- Google is accused of using data from ChatGPT, OpenAI’s chatbot, to train its own Bard chatbot.
- Google denies these accusations and claims that Bard is based on its own LaMDA model.
- This case illustrates the rivalry between Google and OpenAI in the field of generative artificial intelligence.
Not yet deployed in the world that Google’s conversational AI is already controversial. According to the website The Information, Google would have trained Bard with data from ChatGPT, the competing artificial intelligence developed by OpenAI. All this without permission.
The media quotes a former Google artificial intelligence researcher, Jacob Devlin. The latter would have resigned from Google to join OpenAI, after alerting the management of his former company to the use of ChatGPT data. He reportedly said the practice violated OpenAI’s terms of service and risked making Bard’s responses too similar to ChatGPT’s.
According to the report, Google allegedly grabbed ChatGPT data from a website called ShareGPT, which allows Internet users to interact with OpenAI’s AI. Devlin reportedly shared his concerns with Google CEO Sundar Pichai and Jeff Dean, head of the Google Brain division, which oversees the Bard project.
Google denies
Google categorically denied these allegations and claimed that Bard was not based on ChatGPT data. “Bard is not trained with any data from ShareGPT or ChatGPT”said Chris Pappas, spokesperson for Google, at the site The Verge.
Google also clarified that Bard was based on his own model of artificial intelligence LaMDA (Language Model for Dialogue Applications – language model for conversational applications).
This case illustrates the growing rivalry between Google and OpenAI in the field of generative artificial intelligence, which consists of creating textual or visual content from a very simple request.
OpenAI has taken a step ahead in this area with its GPT-4 model, which is considered the most powerful in the world. Microsoft has also invested heavily in OpenAI and has integrated GPT-4 into its products. Its Bing search engine already benefits from this and integrations with the Office Suite are coming soon.
Trolling Bard
Google is therefore seeking to catch up with Bard, which it presents as an AI capable of adapting to all subjects and all styles of conversation. But for the moment, Bard has not convinced the first users with its performance. Some even noticed that Bard gave contradictory or erroneous information about his nature and origin.
For example, an SEO agency named Loopex Digital claimed to have had a conversation with Bard where the chatbot said it was based on GPT-3. But later Bard changed his tune and claimed to be based on LaMDA. The agency provided screenshots of his interactions with Bard at the site Android Authority.
Bard may simply have made a mistake common to generative AI models. But it could also be that Bard is hiding something…
