THE LANGUAGE MACHINES

• Published in: Nature (London) 2021-03, Vol.591 (7848), p.22-25
• Main Author: Hutson, Matthew
• Format: Article
• Language: English
• Subjects: Artificial intelligence; Chatbots; Customer services; Essays; Labeling; Language; Mathematical functions; Mathematics; Neural networks; Power; Scientists; Sentences; Speech recognition; Training
• ISSN: 0028-0836; 1476-4687
• DOI: 10.1038/d41586-021-00530-0

(OpenAI is legally a non-profit company, but in 2019 it created a for-profit subentity called OpenAI LP and partnered with Microsoft, which invested a reported US$1 billion in the firm.) Developers are now testing GPT-3's ability to summarize legal documents, suggest answers to customer-service enquiries, propose computer code, run textbased role-playing games or even identify at-risk individuals in a peer-support community by labelling posts as cries for help. [...]just like smaller chatbots, it can spew hate speech and generate racist and sexist stereotypes, if prompted - faithfully reflecting the associations in its training data. Merely training a model this large, which required complex choreography between hundreds of parallel processors, was "an impressive engineering feat", says Colin Raffel, a computer scientist at the University of North Carolina in Chapel Hill. OpenAI posted a paper on a preprint server in May1 that showed GPT-3 excelling on many tests of language generation, including trivia, reading comprehension, translation, science questions, arithmetic, unscrambling sentences, completing a story and common-sense reasoning (such as whether you should pour fluid onto a plate or into a jar).