OpenAI stops training AI models on the internet

OpenAI has announced that it will no longer use the live internet for training its AI models. This decision comes a few weeks after comments from an engineer at the company, who questioned the effectiveness of this approach. OpenAI points to the significant risks associated with training models on current internet data, which led to this drastic decision.

The decision to no longer train AI models on the live internet marks an important moment in the development of artificial intelligence. It allows OpenAI to have more control over the data used for training and to ensure the quality and safety of the models. According to the company, the unbounded and dynamic nature of internet data can lead to accidents or undesirable outcomes, which poses potential harm to users and society.

Moreover, OpenAI emphasizes the importance of ethics and responsibility in AI development. By basing training solely on carefully selected and controlled datasets, the company hopes to improve the accuracy and reliability of its models, while also minimizing the risks that arise from using uncontrolled information. This change may also have implications for other AI companies that employ similar methods.

It is still unclear how this change will affect OpenAI's future research and development landscape. The organization remains committed to innovation and safety within the AI sector, but the focus on controlled training will undoubtedly transform the way models are developed and implemented.

Read the full article from Techcrunch.