OpenAI Holds Back New AI Text Generator Over Misuse Fears
OpenAI creates a highly convincing text generation AI but refuses to release the full model to the public due to concerns about malicious abuse.
Researchers at the Elon Musk-backed non-profit OpenAI develop an advanced text generation software that exceeds their initial expectations. The AI system, trained on 40GB of data from 8 million web pages, predicts the next word in a sentence so accurately that it produces highly convincing, human-like writing based on short prompts.
Because of its potential for misuse, OpenAI deliberately breaks from its standard practice and withholds the fully trained model from the public. The organization only releases a smaller version for researchers to test, sharing Musk's belief that artificial intelligence poses a significant existential threat if it falls into the wrong hands.
The software struggles with highly technical subjects but generates realistic conversational text about half of the time. In one striking test, it takes a two-sentence prompt about unicorns living in the Andes Mountains and effortlessly writes a believable nine-paragraph news story that perfectly mimics human journalism.