Google is using the AI model Claude from Anthropic to enhance the performance of its own Gemini AI. According to internal documents, contractors are comparing responses from Gemini and Claude to determine which are more accurate, reliable, and better formulated. But what does this mean for the development of AI models and the impact on companies that utilize this technology?
Insight into the comparison process
Google does not use Claude to train Gemini, but to evaluate its performance. Contractors assess the responses of both models based on criteria such as truthfulness, completeness, and detail. For each prompt, they are given up to 30 minutes to thoroughly compare the answers. This process is intended to identify and improve weaknesses in Gemini.
However, this raises questions about compliance with terms of use. Anthropic, the creator of Claude, explicitly prohibits the use of their model for competing applications without permission. Despite Google being a major investor in Anthropic, it remains unclear whether this permission has been obtained.
Impact on businesses and technology development
The ongoing competition to develop better AI models has direct implications for companies and organizations. Comparing Gemini with Claude highlights how essential safety and accuracy are. Claude is known for its strict safety guidelines. In some cases, it even refuses to respond to prompts deemed unsafe. In contrast, Gemini responds more frequently, but can sometimes provide risky answers, such as information that is inappropriate or unsafe.
For companies in sensitive sectors, such as healthcare and legal services, such a difference can be decisive. An AI model that does not meet strict safety requirements can lead to misinformation or even legal risks. This underscores the importance of critically evaluating AI models before they are deployed.
Approach and solutions in AI testing
Google states that Claude is used solely for evaluation and not to train Gemini. This approach reflects standard practices within the industry. Using competing models as benchmarks helps developers improve the performance of their own systems.
However, transparency and adherence to agreements remain crucial. Companies must ensure that they respect the terms of their partners and act ethically. This not only increases trust in the sector but also fosters fair competition.
Organizations that rely on AI should conduct thorough research into the safety and reliability of the models they use. Ensuring ethics and safety must remain a priority, not only for developers but also for companies that apply AI technology.