Short answers, pulled from the story.
Researchers at Google published the paper titled Attention Is All You Need on the 12th of June 2017. This document introduced the transformer architecture that powers all GPT models today and replaced older recurrent neural network systems.
OpenAI released GPT-1 with 117 million parameters on the 11th of June 2018. They launched GPT-2 with 1.5 billion parameters on the 14th of February 2019, followed by GPT-3 containing 175 billion parameters later in 2020. ChatGPT arrived using the GPT-3.5 architecture on the 30th of November 2022 before transitioning to GPT-4 on the 14th of March 2023.
Early versions of these models focused exclusively on generating text based on written input. Recent iterations like GPT-4o now process and generate images alongside audio files while models such as o3 allocate additional computation time to analyze complex questions.
Companies adapted these foundations into specialized tools including EinsteinGPT for sales teams, BloombergGPT for financial professionals, Khanmigo for student tutoring, and SlackGPT for instant-messaging discussions. Google Workspace offers add-ons such as GPT for Sheets and Docs to enhance spreadsheet functionality.
The United States Patent and Trademark Office declined to expedite handling of OpenAI's application in April 2023 and stated that GPT was both descriptive and generic by May 2023. The European Union Intellectual Property Office registered GPT as a trademark in spring 2023 but faced challenges by spring 2024 while Switzerland also granted registration in spring 2023 before similar disputes emerged globally.