aided Language Models | Gao et al. | 2023 | ICML | Code as reasoning + computation |
| Tool Learning with Foundation Models | Qin et al. | 2024 | Nature Machine Intelligence | Comprehensive tool-learning survey |
| Tool Use API docs | Anthropic | 2024–2026 | Official docs | Native tool calling for Claude |
| Tools, Connectors, and MCP | OpenAI | 2025–2026 | Official docs | Tool use in Responses API |
| Computer Use | Anthropic | 2024–2026 | Official docs | GUI agent capabilities |
| Code Execution Tool | Anthropic | 2026 | Official docs | Sandboxed code execution for agents |

### MCP Ecosystem
| Source | Org | Year | Type | Why Authoritative |
|--------|-----|------|------|-------------------|
| MCP Specification (v2025-11-25) | Anthropic / LF Projects | 2024–2025 | Formal spec | Canonical MCP protocol definition |
| MCP Introduction & Concepts | Anthropic | 2024–2026 | Official docs | MCP architecture and concepts |
| MCP Servers Repository | Community | 2025– | GitHub repo | Reference MCP server implementations |
| MCP Authorization Spec | LF Projects | 2025 | Formal spec | OAuth 2.0 authorization for MCP |

### Agent-to-Agent
| Source | Org | Year | Type | Why Authoritative |
|--------|-----|------|------|-------------------|
| Agent-to-Agent Protocol (A2A) | Google | 2025 | GitHub repo | Google's inter-agent communication standard |

### Multi-Agent Systems
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation | Wu et al. (Microsoft) | 2024 | COLM | Multi-agent conversation framework |
| MemGPT: Towards LLMs as Operating Systems | Packer et al. | 2023 | arXiv:2310.08560 | OS-like memory management for agents |
| Agentic Memory: Learning Unified Memory Management for LLM Agents | Yu et al. | 2026 | arXiv:2601.01885 | Unified STM + LTM memory policy |

### Agent Benchmarks
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| SWE-bench: Can Language Models Resolve Real-World GitHub Issues? | Jimenez et al. | 2024 | ICLR | Agentic coding benchmark; official leaderboard |
| GAIA: A Benchmark for General AI Assistants | Mialon et al. | 2024 | ICLR | General agent capability benchmark |
| WebArena: A Realistic Web Environment for Building Autonomous Agents | Zhou et al. | 2024 | arXiv | Web-based agent benchmark |
| τ-bench: A Benchmark for Tool-Agent-User Interaction | Yao et al. | 2024 | arXiv | Tool + user interaction benchmark |

### Agent Frameworks (official docs)
| Source | Org | Year | Type | Why Authoritative |
|--------|-----|------|------|-------------------|
| LangGraph | LangChain | 2024–2026 | Official docs | Stateful multi-actor agent framework |
| CrewAI | CrewAI | 2024–2026 | Official docs | Role-based multi-agent orchestration |

---

## 7. RAG Pipelines, Anti-Hallucination Loops

### RAG Foundations
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks | Lewis et al. (Meta) | 2020 | NeurIPS | Original end-to-end RAG; 4000+ citations |
| Retrieval-Augmented Generation for LLMs: A Survey | Gao et al. | 2024 | arXiv:2312.10997 | Comprehensive RAG survey (2024) |
| Precise Zero-Shot Dense Retrieval without Relevance Labels (HyDE) | Gao et al. | 2023 | ACL | Hypothesis-based retrieval |
| Self-RAG: Learning to Retrieve, Generate, and Critique | Asai et al. | 2024 | ICLR | Self-reflective retrieval |
| Corrective Retrieval Augmented Generation (CRAG) | Yan et al. | 2024 | arXiv:2401.15884 | Retrieval quality evaluator + correction |

### Advanced Retrieval
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| From Local to Global: A Graph RAG Approach (GraphRAG) | Edge et al. (Microsoft) | 2024 | Microsoft Research | Graph-based RAG; enterprise adoption |
| RAPTOR: Recursive Abstractive Processing for Tree-Organized Retrieval | Sarthi et al. | 2024 | ICLR | Hierarchical summarization for retrieval |
| Introducing Contextual Retrieval | Anthropic | 2024 | Anthropic Blog | Context-enhanced chunking |
| ColPali: Efficient Document Retrieval with Vision Language Models | Faysse et al. | 2024 | ICLR 2025 | Vision-based document retrieval |
| Matryoshka Representation Learning | Kusupati et al. | 2022 | NeurIPS | Variable-size embeddings for flexible retrieval |

### RAG Evaluation
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| RAGAS: Automated Evaluation of Retrieval Augmented Generation | Es et al. | 2024 | EACL | Standard RAG eval framework |
| RAGChecker: A Fine-grained Framework for Diagnosing RAG | Ru et al. | 2024 | arXiv:2408.08067 | Fine-grained RAG diagnostics |
| ClashEval: Quantifying the tug-of-war between LLM prior and evidence | Wu, Wu, Zou | 2025 | arXiv:2404.10198 | Internal prior vs. external evidence conflict |

### Anti-Hallucination
| Source | Authors/Org | Year | Venue | Why Authoritative |
|--------|-------------|------|-------|-------------------|
| Chain-of-Verification Reduces Hallucination | Dhuliawala et al. | 2023 | arXiv:2309.11495 | CoVe: generate → verify → revise |
| NeMo Guardrails: Programmable Rails for Safe LLM Applications | Rebedea et al. (NVIDIA) | 2023 | arXiv:2310.10501 | Guardrails framework |
| Reducing Hallucination — Best Practices | Anthropic | 2024–2026 | Official docs | Provider guidance on hallucination reduction |
| LangSmith Tracing & Evaluation | LangChain | 2024–2026 | Official docs | Trace-based hallucination detection |
| Arize Phoenix: LLM Observability | Arize AI | 2025–2026 | Official docs | Open-source hallucination monitoring |
| W&B Weave: LLM Monitoring | Weights & Biases | 2025–2026 | Official docs | ML monitoring for LLM quality |

---

## 8. LLM Evaluation (Evals, Benchmarks)

### LLM-as-Judge
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena | Zheng et al. | 2023 | NeurIPS | Defines LLM-as-Judge methodology |
| G-Eval: NLG Evaluation using GPT-4 with Better Human Alignment | Liu et al. | 2023 | arXiv:2303.16634 | Chain-of-Thought evaluation with GPT-4 |
| Large Language Models are not Fair Evaluators | Wang et al. | 2023 | arXiv:2305.17926 | Position bias in LLM evaluation |
| Who Validates the Validators? | Shankar et al. | 2024 | arXiv:2404.12272 | Meta-evaluation of LLM judges |
| Self-Generated Critiques Boost Reward Modeling | Yu et al. | 2025 | NAACL | Self-critique as reward signal |
| Chatbot Arena: Open Platform for Evaluating LLMs by Human Preference | Chiang et al. | 2024 | arXiv:2403.04132 | LMSys Arena methodology |
| Evalica: Reliable, Reproducible Leaderboards | Ustalov | 2025 | COLING | Statistical significance in leaderboard ranking |

### Key Benchmarks
| Source | Authors/Org | Year | Venue | Why Authoritative |
|--------|-------------|------|-------|-------------------|
| MMLU: Measuring Massive Multitask Language Understanding | Hendrycks et al. | 2021 | ICLR | Most-cited LLM knowledge benchmark |
| BIG-bench: Beyond the Imitation Game | Srivastava et al. | 2022 | arXiv | 204-task collaborative benchmark |
| GSM8K: Training Verifiers to Solve Math Word Problems | Cobbe et al. (OpenAI) | 2021 | NeurIPS | Standard grade-school math benchmark |
| MATH: Measuring Mathematical Problem Solving | Hendrycks et al. | 2021 | NeurIPS | Competition-level math benchmark |
| HumanEval: Evaluating LLMs Trained on Code | Chen et al. (OpenAI) | 2021 | arXiv | Code generation benchmark |
| SWE-bench | Jimenez et al. | 2024 | ICLR | Real-world GitHub issue resolution |
| GAIA | Mialon et al. | 2024 | ICLR | General AI assistant benchmark |
| TruthfulQA | Lin et al. | 2022 | ACL | Truthfulness benchmark |
| HellaSwag | Zellers et al. | 2019 | ACL | Commonsense reasoning |

### Evaluation Frameworks
| Source | Org | Year | Type | Why Authoritative |
|--------|-----|------|------|-------------------|
| DeepEval | Confident AI | 2024–2026 | GitHub repo | Unit-testing framework for LLM outputs |
| Promptfoo | Promptfoo | 2024–2026 | GitHub repo | Prompt eval + red-teaming CLI |
| RAGAS | VibrantLabs | 2024–2026 | GitHub repo | RAG-specific evaluation |
| Braintrust Autoevals | Braintrust | 2024–2026 | GitHub repo | Auto-evaluation for LLM outputs |
| LangSmith | LangChain | 2024–2026 | Official docs | Trace-based eval pipeline |

---

## 9. LLM Security (Prompt Injection, Jailbreaks, OWASP)

### Standards & Taxonomies
| Source | Org | Year | Type | Why Authoritative |
|--------|-----|------|------|-------------------|
| OWASP Top 10 for LLM Applications 2025 | OWASP | 2025 | Formal standard | Industry standard vulnerability taxonomy |
| MITRE ATLAS (Adversarial Threat Landscape for AI Systems) | MITRE | 2024–2026 | Formal framework | AI threat framework; MITRE ATT&CK companion |

### Prompt Injection
| Source | Authors/Org | Year | Venue | Why Authoritative |
|--------|-------------|------|-------|-------------------|
| Not what you've signed up for: Indirect Prompt Injection | Greshake et al. | 2023 | arXiv:2302.12173 | Original indirect prompt injection paper |
| Prompt Injection Attacks and Defenses in LLM-Integrated Applications | Liu et al. | 2024 | arXiv:2402.02393 | Comprehensive injection survey |
| Ignore Previous Prompt: Attack Techniques For Language Models | Schulhoff et al. | 2023 | NeurIPS (SoLaR) | Prompt injection taxonomy |

### Jailbreaks
| Source | Authors/Org | Year | Venue | Why Authoritative |
|--------|-------------|------|-------|-------------------|
| Universal and Transferable Adversarial Attacks (GCG) | Zou et al. | 2023 | arXiv:2307.15043 | Greedy Coordinate Gradient — universal suffix attack |
| AutoDAN: Generating Stealthy Jailbreak Prompts | Liu et al. | 2023 | arXiv:2310.04451 | Automated genetic jailbreak generation |
| Many-shot jailbreaking | Anthropic | 2024 | anthropic.com/research | Long-context jailbreak technique |
| The Crescendo Multi-Turn LLM Jailbreak Attack | Russinovich, Salem, Eldan (Microsoft) | 2024 | arXiv:2404.01833 | Multi-turn escalation jailbreak |
| Mitigating Skeleton Key | Microsoft | 2024 | Security blog | Skeleton key jailbreak technique + mitigation |
| Jailbreaking Leading Safety-Aligned LLMs with Simple Adaptive Attacks | Andriushchenko et al. | 2024 | ICLR 2025 | Adaptive jailbreak; SOTA attack method |

### Red-Teaming
| Source | Authors/Org | Year | Venue | Why Authoritative |
|--------|-------------|------|-------|-------------------|
| Red Teaming Language Models with Language Models | Perez et al. (Anthropic) | 2022 | arXiv:2202.03286 | Automated LLM red-teaming |
| NeMo Guardrails: Programmable Rails | Rebedea et al. (NVIDIA) | 2023 | arXiv:2310.10501 | Guardrails for safe LLM apps |
| PyRIT (Python Risk Identification Toolkit) | Microsoft | 2024–2026 | GitHub repo | Automated red-teaming framework |

### Provider Safety Docs
| Source | Org | Year | Type | Why Authoritative |
|--------|-----|------|------|-------------------|
| Safety & Security documentation | Anthropic | 2024–2026 | Official docs | Claude safety practices |
| GPT-5.4 System Card | OpenAI | 2026 | Official report | Safety evaluation for GPT-5.4 |
| Gemini Safety documentation | Google | 2025–2026 | Official docs | Gemini safety features |

---

## 10. AI-Friendly Code Architecture

### Foundational Principles
| Source | Authors | Year | Type | Why Authoritative |
|--------|---------|------|------|-------------------|
| Clean Code | Martin | 2008 | Book | Industry-standard code quality principles |
| Refactoring, 2nd Ed. | Fowler | 2018 | Book | Refactoring patterns; code improvement |

### LLM-Aware Architecture
| Source | Authors/Org | Year | Type | Why Authoritative |
|--------|-------------|------|------|-------------------|
| AGENTS.md: A standard for AI agent project configuration | Anthropic | 2026 | Official spec | Standard for AI-readable project config |
| Rules for AI | Cursor | 2025–2026 | Official docs | Cursor rules for LLM-guided coding |
| Best practices for agentic coding | Anthropic | 2025 | Official docs | Coding patterns for AI agent workflows |
| copilot-instructions.md | GitHub | 2026 | Official docs | Copilot coding agent customization |

### Code Generation & Understanding
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Enriching Source Code with Contextual Data for Code Completion | van Dam, Izadi, van Deursen | 2023 | MSR | Impact of comments/code structure on completion |
| Dataflow-Guided Retrieval Augmentation for Repository-Level Code Completion | Cheng, Wu, Hu | 2024 | ACL | Repo context graph for code completion |
| SpecRover: Code Intent Extraction via LLMs | Ruan, Zhang, Roychoudhury | 2024 | ICSE 2025 | Extracting developer intent from code |
| Revisiting the Role of Natural Language Code Comments | Gupta et al. | 2026 | arXiv:2601.16661 | Comments vs. LLM code understanding |
| Less is more: DocString compression in code generation | Yang et al. | 2026 | ACM TOSEM | DocString impact on generation quality |
| Beyond the Black Box: Survey on Theory and Mechanism of LLMs | Gan et al. | 2026 | arXiv:2601.02907 | Meta-survey covering inference-stage practices |

### Prompt Pipeline Optimization
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| DSPy: Compiling Declarative Language Model Calls | Khattab et al. | 2024 | ICLR | Optimize LLM pipelines by metric, not manual tuning |
| Rethinking the Role of Demonstrations | Min et al. | 2022 | EMNLP | What makes in-context learning work |
| Calibrate Before Use | Zhao et al. | 2021 | ICML | LLM calibration for better performance |
| Large Language Models as Optimizers | Yang et al. | 2024 | ICLR | Iterative prompt optimization |

---

## 11. Observability (OpenTelemetry, SLO)

### Standards & Specifications
| Source | Org | Year | Type | Why Authoritative |
|--------|-----|------|------|-------------------|
| Semantic Conventions for Generative AI (v1.40.0) | OpenTelemetry / CNCF | 2026 | Formal spec | Industry-standard GenAI observability conventions |
| OpenTelemetry Specification | CNCF | 2019–2026 | Formal spec | Distributed tracing and metrics standard |

### LLM-Specific Tools
| Source | Org | Year | Type | Why Authoritative |
|--------|-----|------|------|-------------------|
| OpenLLMetry | Traceloop | 2025–2026 | GitHub repo | OpenTelemetry-native LLM observability |
| Arize Phoenix | Arize AI | 2025–2026 | GitHub repo | Open-source LLM observability platform |
| LangSmith | LangChain | 2024–2026 | Official docs | Trace-based LLM observability + eval |
| W&B Weave | Weights & Biases | 2025–2026 | Official docs | LLM monitoring and evaluation |

### SLO & Reliability
| Source | Authors | Year | Type | Why Authoritative |
|--------|---------|------|------|-------------------|
| Site Reliability Engineering | Beyer et al. (Google) | 2016 | Book (O'Reilly) | SLI/SLO/Error Budget canon |
| LLMs Get Lost In Multi-Turn Conversation | Laban et al. | 2025 | ICLR 2026 | Multi-turn reliability degradation |
| The Molecular Structure of Thought | Chen et al. | 2026 | arXiv:2601.06002 | Typed traces: deep reasoning, self-reflection, self-exploration |

---

## 12. Multimodal Systems

### Vision-Language
| Source | Authors/Org | Year | Venue | Why Authoritative |
|--------|-------------|------|-------|-------------------|
| CLIP: Learning Transferable Visual Models From Natural Language Supervision | Radford et al. (OpenAI) | 2021 | ICML | Contrastive vision-language pre-training; 15000+ citations |
| Flamingo: a Visual Language Model for Few-Shot Learning | Alayrac et al. (DeepMind) | 2022 | NeurIPS | Few-shot multimodal model |
| LLaVA: Visual Instruction Tuning | Liu et al. | 2024 | NeurIPS | Instruction-tuned vision-language model |
| Visual Instruction Tuning (LLaVA 1.5) | Liu et al. | 2024 | arXiv | Improved multimodal instruction tuning |
| Gemini: A Family of Highly Capable Multimodal Models | Gemini Team (Google) | 2023 | arXiv | Native multimodal architecture |
| Gemini 1.5: Unlocking multimodal understanding across millions of tokens | Reid et al. (Google) | 2024 | arXiv | Ultra-long multimodal context |
| GPT-4V(ision) System Card | OpenAI | 2023 | Official report | GPT-4 vision capabilities |
| Vision API docs | OpenAI | 2024–2026 | Official docs | OpenAI vision capabilities |
| Vision API docs | Anthropic | 2024–2026 | Official docs | Claude vision capabilities |
| Image Understanding docs | Google | 2025–2026 | Official docs | Gemini vision capabilities |

### Audio & Speech
| Source | Org | Year | Type | Why Authoritative |
|--------|-----|------|------|-------------------|
| Audio Understanding | Google | 2025–2026 | Official docs | Gemini native audio processing |
| Realtime API | OpenAI | 2025–2026 | Official docs | Speech-to-speech realtime API |
| Whisper: Robust Speech Recognition | Radford et al. (OpenAI) | 2022 | OpenAI | Speech recognition model |

### Document Understanding
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| ColPali: Efficient Document Retrieval with Vision Language Models | Faysse et al. | 2024 | ICLR 2025 | Visual document retrieval via VLMs |
| DocTR | Mindee | 2023– | GitHub repo | Open-source OCR for documents |

### Multimodal Hallucination
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Evaluating Object Hallucination in Large Vision-Language Models (POPE) | Li et al. | 2023 | EMNLP | Vision-language hallucination benchmark |
| HallusionBench: An Advanced Diagnostic Suite | Guan et al. | 2024 | CVPR | Diagnostic vision-language hallucination |

---

## 13. Fine-Tuning, Post-Training (SFT, RLHF, DPO, LoRA)

### Supervised Fine-Tuning
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Finetuned Language Models Are Zero-Shot Learners (FLAN) | Wei et al. (Google) | 2021 | ICLR | Instruction tuning paradigm |
| Self-Instruct: Aligning LMs with Self-Generated Instructions | Wang et al. | 2022 | ACL | Synthetic instruction generation |
| Textbooks Are All You Need (Phi-1) | Gunasekar et al. (Microsoft) | 2023 | arXiv:2306.11644 | Quality > quantity for training data |
| Orca: Progressive Learning from Complex Explanation Traces | Mukherjee et al. (Microsoft) | 2023 | arXiv:2306.02707 | Learning from GPT-4 reasoning traces |
| LIMA: Less Is More for Alignment | Zhou et al. (Meta) | 2023 | arXiv:2305.11206 | 1000 examples enough for alignment |

### RLHF & Alignment
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Training language models to follow instructions with human feedback (InstructGPT) | Ouyang et al. (OpenAI) | 2022 | NeurIPS | Original RLHF for instruction following |
| Constitutional AI: Harmlessness from AI Feedback | Bai et al. (Anthropic) | 2022 | arXiv:2212.08073 | RLAIF — AI feedback instead of human |
| Scaling Laws for Reward Model Overoptimization | Gao et al. | 2022 | NeurIPS | Reward hacking in RLHF |
| DeepSeekMath: Pushing the Limits of Mathematical Reasoning | Shao et al. | 2024 | arXiv:2402.03300 | GRPO — Group Relative Policy Optimization |
| DeepSeek-R1: Incentivizing Reasoning via RL | Guo et al. | 2025 | Nature vol. 645 | Pure RL for reasoning emergence |

### DPO & Variants
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Direct Preference Optimization (DPO) | Rafailov et al. (Stanford) | 2023 | NeurIPS | RL-free preference optimization; 2000+ citations |
| ORPO: Monolithic Preference Optimization without Reference Model | Hong et al. | 2024 | arXiv:2403.07691 | Reference-free DPO alternative |
| KTO: Model Alignment as Prospect Theoretic Optimization | Ethayarajh et al. | 2024 | arXiv:2402.01306 | Kahneman-Tversky prospect theory alignment |
| SimPO: Simple Preference Optimization with Reference-Free Reward | Meng et al. | 2024 | arXiv:2405.14734 | Length-normalized preference optimization |
| RLHF Workflow: From Reward Modeling to Online RLHF | Dong et al. | 2024 | arXiv:2405.07863 | Online vs offline RLHF comparison |

### Parameter-Efficient Fine-Tuning
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| LoRA: Low-Rank Adaptation of Large Language Models | Hu et al. (Microsoft) | 2021 | ICLR | Dominant PEFT method; 10000+ citations |
| QLoRA: Efficient Finetuning of Quantized LLMs | Dettmers et al. | 2023 | NeurIPS | LoRA + 4-bit quantization |
| IA3: Few-shot Parameter-Efficient Fine-Tuning | Liu et al. | 2022 | ACL | Inference-time PEFT via learned vectors |

### Knowledge Distillation
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Distilling the Knowledge in a Neural Network | Hinton, Vinyals, Dean | 2015 | arXiv:1503.02531 | Original distillation framework |
| Distilling Step-by-Step! | Hsieh et al. | 2023 | ACL | Reasoning distillation via CoT |
| MiniLLM: On-Policy Distillation of LLMs | Gu et al. | 2024 | ICLR | Knowledge distillation for LLMs |

### Provider Fine-Tuning Docs
| Source | Org | Year | Type | Why Authoritative |
|--------|-----|------|------|-------------------|
| Model Optimization (fine-tuning) | OpenAI | 2024–2026 | Official docs | GPT fine-tuning API |
| Reinforcement Fine-Tuning | OpenAI | 2025–2026 | Official docs | RL-based fine-tuning |

---

## 14. Design Patterns for LLM Apps

### Agent Patterns (detailed in book chapters 8–10, 20)
| Source | Authors | Year | Venue | Pattern |
|--------|---------|------|-------|---------|
| Chain-of-Thought | Wei et al. | 2022 | NeurIPS | Reasoning via step-by-step |
| Self-Consistency | Wang et al. | 2023 | ICLR | Majority vote over multiple reasoning paths |
| ReAct | Yao et al. | 2023 | ICLR | Interleaved reasoning + action |
| Reflexion | Shinn et al. | 2023 | NeurIPS | Self-reflection + verbal RL |
| Tree of Thoughts | Yao et al. | 2023 | NeurIPS | Tree search over reasoning paths |
| Graph of Thoughts | Besta et al. | 2024 | AAAI | Graph-structured reasoning |
| Chain-of-Verification | Dhuliawala et al. | 2023 | arXiv:2309.11495 | Generate → verify → revise |
| RAG | Lewis et al. | 2020 | NeurIPS | Retrieve → augment → generate |
| Self-RAG | Asai et al. | 2024 | ICLR | Self-reflective retrieval |
| GraphRAG | Edge et al. | 2024 | Microsoft Research | Graph-based retrieval |

### Architectural Pattern Guides
| Source | Authors/Org | Year | Type | Why Authoritative |
|--------|-------------|------|------|-------------------|
| Building Effective Agents | Anthropic | 2024 | Anthropic Blog | Agent design patterns: workflows → agents |
| LLM Powered Autonomous Agents | Weng (OpenAI) | 2023 | Blog | Comprehensive agent design survey |
| Building Production RAG | LlamaIndex | 2024 | Official docs | Production RAG patterns |

---

## 15. Serving & Inference

### KV-Cache & Memory Management
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Efficient Memory Management for LLM Serving with PagedAttention | Kwon et al. (vLLM) | 2023 | SOSP | PagedAttention — OS-inspired KV-cache; standard in vLLM |
| vLLM Documentation | vLLM Project | 2023–2026 | Official docs | Production LLM serving; disaggregated prefill |
| SGLang: Efficient Execution of Structured Language Model Programs | Zheng et al. | 2024 | arXiv | RadixAttention — prefix-aware KV-cache reuse |
| Prompt Cache: Modular Attention Reuse for Low-Latency Inference | Gim et al. | 2024 | MLSys | Prompt-level KV-cache sharing |

### Speculative Decoding
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Fast Inference from Transformers via Speculative Decoding | Leviathan et al. (Google) | 2023 | ICML | Draft-then-verify; 2–3× speedup |
| Accelerating LLM Decoding with Speculative Sampling | Chen et al. (DeepMind) | 2023 | arXiv:2302.01318 | Independent speculative sampling variant |

### Quantization
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| GPTQ: Accurate Post-Training Quantization for GPTs | Frantar et al. | 2023 | ICLR | OBQ-based weight quantization |
| AWQ: Activation-aware Weight Quantization | Lin et al. | 2024 | MLSys | Activation-aware 4-bit quantization |
| SmoothQuant: Accurate and Efficient Post-Training Quantization | Xiao et al. | 2023 | ICML | Activation + weight quantization |
| QLoRA: Efficient Finetuning of Quantized LLMs | Dettmers et al. | 2023 | NeurIPS | 4-bit + PEFT |
| The Era of 1-bit LLMs (BitNet) | Ma et al. (Microsoft) | 2024 | arXiv:2402.17764 | 1.58-bit LLMs |

### Serving Frameworks
| Source | Org | Year | Type | Why Authoritative |
|--------|-----|------|------|-------------------|
| vLLM | vLLM Project (UC Berkeley) | 2023–2026 | GitHub repo | Most popular open-source LLM serving engine |
| TensorRT-LLM | NVIDIA | 2023–2026 | GitHub repo | NVIDIA-optimized inference |
| Ollama | Ollama | 2024–2026 | Official site | Local LLM serving for devs |
| SGLang | SGLang Project | 2024–2026 | GitHub repo | Structured generation serving |

### Test-Time Compute (Budget Forcing, Scaling)
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| Scaling LLM Test-Time Compute Optimally | Snell et al. (UC Berkeley, Google DeepMind) | 2024 | arXiv:2408.03314 | Adaptive test-time compute allocation |
| Scaling Test-Time Compute Without Verification or RL is Suboptimal | Setlur et al. | 2025 | arXiv:2506.14495 | Verifier necessity for test-time scaling |
| All Roads Lead to Likelihood: The Value of RL in Fine-Tuning | Swamy et al. | 2025 | arXiv:2505.14864 | Solution space constrained by verifier |
| To CoT or not to CoT? | Sprague et al. | 2025 | arXiv:2503.16411 | When CoT helps and when it doesn't |
| Chain-of-Thought Reasoning Without Prompting | Wang & Zhou | 2024 | arXiv:2402.10200 | CoT can happen internally, not just in output |

---

## 16. Agent Orchestration, Durable Execution

### Durable Execution
| Source | Authors/Org | Year | Type | Why Authoritative |
|--------|-------------|------|------|-------------------|
| Sagas | Garcia-Molina & Salem | 1987 | ACM SIGMOD | Foundational compensating transaction pattern |
| Temporal: Durable Execution | Temporal Technologies | 2020–2026 | Official docs | Durable execution platform |
| Inngest: Durable Functions for AI Workflows | Inngest | 2023–2026 | Official docs | Serverless durable execution for AI |
| ADK: Long-Running Tasks | Google | 2025 | Official docs | Google Agent Development Kit |

### Agent Lifecycle & Memory
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| MemGPT: Towards LLMs as Operating Systems | Packer et al. | 2023 | arXiv:2310.08560 | Virtual memory for agent context |
| Agentic Memory: Learned Unified Memory Management | Yu et al. | 2026 | arXiv:2601.01885 | Learned STM + LTM memory policy |
| Generative Agents: Interactive Simulacra | Park et al. | 2023 | UIST | Agent memory + reflection patterns |
| LLMs Get Lost In Multi-Turn Conversation | Laban et al. | 2025 | ICLR 2026 | Multi-turn drift; checkpoint summaries |
| Codex: Background Tasks & Sandbox Execution | OpenAI | 2026 | Official docs | Persistent agent execution |
| Claude Code: Headless & Background Mode | Anthropic | 2026 | Official docs | Background agent workflows |

### Multi-Agent Orchestration
| Source | Authors/Org | Year | Type | Why Authoritative |
|--------|-------------|------|------|-------------------|
| AutoGen | Wu et al. (Microsoft) | 2024 | COLM | Multi-agent conversation framework |
| CrewAI | CrewAI | 2024–2026 | Official docs | Role-based orchestration |
| LangGraph | LangChain | 2024–2026 | Official docs | Graph-based stateful agent orchestration |

---

## 17. LLM Landscape 2026

### Major Model Families (Official Docs)
| Source | Org | Year | Type | Why Authoritative |
|--------|-----|------|------|-------------------|
| GPT-5.4 System Card | OpenAI | 2026 | Official report | SOTA frontier model (May 2026) |
| Claude models overview (Claude Opus 4.6) | Anthropic | 2026 | Official docs | Frontier model with extended thinking |
| Claude Mythos Preview System Card | Anthropic | 2026 | Official report | Next-gen Claude (Project Glasswing) |
| Gemini API models | Google DeepMind | 2026 | Official docs | Gemini 3.x family |
| Gemma 4 | Google DeepMind | 2026 | Official docs | Open-weight Gemma family |
| Llama 4 | Meta AI | 2025–2026 | Official site | Meta open(-weight) model |
| DeepSeek models | DeepSeek-AI | 2024–2026 | GitHub + arXiv | DeepSeek-V3, DeepSeek-R1 |
| Qwen3 / Qwen3.5 / Qwen3.6 | Alibaba | 2025–2026 | GitHub + blog | Qwen open-weight family |
| GLM-5.1 | Zhipu AI / Z.ai | 2026 | Official site | Agentic-optimized model |
| Grok 4.20 | xAI | 2026 | Official docs | Grok family |
| MiniMax-M2.7 | MiniMax | 2026 | Official site | Self-evolution claims |

### Reasoning Models
| Source | Authors/Org | Year | Venue | Why Authoritative |
|--------|-------------|------|-------|-------------------|
| DeepSeek-R1 | Guo et al. / DeepSeek-AI | 2025 | Nature vol. 645 | Pure RL for reasoning emergence |
| o3 System Card | OpenAI | 2025 | Official report | Frontier reasoning model |
| GPT-5.4 System Card | OpenAI | 2026 | Official report | Reasoning + tool use integration |
| Qwen3.6-Plus: Towards Real World Agents | Qwen Team | 2026 | Research blog | Agentic reasoning capabilities |
| Training LLMs to Reason in a Continuous Latent Space (Coconut) | Hao et al. | 2025 | arXiv:2412.06769 | Continuous latent reasoning |

### MoE & Hybrid Architecture
| Source | Authors/Org | Year | Venue | Why Authoritative |
|--------|-------------|------|-------|-------------------|
| Mixtral of Experts | Mistral AI | 2024 | arXiv:2401.04088 | Production MoE (8×7B) |
| DeepSeek-V3 Technical Report | DeepSeek-AI | 2024 | arXiv:2412.19437 | 671B MoE production model |
| Jamba 2 | AI21 Labs | 2026 | Official blog | Hybrid SSM-Transformer-MoE, 256K |
| Transformers are SSMs (Mamba-2) | Dao & Gu | 2024 | ICML | Unified hybrid framework |

### SLM & Edge
| Source | Authors/Org | Year | Venue | Why Authoritative |
|--------|-------------|------|-------|-------------------|
| Phi-4 Technical Report | Microsoft | 2025 | arXiv | MSR's small language model family |
| SmolLM3 | HuggingFace | 2025 | HuggingFace | Open lightweight LM (3B) |
| Qwen3.5-0.8B | Alibaba / Qwen Team | 2026 | HuggingFace | Sub-1B capable model |
| The Era of 1-bit LLMs (BitNet) | Ma et al. (Microsoft) | 2024 | arXiv:2402.17764 | 1.58-bit extreme compression |

### Emerging: Diffusion LLM
| Source | Authors | Year | Venue | Why Authoritative |
|--------|---------|------|-------|-------------------|
| MDLM | Sahoo et al. | 2024 | NeurIPS | Masked diffusion for language |
| SEDD | Lou et al. | 2024 | ICML | Score entropy discrete diffusion |
| Mercury | Inception Labs | 2025–2026 | Official site | First production diffusion LLM |
| LLaDA | Nie et al. | 2025 | arXiv:2502.09992 | 8B diffusion matching LLaMA3 |

### MCP Ecosystem
| Source | Org | Year | Type | Why Authoritative |
|--------|-----|------|------|-------------------|
| MCP Spec (v2025-11-25) | LF Projects / Anthropic | 2024–2025 | Formal spec | Canonical protocol specification |
| MCP Connector (Anthropic) | Anthropic | 2025–2026 | Official docs | Native MCP client in Claude |
| Tools, Connectors, and MCP | OpenAI | 2025–2026 | Official docs | MCP support in Responses API |
| A2A Protocol | Google | 2025 | GitHub repo | Inter-agent protocol |

### Model Technical Reports (Official)
| Source | Authors/Org | Year | Type | Why Authoritative |
|--------|-------------|------|------|-------------------|
| GPT-4 Technical Report | OpenAI | 2023 | arXiv:2303.08774 | GPT-4 architecture overview |
| The Claude 3 Model Family: Opus, Sonnet, Haiku | Anthropic | 2024 | Anthropic | Claude 3 family |
| The Llama 3 Herd of Models | Meta AI | 2024 | arXiv:2407.21783 | Llama 3 family (8B, 70B, 405B) |
| Gemma 2: Improving Open Language Models at a Practical Size | Google DeepMind | 2024 | arXiv:2408.00118 | Gemma 2 family |
| GLM-5: from Vibe Coding to Agentic Engineering | Du et al. (Zhipu) | 2026 | arXiv:2602.15763 | GLM-5 technical report |

---

## Key Benchmark Official Pages

| Benchmark | Official URL | What It Measures |
|-----------|-------------|-----------------|
| LMSys Chatbot Arena | https://chat.lmsys.org/ | Human preference ELO ranking |
| MMLU | https://github.com/hendrycks/test | Multi-task language understanding |
| SWE-bench | https://www.swebench.com/ | Real GitHub issue resolution |
| GAIA | https://huggingface.co/gaia-benchmark | General AI assistant capabilities |
| HumanEval | https://github.com/openai/human-eval | Code generation |
| GSM8K | https://github.com/openai/grade-school-math | Grade-school math |
| MATH | https://github.com/hendrycks/math | Competition math |
| TruthfulQA | https://github.com/sylinrl/TruthfulQA | Truthfulness |
| BIG-bench | https://github.com/google/BIG-bench | 204 diverse tasks |
| MT-Bench | https://github.com/lm-sys/FastChat | Multi-turn conversation |
| HellaSwag | https://rowanzellers.com/hellaswag/ | Commonsense reasoning |
| SimpleQA | https://openai.com/index/simpleqa/ | Short-form factuality |
| HELMET | https://github.com/princeton-nlp/HELMET | Long-context evaluation |

---

## Notes

- **Venue abbreviations:** NeurIPS, ICML, ICLR, ACL, EMNLP, NAACL, AAAI, TACL, TMLR, SOSP, MLSys, COLM, MSR, CVPR, EACL, COLING, SIGMOD, SOSP.
- **Official docs** include Anthropic Platform docs, OpenAI API docs, Google AI docs, and official GitHub repos.
- **"Official blog"** refers only to research blogs from OpenAI, Anthropic, Google DeepMind, Meta AI, Microsoft Research — not personal blogs.
- Sources already cited in the book's chapter `## Источники` sections are all included above, with additions for completeness.
- For fact-checking: verify claims against the *most recent version* of official docs (Anthropic Platform, OpenAI API, Google AI) as they update frequently.