Chinchilla AI/ In the world of technology, artificial intelligence has become one of the most important and most relevant aspects currently in the popularity statistics of the digital world, taking into account the latest related advances and the new proposals that are emerging today. Every day they are launched into the world, there is no doubt that this is one more step towards automation.
In this sense, chatbots or virtual assistants are a topic of enormous relevance today, as a consequence of the popularization of OpenAI and its famous ChatGPT, new proposals have been developed as a way to enhance and take advantage of the benefits of GPT development. 3 and GPT-3.5, so that the automation of processes and the synthesis of large amounts of text becomes faster and easier every day.
Taking this approach, in this article we tell you about Chinchilla AI, a proposal from DeepMind to revolutionize automatic language models.
Origin of the ChatBot: The customer service model
In order to understand what proposals based on intelligent systems such as ChatGPT or Chinchilla AI are about, first of all, it is necessary to return to the concept of a virtual assistant. Taking a look at the past, the first virtual assistant was created in 1966 under the name Eliza, developed by the Massachusetts Institute of Technology (MIT) whose objective was focused on conversation. At that time, the institute's staff often carried out intimate conversations and secrets with the artificial intelligence, and they said they felt listened to.
Today, the proposal of virtual assistants has become one of the greatest interests in the world of technology and software development, carrying out much more sophisticated ideas with a greater innovative scope. A clear example of this is virtual assistants, such as Siri or Alexa, as well as smart homes that take advantage of artificial intelligence to automate day-to-day tasks.
What is a chatbot?
Although we have an exact understanding of what artificial intelligence means, encompassing the use of intelligent algorithms, today companies such as Apple, Google and Amazon have dedicated themselves to developing their own artificial intelligence or Chatbot that can provide their users with a new form of interaction with your brand in addition to a high-level service.
Now, taking a descriptive approach, predict that a chatbot is a software whose purpose is oriented to online conversation, improving human-machine communication, generating a natural language capable of interpreting the user's demands and queries, providing relevant answers. and current.
Although chatbots have functioned as ways to speed up certain searches and resolutions, there is a much greater scope now that this technology has expanded to the industrial and business sector. It is from this particular point that new research is developed such as Chinchilla, the DeepMind language model.
What is Chinchilla AI?
Taking a futuristic approach, DeepMind researchers and developers focused their efforts on developing an optimal computer model under the name Chinchilla AI.
DeepMind claims that its language model is built around features that claim to be much faster and more accurate than competitors, such as ChatGPT and Gopher.
In this regard, Chinchilla AI presents a system that offers 7% more accurate results at the same cost as other language generators, significantly surpassing Gopher (280B parameters), GPT-3 (175B parameters), Jurassic-1 ( 178B parameters) and Megatron-Turning NLG (530B parameters) regarding the performance of tasks or assessments.
DeepMind states that Chinchilla AI uses less energy and computing systems in terms of inference configuration, which improves and enhances its utilization.
Chinchilla AI was launched in 2022, during the month of March, and so far its accuracy is around the average of 67.5% MMLU.
Chinchilla AI Features
When we talk about artificial intelligence models, we know that it is necessary to regulate the computing budget, so that it can become competent at the same time that it is integrated among the other models whose functionalities and power are determined according to the amount of money invested for the new advances and system updates.
In this regard, Chinchilla AI proposes a fully functional model capable of reaching the scope of the competition with an equivalent computing cost, adding the following aspects to achieve this objective:
– Fixed model size: In the first instance, DeepMind developers carried out an experimentation process with a fixed set of model sizes between 70M and 60M, with a variation in the total number of markers to boost performance (4 variations) Subsequently, the best computing resource was selected, creating a computationally powerful model with more than 67B parameters.
– IsoFLOP: We know that DeepMind's goal was to find a fixed model size with a lower computing cost, yet whose features were equivalent to the competition. In this sense, a computational model was determined with more than 67B parameters and 1.4 billion markers.
– Development of parametric loss functions: once the experimentation was completed regarding the two previous aspects, the DeepMind developers implemented a model with losses in parametric functions in relation to size, generating a computation that would have 40B of parameters, in pairs. with the competition.
Chinchilla AI is recognized due to its extensive capacity in processing large amounts of data quickly and accurately, so its implementation is recommended when dealing with a large amount of information, such as customer feedback or current industry trends. market.
Chinchilla AI is capable of adapting, and learns largely from the user's type of language thanks to the interpretation of new tasks, choosing the ideal artificial intelligence for you revolves around your needs, as well as the functions you want to carry. just.