
Google is reportedly utilizing Claude, an AI model developed by Anthropic, to help enhance the performance of its own AI system, Gemini. According to reports, contractors hired by Google have been tasked with comparing responses from both AI models to the same user prompts. Their role is to evaluate the output quality based on criteria such as accuracy, clarity, and verbosity.
The process involves contractors reviewing responses from both Gemini and Claude for specific prompts, with up to 30 minutes given to assess and rate the answers. The feedback collected helps Google pinpoint areas where Gemini can be improved. Interestingly, during this evaluation, contractors noticed some unusual patterns. In certain instances, Gemini’s responses referenced Claude, stating, "I am Claude, created by Anthropic," raising questions about the connection between the two systems.
One notable difference between the models was Claude’s strict adherence to safety standards. Contractors observed that Claude often refrains from responding to prompts it considers unsafe, ensuring a high level of ethical compliance. In contrast, Gemini flagged unsafe prompts but provided more detailed responses, which some contractors felt were less stringent. For instance, when handling prompts involving nudity or bondage, Claude refused to engage, while Gemini identified the input as a major safety concern.
To facilitate these comparisons, Google uses an internal platform that allows contractors to test and review multiple AI models side by side. However, the use of Claude has raised questions regarding the terms of service of Anthropic, which prohibit using the model to create or train competing AI systems without approval. While Google, a financial backer of Anthropic, is reportedly using Claude for comparison, it remains unclear whether this falls within permissible use.
In response to speculation, Shira McNamara, a spokesperson for Google DeepMind, clarified that comparing AI models is common practice in the industry and essential for improving performance. She denied claims that Google is using Claude to train Gemini, calling such suggestions inaccurate. This situation highlights the growing competitiveness of the AI industry and the balance companies must strike between collaboration and competition.
See What’s Next in Tech With the Fast Forward Newsletter
Tweets From @varindiamag
Nothing to see here - yet
When they Tweet, their Tweets will show up here.