TIME

This article was originally published in TIME’s newsletter Into the Metaverse. Subscribe to receive a weekly guide on the future of Internet. Past issues can be found here.

Imagine your dream home. Perhaps it has high ceilings and a fireplace, or large windows that overlook a tranquil lake. Perhaps it is a metallic dome with breathing metal that sits on a fiery world and is populated by alien butlers. Imagine if you could write about these houses and then instantly enter virtual versions with all your friends.

Recent developments in AI such as ChatGPT or DALL-E have made it possible for users to create their own immersive worlds. Three new text-to-3D generators for text were announced in the fall: GET3D from Nvidia and Make-a-Video by Meta. DreamFusion is from Google.

Metaverse builders already use text generators such as ChatGPT, which responds to texts with remarkable poise and intelligence. Visual generators such DALL-E, which creates images from text prompts, are also used to generate new worlds and designs. According to industry insiders, these AI technologies are crucial in building virtual worlds that can be customized and detailed. They also hold the key to creating metaverses that people want.

Rev Lebaredian, Nvidia’s VP for Omniverse, says that “we are able to fill it with interesting stuff because everyone is capable of taking pictures, recording videos, and writing words.” “If you want to create a 3-D internet then you must have people creating content. The only way we can make that happen is with AI.

Change workflows Although AI tools are not yet fully integrated into metaverse creation, they are playing an important, though somewhat minor, role. ChatGPT is used by metaverse builders for brainstorming, writing code, and composing decks and emails.

To find out if ChatGPT had impacted Sean Ellul, co-founder of Metaverse Architects’ 3D development studio Metaverse Architects. The email he sent was a five-paragraph, well-written piece about his experiences with the technology. The catch was that the email contained a fourth paragraph that revealed ChatGPT had written it. Ellul had entered the following prompt into ChatGPT and then sent it (with very little editing):

Write an email to Andrew from TIME about how Metaverse Architects uses chat GPT to create articles, brainstorm code and plan new projects. It’s even used to send emails like this one!

The email that followed was stoic, but completely credible and informative. This email was a testament to the basic powers of ChatGPT as well as the many ways Ellul has integrated it into his daily life. Ellul claims that ChatGPT is used by him to modify design ideas, solicit market techniques, create architectural blueprints and many other tasks.

Ellul states, “I cannot stress how helpful that has been to us.”

Ellul responds to my question “How often do you use ChatGPT?” “How often do you search for something on Google?” The AI is now integrated into his workflow, and the way that he receives information.

A Midjourney image was created using the AI program Midjourney. Sean Ellul asked for a floor plan. Add furniture.

Sean Ellul

Artificial intelligence services have a significant impact on metaverse creation, in particular by bringing characters and worlds to life. The virtual worlds of the future must have rich details in their architecture, landscapes, and objects. Artificial intelligence can build these environments much quicker than humans can by using prompts. Ellul describes the process as an enhanced version building a Pinterest moodboard. You search for an aesthetic or design and then narrow in on a particular iteration to build a larger vision.

AI users don’t need to have a background in architecture or design. They can create virtual environments, first rooms, then buildings, and then entire worlds. Ellul states that AI will be a key component in creating these generative experiences that can easily be customized to meet the needs, interests, and desires of users.

Virtual people will be able to fill Metaverses just as video games have non-player characters (NPCs). However, unlike NPCs who had pre-set scripts for their responses, AI characters can respond organically to your questions, much like ChatGPT.

These types of AI characters are already available in text form. The website character.ai allows you to converse with AI versions such as Billie Eilish, Socrates, and Elon Musk. This type of AI could be placed in a virtual world where they can act as guides, companions, or nemeses.

Ellul claims that his company now generates new characters for video games faster than ever before. He shows me an image of a woman with features that almost look like photographs. He created her using nine people on the Midjourney text-to-image algorithm generator. The algorithm was tweaked bit by bit. After he is satisfied with his face, he logs into ChatGPT and typed: “Generate a backstory of a 24-year old woman from a small village in a fantasy setting, who works at an Inn.” Within seconds, Elspeth’s entire history has been created by the AI. She was a “farmhand, blacksmith’s apprentice and even a barmaid in bustling cities.” She always felt that something was missing, no matter where she went.” (All in, a fairly generic but good starting point for a fairytale.

However, Elspeth can be created as a fully-fledged 3D figure that conforms to physics and moves fluidly. Google and Nvidia have both announced text-to-3D models, although they are still in the early stages. Nvidia researchers announced Magic3D in November. This AI can create 3D models using prompts such as “A blue poison-dart snake sitting on a waterlily.” However, it still takes 40 minutes and has limited data about the frog.

Lebaredian said that Nvidia is currently training AIs in Omniverse’s 3D simulation engine, so they can quickly learn how to label and build objects, navigate through spaces and respond to user questions. When Lebaredian was asked when a regular user would be capable of building the lake house I pictured above, he said that AI progress doesn’t happen linearly. There are lulls as well as explosions that make it hard to predict. He says, “We’re pushing all the boundaries and it’s happening very quickly.” “I think that in ten years, most people will be able create high-quality 3D content by simply talking to the computer. It’s likely to happen sooner than we think.

A illustration by the AI Midjourney. This was created from Sean Ellul’s prompt “4k HD portait beautiful blonde woman with deep stare and electric eyes”

Sean Ellul

Warning signs It is easy to see people being abused by AIs, getting bad advice or developing emotional dependence on them. Ellul admits that this area is fraught with serious risks that must be taken into consideration. He says that deepfakes will likely reach a new level. “This is where the potential misuse begins: We are starting to realize that it is possible to just create identities on-the-fly.”

ChatGPT responded to my colleague Billy Perrigo when he asked about ChatGPT’s human-like abilities. He said: “It is important to remember we are not humans, and we shouldn’t be treated as such. While we can help with information and assistance, our AI should not be relied upon for complex decisions or difficult tasks.

AIs can have all kinds of biases, based on what data they are trained from, and produce a lot of harmful content including hate speech and misinformation. TIME recently discovered that OpenAI, which created ChatGPT, and DALL-E used Kenyan workers to review harmful content for their algorithms.

Read more: OpenAI used Kenyan workers on less than $2 per hour to make ChatGPT less toxic

Despite billions of dollars in investment, AI companies continue to move forward. OpenAI is currently developing its next-generation AI, GPT-4. This could be available as early as this quarter, according to the New York Times. It is being funded by Microsoft in a multi-billion-dollar investment. OpenAI’s tools will gain more use, so the company’s approach to ethical issues will be more important. This is similar to how Facebook’s speech policies played a major role in global politics in the past decade.

Ellul states, “I agree that better technology would make bad players more effective.” “Bad and good actors can have influence if the world is more connected,” Ellul says.

It is still unclear whether AI technology will become a tool for ultimate empowerment, turning idea-generators into virtual creators or writers into engineers. Or if it will be used as a vehicle to commit identity theft and abuse. Ellul states, “It is important to educate ourselves so we don’t either suppress these technologies when they have potential benefits or allow them to go too far when there are potential risks.”

TIME: More Must-Reads

Write to us at letters@time.com

————————————————————————————————————————————————————————————

By: unknown
Title: What ChatGPT Could Mean for the Metaverse – TIME
Sourced From: time.com/6250249/chatgpt-metaverse/