Stanford created an AI similar to ChatGPT for less than $600, revolutionizing communication and content creation. Find out how the developers did it and the consequences of this feat.
Artificial intelligence has advanced significantly with the creation of cutting-edge language models such as OpenAI’s ChatGPT. However, the complexity and high cost of building these models often make them inaccessible to researchers and independent developers. This changed recently with Stanford’s announcement of the creation of an AI for less than $600. It’s a feat that could revolutionize the way machines communicate and create content.
Stanford’s Alpaca: an affordable AI competitor to ChatGPT?
Stanford has developed a AI named Alpaca which showed results identical to those of the ChatGPT language model. However, Alpaca was developed from a free algorithm model. Therefore, its creation cost less than 600 dollars. These AIs, which seemed to be expensive and inaccessible technologies, are now cheap and easy to reproduce.
Just six months ago, only researchers and experts were following the development of these language models. Today, they are developing at an exceptional pace, as witnessed by the launch of GPT-4. In the long term, AI technology could change the way human society functions. This is reflected in theautomation of professional tasks previously considered impossible, particularly among white-collar workers.
Until six months ago, only experts were aware of the great advances made in language models. However, with the launch of ChatGPT, machines are now able to respond in a similar way to humans. They are capable of producing written content, even computer code, in a wide range of domains in record time, and often at a very low cost. remarkably high level of quality.
Alpaca: how did Stanford create this high-performance, low-cost AI?
Stanford scientists implemented this AI using affordable and readily available tools. They exploited LLaMA 7B, an open-source language model from Meta, Mark Zuckerberg’s company, which is one of the smallest and most currently available..
Using an Application Programming Interface (API), they were able to generate conversations using hundreds of instructions written by humans. Finally, the algorithm accumulated 52,000 examples of conversations in record time and at a minimal cost of $500. The data collected was then used to train the LLaMa model.
Using A100 computers, the researchers were able to complete this task in just three hours. As a result, the development cost was less than $100. Named Alpaca, the model thus created was tested in various domains against ChatGPT and managed to outperform it. Although the process has not been optimized, the researchers have demonstrated that it is possible to create efficient AIs at low cost.