OpenAI’s ChatGPT is one of the most advanced and popular chatbots in the world. Powered by a massive language model called GPT, it can generate coherent and engaging content in the form of a conversation about various topics. You can imagine how much time and money was spent to create the chatbot, but what if you could make something similar for a fraction of the cost?
Well, that’s what a group of Stanford scientists did with their project called Alpaca 7B. They cloned ChatGPT for only $600 (roughly ₱32,800+), and “fine-tuned” the open-source LLaMA 7B language model using OpenAI’s GPT API.
The result? Alpalca AI exhibited many behaviors similar to ChatGPT and even outperforms it slightly on some tasks. The researchers were surprised by how well it performed given the relatively small amount of instruction data it received. They also claimed that the chatbot was comparable to GPT-3.5.
However, it is not without flaws. The researchers acknowledged that it “suffered from several common deficiencies of language models such as hallucination (making up facts), toxicity (being rude or offensive), and stereotypes (reinforcing biases)” more than ChatGPT.
See also: Can ChatGPT be used in AI-led cyberattacks?
As impressive as Alpalca is, the project showed how relatively easy it was to replicate ChatGPT with minimal cost and effort. It also raises questions about the safety and ethics of creating such chatbots without proper supervision and regulation. As more companies and researchers develop their own versions of artificial intelligence, we might see more clones like Alpaca 7B emerge in the near future.
Source, 2
Image: Karolina Grabowska (Pexels)
