Language models are not the only way to implement artificial intelligence (AI) tools. D Transformer gave them the impetus they needed to develop and reach the level of sophistication and refinement they currently have, but There are other strategies which explore other ways of providing machines with capabilities that transcend language
One of them is to model the world (model of the world) able to learn how reality works and simulate how it evolves. Mirror Particle, an American startup, wants to take this idea a step further. And it’s building a model of the world that seeks to understand and predict human behavior. Neither more nor less.
As we can assume, their proposal is radically different from the language model that Anthropic, OpenAI, Google or Dipsic currently offer us. Instead of focusing on identifying relationships between words and text creation, Mirror Particle aims to develop a system capable of identifying the patterns that govern our decisions, understanding how our motivations change, and predicting how we will behave in certain situations. Ambition, of course, is not lacking.
If this project goes well, it could revolutionize marketing. It is what he follows
To achieve its purpose, this emerging company needs more than analyzing what we say: it needs to learn what we do and why we do it. D Transformer A language model is an architecture designed to ensure that a machine can understand the relationships that exist between words or different pieces of information. Its purpose is to identify what is really important to explain the set of data being processed.
The goal of mirror particles is to predict what we might do in a given situation and why.
the heart Transformer This is the process of attention, which, very broadly, allows a model to identify which words in a sentence, what relationships exist between them, and what each one means, as well as the meaning of the whole should be evaluated. The foundational model that mirror particles are working on, however, works in a very different way. It doesn’t work with text, at least not like the language model; It is fed by data that reflects how people behave, how we make decisions and how our preferences change.
Its purpose is to learn patterns that allow our actions to form a representation of human behavior that allows it to be simulated and predicted. Instead of predicting which word will appear next, as language models do, mirror particles aim to predict what we might do in a given situation and why. It sounds interesting, but it is inevitable that we wonder what applications this technology could have.
The most obvious is to help brands predict which products their customers might be interested in and why. If this technology manages to reach the same level of development as language models now, can make a revolution Marketing. However, its creators have assured that a better understanding of humans can improve the interaction between humans and AI systems in the future.
In any case, this technology raises several questions that we cannot yet answer. How do you treat our privacy? To what extent can you avoid the biases and prejudices present in the data you learn to interpret human behavior? This is all boring at the moment. We’ll see where the Mirror Particle plan goes.
Picture | Guillermo Esqueda
More information TechCrunch
In Xataka | Two invisible settings that decide whether your AI is right or wrong
In Xataka | “We’re already at the AI singularity”: Sam Altman, CEO of OpenAI
