Google Bard has been accused of training ChatGPT responses

The AI ​​wars are very much on, with Google a little late to the party. In a new report, Google’s Bard has been accused of using ChatGPT responses shared online as training data, but Google denies the claim.

Google Bard is an LLM (Large Language Model) that can generate content based on prompts. This can include explaining topics, answering questions, or creating paragraphs of text at a simple user request. The system works a lot like ChatGPT, the generative AI that captivated the world last year and whose technology also powers the new “Bing” chatbot.

In its first week in the public eye, Bard worked quite similarly to ChatGPT, which means it’s pretty rough around the edges in a number of ways. Bard often gets factual details wrong, sometimes “hallucinating” and creating nonsense, and not mentioning her sources in any capacity.

Related: Hands-on: Bard AI is just as rough as Google put it

But the biggest problem behind the scenes may be how to train Bard. According to a report from the informationJacob Devlin, now a former Google AI engineer, got into trouble with Google by using data from ChatGPT to train Bard.

Devlin reportedly thought the Bard team was “relying too heavily” on the ChatGPT responses posted on the ShareGPT website, where users often share responses they received from an OpenAI chatbot. Devlin also felt that such training could result in Bard’s responses being similar to those from ChatGPT.

After sharing his concerns with Sundar Pichai, Devlin stepped down and now works at OpenAI. The report adds that Google has also stopped using such data to train Bard.

Devlin resigned after sharing his concerns with Pichai, Dean, and other senior managers that Bard’s team, which had assistance from Brain employees, was training its machine learning model using data from OpenAI’s ChatGPT. Specifically, Devlin thought that the Bard team seemed to rely heavily on information from sharinga website where people post conversations they have with ChatGPT.

the information

Other Googlers who were aware of the situation apparently felt that this use violated OpenAI’s Terms of Service, which prohibit the use of ChatGPT output to “develop OpenAI competing models”.

Since this report appeared, Google has issued a brief statement to the edge He says Bard is not trained with data originating from ChatGPT.

Bard has not been trained on any data from ShareGPT or ChatGPT.

Google spokesperson

Google’s statement doesn’t seem to definitively rule out whether or not the data was from ChatGPT never It was used to train the Bard, but that seems not to be the case anymore in the least.

the information The report goes on to say that Google’s Brain AI Group and DeepMind, a company owned by Google subsidiary Alphabet, are teaming up to better compete with OpenAI. The project is known as “Gemini” and is an attempt to “try to match OpenAI’s GPT-4 capabilities,” according to the report. This would include access to 1 trillion parameters – how to measure computations in a machine learning model – identical to those of GPT-4.

More on Google AI: