GPT-2 is a 2019 OpenAI model, 1.5 billion parameters in size, trained with one goal: guess the next word of web text.
In plain words
GPT-2 took the original GPT and made it more than ten times bigger, training it on text from about 8 million web pages. With that single goal of guessing the next word, it learned to write long, fluent passages and began to handle tasks like translation and summarising without task-specific training. Concerned it could be misused, OpenAI first shared only a smaller version and released the full model in stages.
A quick example
Give GPT-2 the opening sentence of a news story and it continues writing several paragraphs in the same style. It can still go wrong: OpenAI saw it repeat itself, switch topic oddly, or describe a fire burning under water.
Where you'll meet it
- Research and experiments, using OpenAI's public GitHub repository.
Related
Sources
- DocsBetter language models and their implications, OpenAI · read 29 Sept 2026
- Docsgpt-2, OpenAI (GitHub) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.