ALPaCA AI: Stanford launches ChatGPT clone – for just $600

L’artificial intelligence (AI) it is revolutionizing many industries and applications, but the high cost of development and its complexity still make it inaccessible to many. However, a team of researchers from the Stanford University developed a new method to create cheap and easy-to-use AI models inspired by alpacas. Indeed, with ALPaCA AI, Stanford researchers have been able to offer an alternative to ChatGPT developed for only 600 dollars.

Alpaca AI, Stanford’s ChatGPT clone developed for $600

Alpacas are animals native to South America, known for their soft wool and their docile temperament. But what do they have to do with AI? According to the researchers, alpacas have a very useful feature for machine learning: they are able to adapt quickly to new environments and situationsthanks to their short-term memory.

Traditional AI models, on the other hand, require a lot of computational resources and data to train and update. Also, they tend to forgetting previous information when learning new ones, a phenomenon known as “catastrophic forgetfulness”. This makes them inflexible and scalable.

To overcome these limitations, researchers have devised a new method called ALPaCA (Adaptive Learning with Partially Available Contextual Information)which allows you to create AI models capable of learning from a few examples and to maintain the acquired knowledge. The method is based on two main components: a Bayesian neural network and a key-value memory.

The Bayesian neural network is a technique that allows you to estimate the probabilities of the different hypotheses given the available information. This way, the model can make accurate predictions even in the presence of incomplete or noisy data. There key-value memory is a tool that allows you to store associated data pairs. In this way, the model can retrieve the relevant information for the current context without forgetting the previous ones.

By using these two components together, researchers have succeeded in creating AI models that can adapt quickly to new tasks and domains with little data and resources. For example, they have demonstrated that their method can generate coherent texts on various topics with only 10 examples for each topic. Furthermore, they showed that their method can predict the air temperature in different cities of the world with only 20 observations for each city.

The researchers believe their method could have many practical applications in different industries such as medicine, engineering or education. In fact, their goal is to make AI more accessible and democratic to all users. When we think of the billions invested in other artificial intelligences, this open source solution seems like a truly momentous step in democratizing AI.