Stanford Shuts Down Low-Cost Alpaca AI Demo Due to Safety and Hosting Costs
Stanford University takes its Alpaca generative AI chatbot offline after just a few days, citing unexpected hosting expenses and inadequate content filters. The model, based on Meta's LLaMA system, originally launches to prove that cheap AI replication is possible despite known accuracy issues.
Stanford University removes the web demo for its Alpaca generative AI model after only a few days of public access. The researchers decide to shut down the chatbot due to unexpected hosting expenses and failures in the system's content filters. Alpaca builds upon Meta's LLaMA AI model, a tool designed specifically to help researchers study and mitigate issues with wrong or inappropriate text generation.
The Stanford team initially creates Alpaca to prove that replicating models like OpenAI's text-davinci-003 requires very little money. By fine-tuning the smaller LLaMA-7B model, the researchers build the entire system for just $600. Despite this incredibly low construction cost, the team discovers that hosting the AI demo for public use is far too expensive to maintain.
Safety concerns also play a major role in the decision to take the demo offline. The researchers openly acknowledge that Alpaca hallucinates incorrect facts and easily produces toxic language or misinformation. Although the original goal is to share the research in an accessible way, Stanford ultimately concludes that the content filters are inadequate and takes the entire project down.