Researcher

Mike Lewis

Mike Lewis is an AI researcher whose work spans negotiation dialogue, language-model pretraining, retrieval, and alignment. His PhD research at Edinburgh explored symbolic and distributed semantics, and he won an EMNLP Best Paper award in 2016. His papers include RoBERTa, BART, retrieval-augmented generation, LIMA, The Llama 3 Herd of Models, and Context Language Models. As of October 2026, he is a Research Scientist at Meta FAIR.

Timeline

  1. 2016

    Edinburgh semantics and an EMNLP award

    His PhD research at Edinburgh explores symbolic and distributed semantics. In 2016, he wins an EMNLP Best Paper award.

  2. 2017

    Negotiation dialogue

    He coauthors Deal or No Deal? in 2017, an end-to-end model that learns negotiation without annotated dialogue states and handles both language and reasoning.

  3. 2019

    RoBERTa revisits BERT

    In 2019, he coauthors RoBERTa, a replication study of BERT pretraining that measures how hyperparameters and training-data size affect performance.

  4. 2020

    BART and retrieval-augmented generation

    In 2020, he coauthors BART, a denoising autoencoder that corrupts text and learns to reconstruct it for sequence-to-sequence generation and comprehension. He also coauthors retrieval-augmented generation (RAG), which combines a pretrained sequence-to-sequence model with a retriever over a Wikipedia vector index to improve open-domain question answering.

  5. 2023

    LIMA alignment training

    At NeurIPS in 2023, he coauthors LIMA, a 65-billion-parameter LLaMa model fine-tuned with 1,000 curated prompts and responses, without reinforcement learning or human preference modeling.

  6. 2024

    The Llama 3 models

    In 2024, he coauthors The Llama 3 Herd of Models, which presents a family of foundation models for multilingual use, coding, reasoning, and tool use.

  7. September 2026

    Context Language Models

    In September 2026, he coauthors Context Language Models, which proposes language models that manage their own context by treating it as a file the model can update.

  8. As of October 2026

    Research Scientist at Meta FAIR

    As of October 2026, he is a Research Scientist at Meta FAIR.

Sources and credits

Photo: We have no published photo of Mike Lewis yet. Send us a free-licensed photo.

  1. The Researchers | Next Play nextplay.so. Checked 21 October 2026.
  2. Deal or No Deal? End-to-End Learning for Negotiation Dialogues arxiv.org. Checked 21 October 2026.
  3. RoBERTa: A Robustly Optimized BERT Pretraining Approach arxiv.org. Checked 21 October 2026.
  4. BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension aclanthology.org. Checked 21 October 2026.
  5. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks papers.neurips.cc. Checked 21 October 2026.
  6. LIMA: Less Is More for Alignment papers.nips.cc. Checked 21 October 2026.
  7. The Llama 3 Herd of Models arxiv.org. Checked 21 October 2026.
  8. Context Language Models arxiv.org. Checked 21 October 2026.