Martian's model router is designed to help companies achieve higher performance and lower costs by intelligently switching between different LLMs depending on the task at hand. The tool can estimate a model's performance without running it, allowing it to route to cheaper models when they perform similarly to more expensive ones. The company claims it has already seen uptake from multi-billion-dollar companies.
Key takeaways:
- Shriyash Upadhyay and Etan Ginsberg, AI researchers from the University of Pennsylvania, have founded a company called Martian, which has raised $9 million in funding. The company aims to focus on interpretability research in AI, rather than capabilities research.
- Martian's first product is a "model router", a tool that routes prompts intended for large language models (LLMs) to the most suitable LLM based on factors such as uptime, skillset, and cost-to-performance ratio.
- The model router can help companies achieve higher performance and lower costs by using a team of models in an application, rather than relying on a single, potentially expensive LLM.
- Despite similar technology existing in the market, Martian's founders claim their model router is unique due to its ability to understand how models fundamentally work, and it has already seen uptake among multi-billion-dollar companies.