Large Language Models and Foundation Architectures

Part of the AI People directory: a structured reference to the individuals shaping artificial intelligence across research, industry, governance, ethics and public discourse.

Ashish Vaswani, Noam Shazeer, Jakob Uszkoreit, Llion Jones, Niki Parmar, Aidan Gomez, Illia Polosukhin and Lukasz Kaiser, transformer architecture, United States and Canada. The eight co-authors of Attention Is All You Need (2017), the paper that introduced the transformer architecture. This single innovation enabled GPT, BERT and virtually every modern language model. Several have since founded AI companies, including Shazeer (Character AI), Gomez (Cohere) and Polosukhin (NEAR Protocol). Key works: Attention Is All You Need (2017).

Alec Radford, language models, United States. Principal author of GPT-2 and a key contributor to GPT research at OpenAI. His work on unsupervised language-model pretraining demonstrated that scaling up simple models on large text corpora could produce surprisingly capable systems. Key works: the GPT-2 technical report; CLIP.

Jacob Devlin, pretraining methods, United States. Created BERT at Google, introducing bidirectional pretraining that dramatically improved NLP benchmarks. BERT's approach of pretraining on masked language modelling became standard practice. Key works: BERT: Pre-training of Deep Bidirectional Transformers (2018).

Christopher Manning, NLP foundations, United States. Co-authored foundational NLP textbooks and led Stanford's NLP group for decades, bridging linguistic theory and practical systems and building infrastructure used throughout the field. Key works: Foundations of Statistical Natural Language Processing (1999); Stanford NLP tools.

Tomáš Mikolov, word embeddings, Czech Republic. Created word2vec, demonstrating that neural networks could learn meaningful word representations capturing semantic relationships, reshaping how NLP systems represent meaning. Key works: the word2vec papers (2013).

Percy Liang, model evaluation and research infrastructure, United States. Leads Stanford's Center for Research on Foundation Models and created HELM, the most comprehensive evaluation framework for language models, providing essential infrastructure for understanding what foundation models can and cannot do. Key works: the HELM benchmark; foundation-model evaluation; CRFM leadership.

Andrej Karpathy, ML education and applied AI, United States. Former Tesla AI Director and OpenAI researcher whose educational content, from lectures to nanoGPT, has taught millions to understand deep learning, bridging frontier research and practical engineering with exceptional clarity. Key works: the CS231n lectures; nanoGPT.

Jason Wei, prompting methods and emergent abilities, United States. Pioneered chain-of-thought prompting, showing that asking language models to reason step-by-step dramatically improves performance on complex tasks, and his research on emergent abilities shaped understanding of what scaling enables. Key works: chain-of-thought prompting; emergent-abilities research.