| Different models have different strengths.
As heavy AI users, we constantly bounce between different subscription-based platforms like ChatGPT, Claude, and Gemini to complete various tasks. Not only is this annoying - it is expensive. There are many practical (but pricey) AI models, each offering distinct capabilities. Choosing one can be difficult; it's like picking a smartphone based on camera quality—they're all similar, but only an expert can tell the difference. We built PreCog to simplify this process by automatically selecting the most suitable model for any task. This allows users to focus on achieving goals efficiently and creatively without being locked into one chatbot experience. What makes PreCog different?
PreCog isn't just one model or AI chatbot; Instead, It acts as an AI "middleman," directing/matching prompts to the most suitable AI model based on the model's strengths and weaknesses relative to thuser'sr's query. AI Model Matchmaking: PreCog automatically selects the most appropriate AI model for each request. This feature conserves time and enhances the accuracy of generated responses. Versatile Assistance: PreCog is adaptable to various tasks, from coding projects to creative writing. It provides reliable support for diverse needs and ensures the availability of the right tools for multiple applications. Continuous Model Improvements: PreCog incorporates the most recent advancements in AI technology by utilizing models informed by the latest LLM leaderboard. This ongoing improvement ensures that users experience an up-to-date and efficient AI service. How does the PreCog work?
We rank the performance of select LLMs on our Model Leaderboard, which powers PreCog. Rankings are based on over 1,000,000 human pairwise comparisons, evaluated using the Bradley-Terry model, and displayed in an Elo-scale rank. The dataset contains AI battles scored by humans. It is maintained by Chatbot Arena (lmarena.ai), an open-source platform for evaluating AI through human preference developed by researchers at UC Berkeley SkyLab. Our evaluation focuses on a curated subset of models used explicitly within the Ubik ecosystem, providing some insight into the capabilities behind PreCog. We would love your feedback. We originally built PreCog as an internal tool to centralize our AI usage but wanted to share its capabilities with like-minded people. Feel free to reach out with any questions or critiques, as they really shape our dev cycle! |