OpenAI Slowly Releases Powerful Fake News AI After Staged Rollout Experiment
OpenAI is gradually releasing larger versions of its deceptive text generator GPT-2 after a six-month phased rollout. A new report shows this cautious approach successfully helps researchers anticipate and study the AI's malicious applications.
OpenAI releases a larger version of its controversial text-generating AI, GPT-2, marking the latest step in a carefully staged rollout. The San Francisco lab initially withholds the full model in February due to its impressive ability to produce convincing fake news, sparking widespread debate over whether the decision is a genuine safety precaution or a publicity stunt.
The organization now shares a version of the model that is half the size of the original, along with a new report detailing the results of this incremental release strategy. OpenAI partners with selected research institutions to study the full model's implications, finding that this staggered approach successfully helps the lab understand and anticipate potential malicious uses of the technology.
Collaborating researchers discover that readers frequently mistake GPT-2's automated outputs for genuine news articles, confirming previous speculations about the model's deceptive capabilities. Meanwhile, outside researchers actively develop new methods to detect machine-generated text as OpenAI continues to evaluate the best practices for handling high-stakes AI research.