Google’s Bard might have gotten a little training help from the competition
Google Bard is Google’s answer to OpenAI’s ChatGPT, which means that on the surface, we are looking at competing platforms. But interestingly enough, it seems that Bard might have gotten a little help from an unexpected source – ChatGPT.
This is according to a report from The Information (paywall) in which former AI researcher Jacob Devlin claims Bard was being trained using the competition’s data. Devlin is said to have quit after he expressed his concerns to Alphabet’s CEO, Sundar Pichai, and other executives over this training method.
Devlin believed that Bard was trained using ShareGPT, which is a platform where users post and share their exchanges with ChatGPT. Apparently the researcher was concerned about how this might have violated OpenAI’s terms of service, but also how the end result could create an AI that would look and sound too similar to the competition.
That being said, it should be noted that Google has denied Devlin’s claims. In a statement made to The Verge, a Google spokesperson denied that Bard was not trained on any data from either ShareGPT or ChatGPT.
There might be some truth to Google’s denial. When Bard was first released to testers, it was discovered that it actually gave wrong answers to questions that were asked. Presumably had it used ChatGPT’s data, it might have given more accurate responses. It could be due to the fact that it isn’t 100% ready for primetime yet, but either way, it’s rather interesting.
Source: Android Central