Exploring the Evolution of Language Models: From Llama 1 to Open Router Marketplace

As AI bug bites, [Speaker] ventures into open-source language models, founding Open Router to democratize access and unlock the potential of multimodel inference.

  • 1. The speaker started Open Router in early 2023 to explore the new market of AI software.
  • 2. They wanted to answer the question: "Will this market be winner-take-all?"
  • 3. Open AI was the leading model, with a few others coming up on its tail.
  • 4. The speaker built prototypes and investigated open source options.
  • 5. In January, users showed interest in different types of models, starting with moderation.
  • 6. Users wanted to understand moderation policies and avoid being deplatformed.
  • 7. Some users generated content that was questionably against terms of service.
  • 8. In February, the open-source race began, with models like Bloom 176b and OPT by Facebook.
  • 9. Llama 1, a smaller model, outperformed GPT3 on most benchmarks in February.
  • 10. Alpaca, the first successful distillation of LLM style and knowledge onto a small model, was created in March.
  • 11. Alpaca was made for less than $600 and marked a significant breakthrough.
  • 12. The speaker saw potential in making unique data available as a service through language models.
  • 13. Open Router initially aimed to collect all the models and help people understand how to use them.
  • 14. In April, the Window AI Chrome extension was launched, allowing users to choose their model for web apps.
  • 15. Open Router officially launched in May, co-founded by the creator of Plasmo.
  • 16. Open Router now offers an API that provides access to all language models and data about usage trends.
  • 17. The platform has grown 10 to 100% month over month for two years and hosts over 400 models from more than 60 providers.
  • 18. Open Router initially functioned as a collection of models with limited providers, but the ecosystem soon became more complex.
  • 19. The platform now aggregates providers at different price points and helps developers maintain uptime for open source or closed source models.
  • 20. Real-world data on latency and throughput is provided for each model, making Open Router a marketplace and an optimization tool.
  • 21. The speaker believes that the future of AI will involve multimodal models tailored to various purposes.
  • 22. Inference will be a commodity, with selection and routing being crucial for developers.
  • 23. Active model usage on Open Router has steadily grown, and the platform aims to make the ecosystem more homogeneous and easier to work with.
  • 24. The speaker's company is working on adding new modalities like image generation and improving prompt observability, discovery, and pricing.

Source: AI Engineer via YouTube

❓ What do you think? What implications do the democratization of AI models and the rise of multimodel inference have on the future of software development, data analysis, and decision-making? Feel free to share your thoughts in the comments!