Google partners with Anthropic's Claude to refine Gemini AI: What’s happening?

In a surprising move, Google is leveraging Anthropic’s AI model, Claude, to enhance the performance of its own AI system, Gemini. According to a report by TechCrunch, contractors hired by Google are comparing outputs from Gemini and Claude, evaluating factors like truthfulness, clarity, and verbosity to provide feedback for improvement.

Google and Anthropic Join Forces to Improve AI Models

How the evaluation process works

Contractors are presented with responses from both Gemini and Claude to identical prompts. They meticulously assess and rate these responses within 30 minutes, focusing on how well the outputs meet evaluation criteria. This feedback is crucial for identifying areas where Gemini can improve.

Interestingly, contractors noticed occasional quirks in Gemini's responses, with some outputs including references like, “I am Claude, created by Anthropic.” This has fueled speculation about the connection between the two systems.

Safety standards: Claude vs Gemini

One standout observation during these comparisons was Claude’s strict adherence to safety. Claude consistently refuses to engage with prompts it deems unsafe, maintaining high ethical standards. Gemini, while also flagging unsafe prompts, offers more detailed explanations, though it appears less rigid in its approach.

For instance, in one case involving prompts related to nudity and bondage, Claude refused to respond, while Gemini flagged the input as a significant safety violation with a detailed explanation.

Addressing ethical and competitive concerns

Google uses an internal platform to facilitate these comparisons, allowing contractors to evaluate multiple AI models side by side. However, the involvement of Claude has raised eyebrows, as Anthropic’s terms of service prohibit using Claude to train competing AI models. It remains unclear if this restriction applies to Google, given its financial backing of Anthropic.

Shira McNamara, a spokesperson for Google DeepMind, addressed these concerns, stating that comparing AI models is a common industry practice. She firmly denied claims that Google used Claude to train Gemini, calling such accusations inaccurate.

The collaboration highlights rising competition in AI

This collaboration underscores the increasing competitiveness of the AI industry, where giants like Google are blending internal innovation with external partnerships to refine their models. However, it also raises critical questions about ethical boundaries and the delicate balance between collaboration and competition in the fast-evolving AI landscape.