OpenAI announces GPT-2 and withholds the full model over misuse concerns
Confirmed
Importance: major (4 of 5)The takeaway
GPT-2, a 1.5B-parameter language model trained on 40GB of web text, generated strikingly coherent paragraphs; OpenAI initially released only smaller versions, citing misuse risk, and released the full model in November 2019.
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- Major (4 of 5)
- Last verified
- 29 September 2026
Your AI and this story
- GPT-6 AstraIn its training data
- Claude Opus 5.5In its training data
- Gemini 4 HydrogenIn its training data
- Grok 4.7In its training data
It happened before the cutoffs of all four assistants, so it can be in their training data.
Key facts
- 1.5 billion parameters
- Trained on WebText (~8M web pages, ~40GB)
- Staged release: full 1.5B model published 5 November 2019
- Paper: ‘Language Models are Unsupervised Multitask Learners’
What happened
OpenAI showed that scaling a language model produced zero-shot abilities on many tasks, and experimented with staged, responsible release.
Why it matters
First public glimpse of what scaling LLMs could do and the first major debate over whether to release model weights.
Sources
3 sources from 2 sites. Numbers match the chips in the text.
3 sources: 3 primary
Primary
- Better language models and their implications (OpenAI)openai.com, official
- GPT-2: 1.5B release (OpenAI)openai.com, official
- openai/gpt-2 (code)github.com, code
Changes
- Filed