Lab city, United States
Seattle and Redmond
Microsoft, Amazon and the Allen Institute for AI: Copilot, Phi, Amazon Q and the fully open OLMo.
The Seattle area combines two of the largest cloud companies with a strong research scene. Microsoft, in Redmond, backs OpenAI and builds Copilot and the small Phi models; Amazon, in Seattle, runs AWS and builds the Nova models. The Allen Institute for AI (Ai2), founded by Paul Allen in 2014, and the University of Washington release some of the most open models anywhere.
Microsoft and Amazon
Microsoft invested in OpenAI from 2019 and launched Bing Chat, later Copilot, in February 2023; early long conversations, in which the bot called itself Sydney, drew wide attention. Its researchers build the Phi series of small open models. Amazon offers many models through its Bedrock service, launched the Amazon Q assistant for businesses in November 2023 and its own Nova models in December 2024, and is a major investor in Anthropic.
Ai2 and the university
Ai2 released ELMo in 2018, an early model of word meaning in context, and the OLMo family from February 2024, publishing weights, training data, code and checkpoints together. Its Tülu recipes for post-training are open as well. The University of Washington's Paul G. Allen School produced QLoRA (2023), a method for fine-tuning large models on a single graphics card, and trains many of the field's researchers.
Models of Seattle and Redmond
Still to be written: Sydney (Microsoft), OLMo (language model).