“The physics of my days has died”. No lo digo yo, lo dice Einstein. Well, no el de verdad, sino el virtual avatar of Albert Einstein created for Character.ai. This company created by two former Google engineers wants to give a surprising turn to the revolution generated by ChatGPT with one of the best projects based on this technology.
What is Character.ai. It is a project that, according to its creators, “is making reality from the dream of science fiction of open conversations and collaborations with computers”. The platform allows you to choose a bunch of characters of great historical and actual relevance, and tries to simulate the form in which they will answer our questions.
Imitando a cualquiera. It is systematic if it is based on the use of neuronal language models: a supercomputer is trained with large amounts of text transcribed or published by these personalities, and learns to generate coherent responses linked to this training.
El origen. In 2015 Daniel de Freitas, one of the two creators of Character.ai, was working at Microsoft. He read a study published by Google Brain scientists. In it he detailed how a neuronal network allowed him to learn to converse by analyzing transcriptions of dialogues and diverse texts.
De Freitas ended up working at Google in 2017, and there he worked on his own chatbot in parallel to his official work. The result was Meena, who ended up becoming LaMDA. A companion of his came to affirm that LaMDA had conscience of itself and was fired. Google, which knows well its potential, has preferred not to take this project out of the laboratory in order not to damage its reputation. De Freitas and his company Noam Shazzeer did not think so, and they left Google to work on this project.
It’s not real, it’s just entertainment. The service managers themselves advise: “We have to say that an hallucinatory supercomputer is not a source of reliable information. Even so, we hope that Character.AI will turn out to be a useful tool for imagination, brainstorming, language learning and an infinite number of other fines that nosotros mismos aún no hemos imaginado”
But the answers are credible. It may only be entertainment, but the answers provided by this chatbot with multiple personalities are really reasonable. We asked Einstein if conventional physics had died, and he replied that the quantum theory had changed everything… but that the physics of his days had died.
Shakira is not a saint. The same thing happened when conversing with Shakira’s avatar, to which we asked him why he had created no one, but several revenge songs against Piqué. La máquina nos contestaba que “necesitaba express emociones en letras como siempre hago… No tengo deseos de venganza. No es mi naturaleza. Pero tampoco soy una santa”. That last sentence gives a certainly believable touch to the form of expression of this chatbot.
Design your own avatar. In addition to conversing with famous fictional characters, you can create your own. This option can be really remarkable, since we can establish different attributes and “train” it in conversations by assigning stars to the questions that best fit that personality. At the moment, that’s right, it doesn’t seem possible to force the training with texts that one could find with this character.
Privacy and other problems. The technology is surprising, but it is not exempt from the same problems that have affected CoPilot or Stable Diffusion. If you can train freely, will you move? Does this imply problems of copyright or intellectual property?
Not only that, of course: doubts also arise when casual conversations end up being used for other needs: these engines can also be used in harmful ways, helping to generate fake news or more credible deepfakes that work in text mode.