Documents reveal what tool Google used to try to beat ChatGPT: ChatGPT itself

In a notable turn of events, Google has taken significant steps to bolster its AI capabilities by utilizing insights from ChatGPT. A trove of documents obtained by Business Insider reveals that Google enlisted the help of contractors from Scale AI to refine Bard, its AI chatbot, which was initially perceived as lacking upon its release earlier in 2023. The internal documents indicate that Scale AI contractors generated thousands of responses from ChatGPT, meticulously comparing them with Bard's outputs. This effort aimed to enhance Bard's performance, evolving the chatbot into what is now known as Gemini. The contractors were instructed to analyze why GPT-4's responses were superior and to ensure that Bard's rewrites met or exceeded those standards. Notably, the documents flagged several contractors whose work consistently fell short of ChatGPT's quality, with performance bonuses offered for those who could outperform the renowned AI. Scale AI, a startup based in San Francisco, plays a pivotal role in AI development for major tech firms. The company relies on a workforce of human contractors to perform essential tasks, including refining chatbot responses. Recent reports suggest that Meta may invest $15 billion in Scale AI, highlighting the startup's growing influence in the AI landscape. Despite the close scrutiny of ChatGPT's capabilities, both Scale AI and Google have denied using ChatGPT for direct model training. A spokesperson from Scale AI clarified that the analysis conducted was part of standardized evaluation practices rather than a method of training models. Google echoed this sentiment, asserting that no outputs from competing companies were incorporated into Gemini's training. The documents also reveal the rigorous evaluation process undertaken by Scale AI, which included creating spreadsheets that compared Bard's rewrites against ChatGPT's responses. This meticulous approach was aimed at identifying weaknesses in Bard’s outputs and improving them, particularly in specialized domains like engineering and physics. However, the exposure of these internal documents raises concerns about data privacy, as they inadvertently disclosed sensitive contractor information. Scale AI is currently investigating how this data became publicly accessible and is taking measures to rectify the situation. Since the initial challenges faced by Bard, Google has revamped its AI strategy, launching over 100 new AI products and features in recent months. CEO Sundar Pichai has emphasized the company’s commitment to rapid innovation in the AI space, making it clear that Google is determined to lead the industry in AI advancements.