Founders Who Only Date Sequoia
YC founders want Sequoia or nothing. Sequoia wrote a $1 billion check to someone else.
Read the storyNoam Shazeer
AI researcher; co-founder, Character.AI
Noam Shazeer is a researcher whose work spans neural language models, efficient architectures and conversational AI. He co-authored the original Transformer paper, early sparsely gated mixture-of-experts work, T5 and Switch Transformers, and authored research on gated feed-forward layers. He also co-founded Character.AI with Daniel de Freitas, translating language-model research into a platform for interactive characters and personalized conversation.
Follow Noam Shazeer
Explore the coverage Get AI Secret’s free daily briefing, including Noam Shazeer and the wider AI industry.
At a glance
Noam Shazeer is a researcher whose work spans neural language models, efficient architectures and conversational AI. He co-authored the original Transformer paper, early sparsely gated mixture-of-experts work, T5 and Switch Transformers, and authored research on gated feed-forward layers. He also co-founded Character.AI with Daniel de Freitas, translating language-model research into a platform for interactive characters and personalized conversation.
- Known for
- Transformer; mixture-of-experts; T5; Switch Transformers
- Company founded
- Character.AI, with Daniel de Freitas
- Research focus
- Language-model architectures and computational efficiency
Sources checked . Selected career background and milestones.
Career, in context.
Selected research, leadership and career milestones, newest first.
Co-introduces Character.AI’s vision
Shazeer, Daniel de Freitas and their team describe Character.AI’s full-stack approach to personalized conversational systems, reporting a public beta that had launched in September.
Source: Character.AI: Introducing Character ↗Co-authors Switch Transformers
With William Fedus and Barret Zoph, Shazeer introduces a simplified sparse expert-routing approach, studying more efficient training and models with up to a trillion parameters.
Source: Fedus, Zoph and Shazeer: Switch Transformers ↗Publishes GLU Variants Improve Transformer
Shazeer evaluates gated linear-unit variants in Transformer feed-forward layers, showing that some alternatives improve model quality compared with commonly used ReLU or GELU activations.
Source: Shazeer: GLU Variants Improve Transformer ↗Co-authors the T5 study
Shazeer and collaborators introduce a unified text-to-text framework and compare transfer-learning methods across language tasks, releasing models, code and the Colossal Clean Crawled Corpus.
Source: Raffel et al.: Unified Text-to-Text Transformer ↗Co-authors Attention Is All You Need
Shazeer is one of eight authors introducing the Transformer, a sequence-modeling architecture built around attention mechanisms rather than recurrent or convolutional layers, evaluated on machine translation.
Source: Vaswani et al.: Attention Is All You Need ↗Leads authorship of sparse expert-layer paper
Shazeer and six collaborators introduce a sparsely gated mixture-of-experts layer, using a trainable gate to choose a subset of expert networks for each input.
Source: Shazeer et al.: Sparsely-Gated Mixture-of-Experts ↗
Show earlier historyHide earlier history
The latest on Noam Shazeer
Newest firstRelated reporting
Briefs, articles and relevant sections, newest first. Explore older coverage below.
More coverage
8 min read · Members
Character.ai Is Bankrupt
Silicon Valley's golden child didn't live to see its fifth birthday.
7 min read · Members
You Can’t Lock Up a Ghost
Plus: Google Lost Two Giants, Norway Walled Off Childhood From AI
6 min read · Members
A Billion Bodies
Plus: AI Took A Seat At The G7, GPUs Are The New Exit
