Google Enhances Gemini AI Through Collaboration with Anthropic's Claude

Google is using Anthropic's Claude AI for Gemini improvements, sparking debate over competition and ethics.

Google is said to be using Claude, an AI model developed by Anthropic, to enhance its Gemini AI system. According to a report in TechCrunch, Google has engaged contractors to compare responses produced by Gemini and Claude on the same user prompts. The point is to measure how each AI does in terms of truthfulness, clarity, and verbosity.

The process requires the contractors to assess the responses from both AI systems to a given prompt within up to 30 minutes for rating each response. This is helpful in identifying areas in which Google can improve Gemini. During evaluation, contractors noticed some interesting patterns. At times, the output from Gemini included references like, "I am Claude, created by Anthropic," which led to questions about the possible connections between the two systems.

One of the differences that was mentioned was Claude's strict safety approach. Contractors noticed that Claude never responds to prompts that are deemed unsafe, and he always maintains a high ethical standard. On the other hand, Gemini also flagged unsafe prompts, but its responses were considered more elaborate, though slightly less stringent. For instance, when asked about nudity and bondage, Claude wouldn't respond, while Gemini marked it as a significant safety violation.

Google has an internal platform that enables the comparison of these AI models, which contractors can test side by side. However, using Claude raises a concern since Anthropic's terms of service do not allow using its AI model to train competing systems without permission. Google has financial backing in Anthropic, but it is still unknown if this lets them bypass these restrictions.

To these concerns, Shira McNamara, a Google DeepMind spokesperson, clarified that model comparison is standard practice in AI development. She denied claims that Claude was used to train Gemini, stating that the report was inaccurate. This collaboration reflects the competitive nature of the AI industry, where companies balance internal and external development to refine their technologies.