OpenAI Releases Full GPT-2 Text Model After Safety Concerns Fade
OpenAI makes its complete GPT-2 language model available to the public after finding no strong evidence of misuse. The AI generates convincing but flawed text, demonstrating both its capabilities and its limitations.
OpenAI releases the full version of its GPT-2 text-generating language model after withholding it for months due to safety concerns. The research lab initially fears that malicious actors could use the software to automatically produce fake news, spam, and phishing emails. However, OpenAI now states that it sees no strong evidence of misuse so far and makes the complete code freely available.
The AI model demonstrates its capabilities by actually writing portions of the news article about its own release. When fed a writing prompt, GPT-2 generates plausible-sounding but entirely fabricated information, such as falsely claiming that DeepMind creates the model or that the code powers popular video games like Dota 2.
Despite the alarming potential, the generated text contains obvious factual errors that a discerning reader easily identifies as false. The software shows that while artificial intelligence can quickly produce large volumes of coherent text, it still struggles with accuracy and remains susceptible to generating misinformation that could spread rapidly online.