GPT-2
OpenAI published a description of GPT-2.
Why it matters
It showed how a large Transformer could generate coherent continuations from prompts.
OpenAI’s publication also discusses release risks and staged sharing.
Atlas record: https://www.aitimeline.dev/en/events/evt-0013/
Research · February 2019
Event record
- Event date
- February 2019
- Timeline date
- Event date
- Verification
- Sources gathered automatically · September 11, 2026
- Lines
- ID
- evt-0013
Sources
- primary GPT-2
OpenAI · Published February 2019
- primary Language Models are Unsupervised Multitask Learners (Radford, Wu, Child, Luan, Amodei, Sutskever), OpenAI technical report
OpenAI
Related events
Earlier
- Related California: a bot must say it is a bot 2018links to this record
The law was passed five months before GPT-2 was described and is written about automated accounts, not language models; its disclosure duty also covers a bot running on such a model.
- Enables GPT-1 2018links to this record
The direct predecessor of GPT-2, eight months later.
- Builds on Transformer 2017
The GPT-2 technical description identifies Transformer-based language modeling.
- GPT-2 · OpenAI
Later
- Related Scaling laws 2020links to this record
The study measured models of the same architectural family.
- Extends GPT-3 2020links to this record
The same architecture, scaled up a hundredfold.
- Related ChatGPT 2022links to this record
The assistant rests on the GPT line of language models.
- Related gpt-oss 2025links to this record
The company's first open weights since GPT-2 in 2019.