Google is reportedly using Claude, an AI system developed by Anthropic, to enhance the performance of its AI model Gemini. According to TechCrunch, a contractor hired by Google was tasked with evaluating and comparing the output of Gemini and Claude in response to the same prompts. The ratings focus on criteria such as veracity, clarity, and redundancy, and provide Google with valuable insights to improve Gemini’s performance.
The process reportedly involves the contractor receiving responses from both AI models for specific prompts. We then spend up to 30 minutes thoroughly assessing and rating the quality of each response. This feedback helps Google pinpoint areas where Gemini needs improvement. However, contractors observed a unique pattern during these evaluations. Occasionally, Gemini’s responses strangely include references such as “I am Claude, created by Anthropic,” raising questions about the model’s interrelationship.
The report highlights that one notable difference in the comparison is the model’s approach to safety. Claude is known for his firm stance on ethical boundaries, often refusing to engage in prompts he deems dangerous. In contrast, Gemini also flags such input as a violation, but provides a more detailed explanation, albeit in a potentially less rigorous approach. For example, in a case involving sensitive topics such as nudity or bondage, Claude chose to decline completely, while Gemini elaborated on safety concerns.
According to the report, Google is using an internal platform to facilitate comparisons of these models, allowing contractors to test and review AI systems in parallel. However, Claude’s involvement caused controversy. Anthropic’s terms of service prohibit the use of Claude to train competing AI products without prior approval. This restriction applies to outside companies, but it’s unclear whether it also applies to financial backers like Google, which has invested in Anthropic.
Google DeepMind spokesperson Shira McNamara addressed the speculation, saying comparing models is standard practice in the industry, India Today reported. She flatly denied claims that Claude was used to train Gemini and labeled such suggestions inaccurate.
