Google accused of using ChatGPT data to train Bard

A former Google employee said that the company used ChatGPT data taken from the ShareGPT website to train its conversational AI Bard. The Information reports.

According to the publication, former Google AI engineer Jacob Devlin spoke out strongly against this approach. After trying to talk the management out of it, he left the company and joined OpenAI.

Devlin argued that using ShareGPT violates licensing agreements. He added that this training approach would make Bard’s and ChatGPT’s responses too similar.

Another source told The Information that Google stopped using the data after his warnings. It is possible the company removed this portion of training.

In a statement to The Verge, Google’s spokesperson Chris Pappas denied the report.

“Bard was not trained on any data from ShareGPT or ChatGPT,” he said.

Google presented its version of a conversational AI in February 2023. It appears the company was rushing the announcement — the presentation took place a day before Microsoft’s “new Bing”.

At the same time, Google did not release Bard to the public. Preliminary testing of the service among a limited number of participants took place a month and a half after the announcement.

Employees criticised Google for the rushed presentation. The company also did not notice that the chatbot made an error in promotional materials, which cost the parent company’s $100 billion in market capitalisation.

In February, Google’s management asked staff to test Bard and fix poor responses.

Soon, internal chats were filled with memes and AI dialogues about potential layoffs.