Researchers at Stanford University have taken down their short-lived chatbot that harnessed Meta’s LLaMA AI, nicknamed Alpaca AI. The researchers launched Alpaca with a public demo anyone could try last week, but quickly took the model offline thanks to rising costs, safety concerns, and “hallucinations,” which is the word the AI community has settled on for when a chatbot confidently states misinformation, dreaming up a fact that doesn’t exist.
Along with an interactive demo, Stanford’s Human-Centered Artificial Intelligence institute also published Alpaca’s training data, the code for its data generation process, and the training code used to fine-tune the model, according to the Register.
In a press release announcing the initial release of Alpaca , lead author Rohan Taori, a computer science PhD student at Stanford, acknowledged a public test has risks. “Deploying an interactive demo for Alpaca also poses potential risks, such as more widely disseminating harmful content and lowering the barrier for spam, fraud, or disinformation,” the researchers said. Taori said the team implemented a content filter and a watermark system that could identify Alpaca’s words in order to mitigate those risks.
There’s no word about exactly what went wrong. “The original goal of releasing a demo was to disseminate our research in an accessible way. We feel that we have mostly achieved this goal, and given the hosting costs and the inadequacies of our content filters, we decided to bring down the demo,” a spokesperson for the department told the Register. Stanford’s Human-Centered Artificial Intelligence institute didn’t immediately respond to a request for comment.
Despite its apparent failures, Alpaca has some exciting facets that make the research project interesting. Its low upfront costs are particularly notable. The researchers spent just $600 to get it working, and reportedly ran the AI using low-power machines, including Raspberry Pi computers and even a Pixel 6 smartphone, in contrast to Microsoft’s multimillion-dollar supercomputers.
You can’t access a working copy of Alpaca anymore, but the code and underlying data are still up on GitHub.
“We encourage users to help us identify new kinds of failures by flagging them in the web demo,” the researchers said in the press release. “Overall, we hope that the release of Alpaca can facilitate further research into instruction-following models and their alignment with human values.”
Want to know more about AI, chatbots, and the future of machine learning? Check out our full coverage of artificial intelligence, or browse our guides to The Best Free AI Art Generators and Everything We Know About OpenAI’s ChatGPT.
Discussion about this post