New text generator built by OpenAI considered too dangerous to release (2019) (techcrunch.com)

OpenAI has announced its latest language model, GPT-2, but has opted not to release it fully due to concerns over misuse. The model demonstrates remarkable capabilities in generating coherent and contextually relevant text by predicting the next word in a dataset of 40 gigabytes of internet text. While this advancement can enhance applications like conversational bots and improved speech recognition, OpenAI identified significant risks, including the potential for generating fake news and automated harassment on social media platforms. An example of its power is seen when GPT-2 contradicted a prompt about recycling, producing a lengthy argument against it with alarming conviction. The decision to limit the release of GPT-2 has sparked debate within the AI community, with critics accusing OpenAI of going against its founding principles of openness, while some applauded the ethical foresight it represents. As discussions around AI ethics grow, especially in light of recent warnings from government agencies about the societal risks posed by AI technologies, OpenAI has stated that it will revisit the decision in six months and encourages government initiatives to monitor the implications of AI advancements.