GPT-2 and its staged release (Feb 2019)

What it was

On February 14, 2019, OpenAI announced GPT-2, a transformer language model with 1.5 billion parameters trained to predict the next word on 40GB of text from 8 million web pages linked from Reddit. The paper, "Language Models are Unsupervised Multitask Learners," listed Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei and Ilya Sutskever. Amodei, then OpenAI's research director, told The Guardian that the models "were 12 times bigger, and the dataset was 15 times bigger and much broader" than the previous state of the art.

The announcement said, "Due to our concerns about malicious applications of the technology, we are not releasing the trained model." OpenAI released a 117-million-parameter version instead. It then released larger versions in steps: 345 million parameters in May 2019, 774 million in August, and the full 1.5 billion on November 5, 2019, with a detection model and a report from partner researchers.

Read the original

What it changed

The staged release made the question of whether to publish a model's weights a matter of public argument within days. OpenAI said the process was meant as a test case for others; its November 2019 post said it had "continued with our original staged release plan in order to provide the community with a test case of a full staged release process." That post reported that Cornell researchers found people rated the 1.5-billion-parameter model's text 6.91 out of 10 for credibility and that the Middlebury Institute's CTEC had fine-tuned GPT-2 to generate extremist propaganda.

GPT-2 was the last language model whose weights OpenAI published until gpt-oss in August 2025. Its successor GPT-3 (May 2020) was offered only through an API, with no weights released.

The arguments it moved

Who should have access to powerful models?

Within a week, researchers were disputing the decision in public. Hugh Zhang wrote in The Gradient on February 19, 2019 that "Withholding the full GPT-2 model is both unnecessary for safety reasons and detrimental to future progress in AI." Jack Clark, OpenAI's head of policy, defended it to The Guardian on February 14: "We need to perform experimentation to find out what they can and can't do." The episode stayed a reference point. Yann LeCun posted in January 2024, "Remember when GPT-2 was deemed too dangerous to release? That was 5 years ago. The world didn't end," and on September 13, 2026, during the argument over Amodei's call to pace the frontier, he posted that "Dario was already claiming that GPT2 was too dangerous to open source back in 2019. I made fun of them then."

Compare every leader's position on this

Positions it bears on

Sam Altman, Open versus closed models

He has conceded that OpenAI's closed approach put it on the wrong side of history and has released open-weight models, while keeping frontier models proprietary.

GPT-2 was the last OpenAI language model with released weights until gpt-oss in 2025, the gap Altman called being "on the wrong side of history."

Critics and counterpoints
Dario Amodei, Open-weight models

He does not seek a ban on open-weight models, which he calls a public good when they lack dangerous capabilities, but says they are harder to safeguard and cannot be withdrawn once released.

Amodei co-wrote the GPT-2 paper and led research at OpenAI during the staged release; his later view that released weights cannot be withdrawn is the same concern.

Critics and counterpoints
Yann LeCun, Open vs closed models

Insists that foundation models must be open source so that no handful of companies controls the information people receive, and sees the concentration of AI in proprietary systems as a greater danger than the technology itself.

LeCun has cited GPT-2's withholding since 2019 as the example of misplaced caution about publishing models.

Critics and counterpoints
Ilya Sutskever, Open versus closed models

He argues that once models are powerful enough to cause great harm, publishing their details or weights stops making sense, and has called OpenAI's early openness a mistake.

Sutskever, the paper's senior author with Amodei, said in March 2023 that OpenAI's earlier openness had been a mistake and that open-sourcing powerful AI would come to look "just not wise."

Critics and counterpoints

Sources

Last verified 2026-09-22.