Google denies using ShareGPT training for BART and ChatGPT models

source: Interesting Engineering Google denies using the ShareGPT training method to train its BART and ChatGPT language models. The ShareGPT method was proposed by researchers at the University of Washington, and it involves training multiple language models together to improve their performance. According to a Google spokesperson, the company did not use the ShareGPT method for its BART and ChatGPT models, but instead used its own in-house methods. Google also stated that it is committed to responsible AI practices and ensuring that its language models are transparent and unbiased. However, some researchers have criticized Google's lack of transparency and openness about its language model training methods. They argue that sharing training data and techniques can help improve the overall quality and fairness of language models. [caption id="attachment_132757" align="aligncenter" width="1200"] source: GizChina[/caption] Despite the controversy, Google's BART and ChatGPT models have shown impressive results in various language tasks, such as summarization and question answering. The company has also released a public API for ChatGPT, allowing developers to access the model's capabilities for their own applications. While Google denies using the ShareGPT training method for its BART and ChatGPT models, the controversy surrounding language model training methods highlights the importance of transparency and open sharing of data and techniques in AI research.