From Thought to Answer · Sources
Sources (38)
The research, documentation and reporting behind each step. Links checked September 2026; help-center and documentation pages change over time.
Step 1 · Your prompt
- Anthropic — System prompts (release notes) (Claude Platform Docs)
- OpenAI — Memory FAQ (OpenAI Help Center)
Step 2 · Tokens
- Petrov, La Malfa, Torr & Bibi — Language Model Tokenizers Introduce Unfairness Between Languages (NeurIPS 2023)
- OpenAI — tiktoken — GPT-4o uses the o200k_base tokenizer (GitHub)
Step 3 · The map
- Mikolov, Chen, Corrado & Dean — Efficient Estimation of Word Representations in Vector Space (arXiv, 2013)
- Nissim, van Noord & van der Goot — Fair Is Better than Sensational: Man Is to Doctor as Woman Is to Doctor (Computational Linguistics, 2020)
- Mrkšić et al. — Counter-fitting Word Vectors to Linguistic Constraints (NAACL 2016)
Step 4 · Attention
- Vaswani et al. — Attention Is All You Need (NeurIPS 2017)
- Llama Team, AI @ Meta — The Llama 3 Herd of Models (arXiv, 2024 — layer counts; also model sizes and training data in Step 7)
Step 5 · The dice
- Google — Prompt design strategies — temperature (Gemini API docs)
- He & Thinking Machines Lab — Defeating Nondeterminism in LLM Inference (2025)
Step 6 · The loop
- NASA Space Place — Why Is the Sky Blue? (NASA)
Step 7 · Training
- OpenAI — GPT-4 Technical Report (arXiv, 2023 — licensed data)
- Google for Developers — Gradient descent (Machine Learning Crash Course)
- Ouyang et al. — Training language models to follow instructions with human feedback (NeurIPS 2022)
- Bai et al. — Constitutional AI: Harmlessness from AI Feedback (arXiv, 2022)
- DeepSeek-AI — DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning (Nature, 2025)
Step 8 · Honest limits
- Kalai, Nachum, Vempala & Zhang — Why Language Models Hallucinate (OpenAI / arXiv, 2025)
- Hong, Troynikov & Huber — Context Rot: How Increasing Input Tokens Impacts LLM Performance (Chroma, 2025)
- Anthropic — How do usage and length limits work? (Claude Help Center)
- OpenAI — How your data is used to improve model performance (OpenAI Help Center)
- Anthropic — Updates to Consumer Terms and Privacy Policy (2025)
Step 9 · Prompt rules
- Anthropic — Prompting best practices (Claude Platform Docs)
- OpenAI — Best practices for prompt engineering with the OpenAI API (OpenAI Help Center)
- Liu et al. — Lost in the Middle: How Language Models Use Long Contexts (TACL, 2024)
- Hsieh et al. — Found in the Middle: Calibrating Positional Attention Bias Improves Long Context Utilization (Findings of ACL 2024)
- MacCallum & Lee — GPT-4.1 Prompting Guide (OpenAI Cookbook, 2025)
- Wei et al. — Chain-of-Thought Prompting Elicits Reasoning in Large Language Models (NeurIPS 2022)
- OpenAI — Reasoning best practices (OpenAI API docs)
- Chen et al. — Reasoning Models Don't Always Say What They Think (Anthropic, 2025)
Step 10 · Getting accurate answers
- OpenAI — Learning to reason with LLMs (2024)
- Farquhar, Kossen, Kuhn & Gal — Detecting hallucinations in large language models using semantic entropy (Nature, 2024)
- Manakul, Liusie & Gales — SelfCheckGPT: Zero-Resource Black-Box Hallucination Detection for Generative Large Language Models (EMNLP 2023)
- Anthropic — Reduce hallucinations (Claude Platform Docs)
- Xiong et al. — Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMs (ICLR 2024)
- Jaźwińska & Chandrasekar — AI Search Has a Citation Problem (Columbia Journalism Review, 2025)
- Sharma et al. — Towards Understanding Sycophancy in Language Models (ICLR 2024)
- Kim, Garg, Peng & Garg — Correlated Errors in Large Language Models (ICML 2025)