Tracking the Explosive World of Generative AI

Controversy at Google: AI Chatbot Bard Accused of Using OpenAI's ChatGPT Data

Google's AI team faces scrutiny over allegations that their chatbot Bard was trained using data from OpenAI's ChatGPT, sparking internal dispute and the resignation of a senior engineer.

Google is under fire as inside sources reveal its chatbot, Bard, was trained on ChatGPT's outputs. Photo credit: Google / Reuters

🧠 Stay Ahead of the Curve

Google's AI team purportedly employed OpenAI's ChatGPT data in their Bard chatbot's training, resulting in internal disputes and the departure of a senior engineer.
Google's recent endeavors underscore the intense pressure from their "code red" scenario as they strive to match OpenAI's capabilities.
The prevailing arms race propelling tech companies to rapidly create AI solutions provides limited space for addressing responsible and ethical development concerns raised by critics.

By Michael Zhang

March 30, 2023

Google's AI team, responsible for the development of Bard, the company's AI chatbot, has come under fire amid allegations that they may have used data from OpenAI's ChatGPT during the chatbot's training, as reported by The Information.

The decision sparked controversy within Google's product team, prompting senior AI engineer Jacob Devlin to resign in January and then join OpenAI. Before resigning, Devlin voiced concerns to the company's top executives, including CEO Sundar Pichai. Devlin notably contributed to a groundbreaking paper on training machine-learning models to enhance their understanding of sentence clusters—an innovation that OpenAI has integrated into its language models.

Sources within Google claim that Bard's training involved data from OpenAI's ChatGPT, specifically logged conversations from ShareGPT, a public platform where users can post transcripts of their ChatGPT interactions. Several Google employees reportedly flagged that this approach would have contravened OpenAI's terms of service, which expressly forbid using ChatGPT's outputs to develop models that rival OpenAI. The Information's findings also suggest that Google employees were apprehensive that Bard's answers would too closely resemble ChatGPT's responses due to the contentious training.

Ultimately, Google ceased using ChatGPT data to train Bard, according to The Information. Bard was later released in preview access, garnering lukewarm reviews from AI researchers and technology journalists. The New York Times characterized Bard's capabilities as "less detailed," "odd," and "minimalist" in comparison to ChatGPT, which recently began using OpenAI’s GPT-4 large language model (LLM).

In response to ChatGPT's launch, Google has implemented a "code red," acknowledging that they are trailing behind in AI capabilities. The Information reports that the pressure to make rapid advancements in generative AI has become so intense that, for the first time, rival Google teams—Deepmind and Google Brain—have joined forces to create a chatbot that can compete with OpenAI's GPT-4.