We are in a period where AI is a fast-growing technology and the tech world is busy developing different models of it. One guy, a GPT trainer, decided to "clone" his girlfriend using various AI technologies, detailing it in a Twitter thread.
The Process of Making a Digital Clone
Enias Cailliau, the developer behind the Twitter thread, is calling his project GirlfriendGPT. He is attempting to make an AI version of his girlfriend, Sascha Ludwig, and is trying to make the AI version as similar to the person it is based on as much as possible.
He started off by using GPT AI, saying that he has been obsessing about OpenAI's large language model and what it can do. He stated that he was thinking about the ability to create "human-like agents" that behave as real-life humans do but have encountered difficulties.
The developer started off by using Google Bard to describe her personality through her content on YouTube and Instagram, saying that "cute words" give him writer's block. These were used as Cailliau modified the prefix to include his girlfriend's personality.
Given that Ludwig was an influencer, there was a lot of content to go with. Through a text prompt, Cailliau asked Google Bard to look at her videos and provided links to her YouTube channel and Instagram profile.
In order to add a voice to the clone, he proceeded to use Eleven Labs, saying "Clones need voices, so I gave her a voice." He used YouTube Shorts videos posted by Ludwig and then uploaded them to the AI text-to-speech tool, as mentioned in Popular Mechanics.
The AI clone was tested out in certain situations to see how it will respond, and reports say that the GirlfriendGPT was answering more like the person it was trained on. He also added a SelfieTool to his agent to generate selfies using Stable Diffusion.
He even believes that there is a future where everybody has personal AI companions across devices. With that being said, Cailliau also posted the source code so others may use it as well and create their own digital clones.
Read Also : Amazon Products Now Have AI-Generated Reviews
Are There Risks to This Technology?
Like all innovations, there are. For one, it can be used to spread misinformation. A convincing clone can be exploited by threat actors if there are no measures put in place to clearly indicate the users are interacting with AI.
The same concern applies to deep fakes, wherein images, audio, and videos have become more and more convincing and are harder to detect. When the photo of "Balenciaga Pope" spread, people were actually questioning if it was a real photo.
Although professionals and experts have no trouble spotting indicators of AI-generated content, many users online still can't. However, a digital clone can be put to good use, just as technology is intended for.
People can create digital versions of lost loved ones by using old footage of them and photos to create a likeness. Recordings of their voice can also be uploaded to AI text-to-speech tools so that the clone would respond in that person's voice.
As the creator of the GirlfriendGPT pointed out, even though it might look like a "shallow project," it already has features that can be used to build something professional like an automated support agent.
Related : 5 Risks of AI-Generated Content