So you want to build your own open source ChatGPT-style chatbot…

Mozilla's hackathon aimed to build an internal chatbot prototype running on open source large language models (LLMs) and hosted on Mozilla's cloud infrastructure. They chose to avoid third-party AI services to maintain control and privacy. They encountered challenges while using Hugging Face's hosted API, leading them to shift their attention to llama.cpp, an open source project that allowed them to run LLMs on consumer-grade hardware. The project's decision to use LLMs based on Facebook's LLaMA architecture limited their model selection due to licensing restrictions, but they evaluated several models manually to address biases and ensure trustworthiness.