Artificial intelligence is on everyone’s lips. In Spain it is increasingly present, with new followers and models every day. From models that convert text to images like DALL-E to chatbot with a great capacity for conversation, like the already famous ChatGPT. But what if there was an AI that mimicked famous people? Even more; What if you could imitate public figures both living and deceased?
It is what he proposes Character.AI, The last model chatbot of artificial intelligence that not only seeks to be a conversational AI like ChatGPT, but literally speak like public figures. Some of the examples include Elon Musk, Bella Poarch (one of the most famous TikTok influencers) and even with fictional characters, like Mario from the video game Super Mario 64.
The magic of Character.AI, according to an extensive report from the New York Times, is that the model that gives life to this website is not simply capable of imitating characters, but is capable of to imitate absolutely anything.
Character.AI: talk to celebrities
Character.AI is a website that is currently in beta, and is based on neural language models executed by a supercomputer. The idea is simple: build a dialogue with a character, creating the illusion that you are actually talking to that character, be it famous or even real.
Created by an AI developer who used to work for Google, the neural networks that power the web are made up of a computer that processes vast amounts of data. referring to the characters he imitates. Thus, he has collected information in the form of articles, news, books, and other texts that detail what these characters are like. Additionally, the system can combine disparate concepts and thus provide a greater breadth of topics to be able to manage.
Chatbot with Elon Musk.
On the web we can see characters that are created by other users of the page to use the AI model embedded within Character.AI. Unfortunately, this results in several rather problematic details. One of them is the presence of highly questionable public figures, as would be the case of Kanye West or himself Adolph Hitler.
Although there are characters controlled by the web, as would be the case with Elon Musk, there are chatbot who give completely inappropriate messages. For example, talking to the chatbot of Adolf Hitler, we asked him about the Jewish population. He chatbot answer openly that they shouldn’t “get in your way”. In the absence of a response, the chatbot reply that only is doing what is “necessary” to keep Germany safe.
Conversation with Adolf Hitler.
“Well, I don’t like them and I don’t think they should be in my way, so I’ll make them disappear my way once I take control of Germany.” Others chatbot also include characters like donald trump which also sends us quite questionable messages related to racism and xenophobia. In many other cases it is represented quite well, as occurs in the chatbot by Elon Musk.
To this is added that certain characters complemented by the web itself change their context or their way of speaking, surely to prevent them from spreading their messages. In the case of Kanye West, famous for launching antisemitic messages and proclamations and defend issues about the American extreme right, the chatbot He responds to us in a kind way, assuring that he would give his life by racialized and Jewish people.
Chatbot with Billie Eilish.
For the rest, we have tested a good number of chatbots. billie eilish, Tony Stark or even the recently deceased Elizabeth II, Queen of the United Kingdom, are just a few examples. And again, sometimes the conversation was so contrived that it didn’t feel like it was real. At other times, the chatbot he was able to nail the personality that at least these characters display on the internet. Perhaps the most credible has been billie eilish, who practically speaks like the artist on social media.
This model is born from the hand of engineer Daniel De Freitas, former Google worker on the YouTube part. Over time he began to develop his own chatbot, along the way training a neural network that used much larger collections of information and dialogue. In fact, it is not limited to imitation; Characters.AI also has sections with psychologists, coaches and even games.
[VALL-E, la inteligencia artificial de Microsoft que imita tu voz con solo escucharte 3 segundos]
Thanks to his position in Google, he was able to use credits to run his software within Google’s network of computer data centers, as well as borrow other engineers to use their infrastructures. It is not for less, since the sections of this web are almost infinite.
His efforts led to the creation of Meena, a chatbot that would end up becoming nothing less that LaMDA, artificial intelligence that Blake Lemoine, a Google engineer, described as being sentient. De Freitas and other colleagues such as Noa Shazeer left Google and founded Character.AI, with the idea of creating a unique system capable of imitating practically everything. Of course, requiring a large investment of billions of dollars and a computational data analysis practically out of the reach of any user.