Google has refuted recent allegations suggesting that its AI chatbot, Bard, was trained on data obtained from OpenAI’s ChatGPT. According to a report published in The Information, former Google engineer Jacob Devlin informed top executives, including CEO Sundar Pichai, that Bard’s machine learning models were being trained using data from ChatGPT. Devlin warned that this could violate OpenAI’s terms of service and result in Bard’s responses closely resembling ChatGPT’s.
Google vehemently denies these claims. In an interview with The Verge, Google spokesperson Chris Pappas stated, “Bard is not trained on any data from ShareGPT or ChatGPT.”
This is not the first time Google has faced controversy regarding its AI chatbot. Shortly after Bard’s public announcement, the chatbot claimed it was trained on Google’s internal data, including Google Search, Gmail, and other products, in response to a query about its dataset. Google later clarified that Bard was not trained on Gmail data and that the response might have been an error made by the experimental language model.
Bard was introduced last month as a competitor to OpenAI’s ChatGPT, which has gained immense popularity. However, the launch was fraught with issues, including factual errors in Bard’s introduction advertisement and a missing demo phone during a press conference in Paris.
Google recently rolled out Bard for public testing, aiming to compete with ChatGPT. Available for selected users in the US and UK, Bard offers similar capabilities, such as writing essays, planning events, creating travel itineraries, and suggesting dinner recipes. However, unlike ChatGPT, Bard lacks multimodal abilities and cannot generate code.
According to The Information report, Devlin joined OpenAI after leaving Google. It suggested that Google might have stopped using ChatGPT data following Devlin’s warnings and possibly discarded that part of the training process.
The recent allegations and Google’s denial highlight the competitive landscape in the AI chatbot space. As more companies develop AI-powered language models, questions surrounding data sources and usage, as well as intellectual property, will continue to arise. For now, Google maintains that Bard was not trained on any data from ShareGPT or ChatGPT, focusing on improving the chatbot’s capabilities and gathering user feedback to fine-tune the system.