Big tech companies have never been more eager to put on the brakes as much as they have in the face of OpenAI’s ChatGPT; some tech behemoths, such as Microsoft, were quick to jump on board, while others, such as Google, were left catching up. The company finally unveiled Bard last month in response to OpenAI’s popular generative AI. Now, we’re learning of accusations claiming its researchers used data from OpenAI’s technology without permission to help develop Bard.
Like other types of generative AI, Bard is built on a combination of machine learning models and training data. In Bard’s case, Google used its Language Model for Dialogue Applications, or LaMDA, to form the chatbot’s syntax. For training data, there are plenty of open-source libraries for AI researchers to use and we don’t doubt that Google has scoured around for them. However, it appears that this public domain data wasn’t enough.
According to The Information, Google trained Bard using data from ChatGPT responses shared publicly through ShareGPT, a website where users can share responses from OpenAI’s chatbot. Jacob Devlin, a Google AI researcher, left the company after warning top executives (including CEO Sundar Pichai) that using ChatGPT data would violate OpenAI’s terms of service. Devlin, who now works for OpenAI, was also concerned that if Google tapped into ChatGPT’s data, Bard might end up with similar responses.
The report also claims that Google ordered its DeepMind division to collaborate with the Brain team on a new initiative known internally as Gemini.
Google sent a statement to WinFuture saying that Bard’s training did not rely on data from ChatGPT. “Bard is not trained on data from ShareGPT or ChatGPT,” the company told the publication.
Android Police has also contacted Google for a statement, but we haven’t heard back yet.
Walking a tightrope between ethical concerns and lack of progress, Google has scrambled and even crunched on its way to create Bard on an advanced timeline. Business Insider previously reported that Pichai asked Google employees to take up to four hours out of their day to iron out a few kinks in its ChatGPT challenger before its release.
It remains unclear at the moment whether data from ChatGPT was used to train Bard at some point prior to its launch and whether Bard operates on any of that training. So far as Google says, neither are the case.
