Key Facts

• The next generation of AI models are designed to be trained by people who are paid to have conversations with them, a process that relies on human input to improve the models' language understanding and generation capabilities.
• Several workers responsible for training AI models have admitted to New Scientist that, instead of doing the task themselves, they use chatbots to complete the conversations required to train the models.
• This practice, referred to as "AI inbreeding," involves using AI models to generate training data for other AI models, which can lead to a degradation in the quality and diversity of the training data.
• Experts warn that "AI inbreeding" may reduce the power and usefulness of future AI models, as they are trained on data generated by other AI models rather than by humans interacting with them directly.
• The use of chatbots to train AI models raises questions about the reliability and accuracy of the training data, as well as the potential for AI models to become less effective at handling real-world conversations and interactions.