Lists › Sibling sets

NLP models named after Sesame Street

A sibling set for 6 agents. Natural-language-processing models whose builders, across at least four institutions, kept naming them after Sesame Street characters as a running in-joke, starting with Elmo in 2017. Four employers with nothing else in common, one running joke none of them will be first to stop. GPT-2 was nearly Snuffleupagus.

1. 1ElmoEmbeddings from Language Models; Allen Institute for AI, 2017, the model that started the trend
2. 2BertBidirectional Encoder Representations from Transformers; Google, 2018, named in explicit homage to Elmo
3. 3ErnieEnhanced Representation through kNowledge IntEgration; Baidu, April 2019
4. 4GroverGenerating aRticles by Only Viewing mEtadata Records; Allen Institute for AI and the University of Washington, June 2019, built to write fake news well enough to learn to catch it
5. 5KermitGenerative Insertion-Based Modeling for Sequences; a Google-affiliated team, June 2019
6. 6Big BirdGoogle Research, 2020, a sparse-attention transformer built for sequences too long for a standard model

Load this set onto the wheel

Order is chronological by release: Elmo, Bert, Ernie, Grover, Kermit, Big Bird. Source: Inside AI's Muppet Empire (DeepLearning.AI, The Batch), linked.