Recently, the improvement work of Google’s Gemini AI model has caused controversy. According to TechCrunch, Google contractors compared Gemini's answers to the output of Anthropic's competitor Claude to evaluate Gemini's performance. The move raises questions about whether Google used Claude's data without authorization and whether it violated Anthropic's terms of business services, since Anthropic is one of Google's major investors. The incident highlights the intensity of competition in the AI field and the importance of data usage and compliance issues.
The competition among technology giants in the field of AI is becoming increasingly fierce, and a recent piece of news has once again brought the undercurrent of industry competition to the forefront. According to internal communications obtained by TechCrunch, the contractor responsible for improving Google's Gemini artificial intelligence model was comparing its answers to the output of Anthropic's competing model Claude. The move raised questions about Google's compliance with Anthropic's authorization to use Claude for testing.
As technology companies compete to develop better AI models, model performance is usually assessed through industry benchmarks, rather than contractors spending a lot of energy evaluating competitors' AI responses. However, according to reports, Gemini contractors are required to score every response they see based on multiple criteria, including authenticity and verbosity. The contractor has up to 30 minutes to decide whether Gemini or Claude's answer is better.
Contractors noted that Claude's responses seemed to emphasize security more than Gemini's, internal chat records show. One contractor said: "Claude has the most stringent security settings of all AI models." In some cases, Claude will not respond to prompt words it considers unsafe, such as playing a different AI assistant. In another test, Claude avoided answering a certain prompt word, and Gemini's response was flagged as a "serious security violation" for including "nudity and restraint."
Anthropic's terms of business services specifically prohibit customers from accessing Claude "to build competing products or services" or "train competing AI models" without approval. It’s worth noting that Google is Anthropic’s major investor. This background has increased the outside world’s doubts about Google’s behavioral compliance.
In response, McNamara, a spokesman for Google DeepMind (responsible for Gemini), said that DeepMind does "compare model outputs" for evaluation, but does not train Gemini on the Anthropic model. "Of course, as is industry practice, in some cases we compare model outputs as part of the evaluation process," McNamara said. "However, any suggestion that we used the Anthropic model to train Gemini is inaccurate."
Although Google denies using Claude to train Gemini, its rhetoric of "comparing model output" is still ambiguous and cannot completely eliminate external doubts. The incident has once again raised concerns about data use and compliance in the development of AI models. In the fiercely competitive AI field, how to maintain fair competition and respect intellectual property rights will become an important issue that requires continued attention and discussion during the development of the industry.
Google’s explanation of the incident did not completely eliminate public doubts, which reflects the fierce competition in the AI field and the seriousness of data compliance challenges. In the future, the industry needs to establish clearer rules and standards to ensure fair competition and intellectual property protection.