OpenAI has released the largest version yet of its fake-news-spewing AI
OpenAI has released its latest version of GPT-2—and published a report that explains why it's doing it in this way.
The news: In February OpenAI unveiled a language model so good at generating fake news that the organization decided it was too dangerous to release the full thing. Now six months later, it has released a version of the model, known as GPT-2, that’s half the size of the full one. Alongside it, the policy team has published a report that examines the impact of releasing it in this way.
Bit by bit: In May, a few months after GPT-2’s initial debut, OpenAI decided it would carry out the staggered release of incrementally larger versions of the model. In February, it published a version of the model that was merely 8% of the size of the full one. It then published another version that is roughly a quarter of the size of the full one.
So, was it worth it? The accompanying report says that the decision to do it this way helped OpenAI better anticipate the possible malicious uses of GPT-2. But some experts argue that withholding the full model actually prevents effective countermeasures being developed.
See the full story here: https://www.technologyreview.com/s/614237/openai-released-its-fake-news-ai-gpt-2/?