Whitepapers & Public Documents

The landmark Transformer architecture paper by Vaswani et al. Introduced the self-attention mechanism that replaced RNNs and became the foundation for all modern LLMs including GPT, BERT, and T5.

Comprehensive annual report tracking AI development, adoption, research output, investment trends, and policy across 100+ countries. Includes AI's environmental impact and public opinion data.

Annual deep-dive into AI research, industry developments, politics, and safety by Nathan Benaich and Ian Hogarth. Widely cited by researchers and investors for its insight into AI trends.

Official NIST publication of the AI Risk Management Framework. Provides voluntary guidance organized around GOVERN, MAP, MEASURE, and MANAGE functions for trustworthy AI.

Official consolidated text of the EU Artificial Intelligence Act (Regulation EU 2024/1689). The world's first comprehensive AI law, 459 articles covering risk-based AI regulation.

OpenAI's technical report on GPT-4 covering capabilities, limitations, safety evaluations, and the training approach. Notably absent of architecture details but rich in benchmark data.

Microsoft Research's 155-page analysis of GPT-4's capabilities across diverse tasks. Argues GPT-4 exhibits early sparks of AGI, sparking significant debate in the AI community.

Anthropic's paper introducing Constitutional AI — a method for training safer AI using AI feedback rather than human feedback for the harmlessness component. Foundation of Claude's safety approach.

McKinsey's analysis of generative AI's potential to add $2.6-4.4 trillion annually across 63 use cases. Covers automation potential by sector and workforce transition implications.

Kaplan et al.'s foundational paper establishing power-law scaling relationships between model size, dataset size, compute, and language model performance. Justification for the race to larger models.

Hoffmann et al.'s paper showing that most LLMs were significantly undertrained relative to compute budget. The Chinchilla optimal training laws reshaped how frontier labs scale models.

Annual critical analysis of AI industry power concentration, labor impacts, and the gap between AI promises and harms. Includes policy recommendations and advocacy positions.

The OECD AI Principles provide recommendations for responsible stewardship of trustworthy AI adopted by 44+ countries. Covers transparency, robustness, accountability, and values-aligned AI.

Gartner's annual assessment of AI technology maturity, adoption, and business value. Identifies which AI technologies are at peak hype vs. productive deployment. Essential for technology investment decisions.

Technical report for Llama 2 and Llama 2-Chat models (7B to 70B parameters). Covers the RLHF training process, safety evaluations, and red-teaming methodology. Most cited open-source LLM paper.