AI Battle Tested 👀: LinkedIn's Crosscheck Experiment 🚀

CROSSCHECK LAUNCH: A NEW AI EVALUATION TOOL
LinkedIn has introduced Crosscheck, a groundbreaking feature within its LinkedIn Labs initiative, designed to provide premium subscribers with a unique opportunity to evaluate competing AI models. This “blind taste test” approach allows users to compare responses from prominent AI developers like OpenAI, Anthropic, Google, MoonshotAI, Mistral, and Amazon, all without the constraints of traditional subscription models or token limits. The core concept centers around presenting users with a prompt and receiving two distinct answers – one from each participating AI model – before they select their preferred response. This removes inherent biases and allows for a more objective assessment of model capabilities.

THE CROSSCHECK MECHANICS AND INITIAL RESULTS
The Crosscheck feature operates primarily through text-based prompts, acknowledging the current limitations of AI models regarding image generation or file uploads. Users initiate the process by submitting a prompt, and immediately receive two distinct answers generated by different AI models. Following this, the user selects the response they find most satisfactory. Crucially, only after this selection is revealed which model produced each answer. Initial testing by LinkedIn has demonstrated support for a diverse range of models, including multiple Anthropic models, Google’s offerings, MoonshotAI, Mistral, and Amazon’s AI systems. This broad support highlights LinkedIn’s intention to create a comprehensive evaluation platform. Furthermore, LinkedIn is developing a dedicated leaderboard to track user ratings across various industries, offering valuable insights into model performance within specific professional contexts. The company’s Chief Product Officer, Hari Srinivasan, emphasizes that Crosscheck is currently an “early product,” acknowledging the need for ongoing improvements in speed and functionality, as well as the expansion of supported model types and question categories.

DATA SHARING AND FUTURE EXPANSION
A key element of the Crosscheck program involves the anonymized sharing of user data with the respective AI model developers. LinkedIn clarifies that no personally identifiable information is shared, but rather aggregated data concerning model performance across different professional fields is collected. This data is intended to assist model builders in understanding how their AI systems are perceived and utilized by various industries, ultimately informing future product development and refinement. Currently, Crosscheck is exclusively available to LinkedIn Premium subscribers in the United States. However, LinkedIn has announced plans to rapidly expand the feature to additional countries and, eventually, free users, signaling a commitment to making this valuable evaluation tool accessible to a wider audience. The company’s strategic approach reflects a belief in the collaborative potential of data-driven insights within the rapidly evolving landscape of artificial intelligence.