What Creators and Brands Need to Know in the Age of AIGC

An interview with Sanxiangshi that went well beyond the technology, into how AIGC can amplify creators and empower brands. May we all use AI to raise our efficiency and income this year, rather than lose our jobs to it.

Translated by AI

Editor’s note

Thanks to Sanxiangshi for the interview, in which we talked about a lot beyond AI technology: how to use AIGC to amplify creators and empower brands. In the new year, I hope everyone can use AI to raise their efficiency and income, rather than have AI take their jobs (tongue firmly in cheek).


Interviews with 100 Creators, livestream no. 2

Our guest was Meng Shengyu,

who talked with everyone about “everything about AIGC”.

Beyond the development and applications of the technology,

how AI and humans can live together in harmony

was a theme that ran through the whole livestream.

This discussion of “reality” and “visions of the future”

had the online audience thoroughly excited.

Simon Meng (Meng Shengyu)

Cross-disciplinary researcher in architecture, art, AI and biology; PhD candidate in architecture at the University of Innsbruck (UIBK); MArch from University College London (UCL); teaching assistant and guest critic for technical courses at UCL and UIBK. His work has been exhibited in London, Edinburgh, Venice and elsewhere; collaborating artist for National Geographic China and Chow Tai Fook digital collectibles.

  • Livestream highlights
  1. How is the AIGC market developing?
  2. Which AIGC tools can raise your productivity?
  3. What new professions is AI creating?
  4. Taming AI with AI? Using AI to create “another me”?
  5. How can brands benefit from AI?
  6. Should brands move into the metaverse? (A big but not distant question)

Open topic: what will consumption look like in a future built on AI?

The anime-avatar generators that recently went viral on Xiaohongshu and Douyin brought AI into public view, and ChatGPT flooding everyone’s feeds let people feel its magic even more directly. Before that, though, AI seemed confined to professionals, and the public couldn’t imagine what it could do…

>>>>>>>>>>>>

Q1

How is the AIGC market developing?

AIGC used to demand a lot from both software and hardware, so it was hard for the public to pick up quickly and stayed mostly an experiment among professionals. Yields were low and quality uneven, so even professionals spent a lot of time and effort producing a piece. As machines got better and the technology advanced, the barrier to AIGC came down and yields and quality improved markedly. Together, these two factors brought AIGC to a tipping point, from an experimental state to something the public can use.

Simon Meng, 2022

Guardians of the Ocean’s Natives × National Geographic China (AI blending a National Geographic photographer’s work with an iridescent oil-film style)

Q2

Which AIGC tools can the public master?

  • Text generation tools
  • Image generation tools

Speaking of mature AI content tools on the market that can raise productivity, let’s first split AI into several areas: image generation, music generation, video generation and 3D asset generation. Going a bit further, there’s also the generation of interactive behaviour in metaverse games, such as character interactions.

Of these, the closest to maturity are conversational text generation and image generation.

Image generation

Everyone knows plenty of image generation tools by now. If we only count those usable in a browser, there’s Midjourney, DALL·E 2, and Stable Diffusion (whose web version is called DreamStudio).

Personally, though, I use Stable Diffusion the most, because it’s the only one of the main models that’s fully open source. As a result, many independent developers build all sorts of derivatives on the open-source version to meet every kind of need: whatever feature you can think of, someone will build it. And you can run it entirely on your own computer, so you’re not limited by server performance or by the service provider’s licence terms.

  • Here’s something many people miss: if you generate an image on Midjourney and you’re not a paying user, then under its cloud service terms you don’t own the full copyright to the image. You only get a non-commercial licence from Midjourney that requires attribution (CC BY-NC 4.0). Many people may not have noticed this.

Simon Meng, The Breathing House

Text generation

This was a bit behind, but then ChatGPT appeared this month (December), and “frightening” really is the right word. Programming, answering questions, writing articles: it handles all of it, in Chinese and in English. It still has one problem, though: it mainly runs in the cloud, and the model is large, reportedly with hundreds of billions of parameters. If a model like that can’t be brought down to individual machines, people’s use will remain limited by the service provider. So judged by its output, ChatGPT is more than good enough; judged by how much an individual can control and customise it, it still falls a little short.

The other areas I mentioned, such as music, video, games and assets, have all made some progress, but none has yet reached the stage of images and text, where the public can take part very easily and get a fairly satisfying result. They’re still semi-experimental.

Q3

What new professions is AI creating?

  • Prompt engineer
  • CAO (Chief Artificial Intelligence Officer)

Today’s AI tools basically respond to the instructions we give them. But the same kind of question doesn’t have just one answer, and the way we ask may not go beyond certain limits. So there’s now a kind of person called a prompt engineer. The job doesn’t even involve changing code; it’s about trying different ways of asking the AI questions. There’s a project on GitHub called awesome-chatgpt-prompts if you’re interested.

Because some large NLP models aren’t open source, people can’t analyse them in depth, so all interaction with these big models is limited to how you ask questions or hold a conversation. That has turned it into a subject worth studying in its own right, since it hasn’t yet become something people pick up naturally and easily.

In future, there may be a new title in companies: “CAO”.

We know about CEOs and COOs; a CAO, with A for AI, would be a role that helps the boss “talk” to AI better. It’s similar to the prompt engineer above, but the “conversation” would cover more dimensions of interaction with AI, such as choosing models, tuning parameters, fine-tuning and setting up how people interact with them.

For now, humans and AI still need someone in this guiding role, so that AI can serve humans better.

These sound like professions with good prospects, but any profession needs time to prove itself. For example, when people first generated images from text, they needed very long prompts to get decent results; now they’ve found a tool for that, since ChatGPT itself can write excellent image descriptions. So if this role appears and is still stable five or six years later, it may truly be a new profession. But it could just as easily be made obsolete by another AI within two or three years, or even one or two. We don’t know yet.

Simon Meng, AI-generated

Cyberpunk London reclaimed by plants, 2021

Q4

Taming AI with AI?

Using AI to create “another me in the world”

  • Tools that generate interactive character behaviour
  • Replicating another version of yourself

As mentioned above, perhaps in future we can tame AI with AI (for example, using a text generator to drive an image generator). Could we then train lots of AI tools to create countless models with the same abilities and behaviour as ourselves, to help us become more efficient?

We discussed many possible paths to this in the livestream; here are a few for discussion:

Among AI tools, there are also applications that generate interactive character behaviour, of two kinds. One is a website called character.ai: once its model can hold a conversation, users can train it with relatively few parameters and samples to shape a conversational AI with a distinct personality. So on that site you can chat with King Arthur, an alien, a celebrity, a detective novelist and so on. Each conversational AI has its own personality.

The other is the many NPCs we already have in the metaverse and in open-world games. People used to think of NPCs as puppets, a byword for lack of intelligence. But if we want to be immersed in these worlds, better graphics aren’t enough; the characters need to be more interactive too. So many games are now experimenting with, and some may already have, NPC interactions driven by AI. These don’t follow fixed response logic; given a role or a range of activity, they have a degree of autonomy.

There are plenty of face-customisation tools now, but they remain at the level of 2D or 3D appearance. If you want to use AI to create characters, shaping them in 2D or 3D matters less than you’d think. What matters most is the NPC’s “personality”: the “emotions” and behaviour patterns you perceive when talking to a virtual person.

On “another me in the world”

There’s a case in which someone used AI tools to recreate a deceased relative, letting the virtual person imitate his tone and personality and talk with his family (source: I “resurrected” my grandfather with AI to say goodbye). But that’s limited to fairly small tasks. Actually replicating a person with AI is complex and involves a great deal of law and regulation. So wholesale replacement is certainly impossible today, but replacing someone in certain parts of a task, in certain emotional roles or in some small tasks is possible.

Simon Meng, Boundless Starry Sky × Chow Tai Fook, 2022

(AI generated 5,000 images of starry skies that don’t exist, and 1,500 videos roaming through them, from NASA telescope images.)

Q5

How can brands benefit from AI?

  • Use AI to improve communication and interaction with customers
  • Use AI to support operations

How can brands use AIGC tools? Simon’s suggestion is to bring AI into interactions with customers.

Many brands now like “co-creation”. Honor of Kings, for example, often runs skin co-creation: it releases two designs and lets you pick one to develop further. But you’ll find that most so-called co-creation just gives consumers a very simple choice, without real customisation. The reason is that, facing a customer base where everyone is different, it’s hard for a brand to offer unlimited high-quality options, since each one takes a lot of time to make.

But with AI generation tools, a brand can have a very efficient, intelligent customer service that guides and inspires consumers to say what they really want.

Often consumers don’t know what they want, or can’t express it. If an AI sits in the middle and quickly expands on or visualises their ideas, I believe it can close that gap. And with pre-set materials and feedback mechanisms, it can keep collecting customers’ genuine reactions, which can serve as data and a basis for improving products.

For example: “What kind of moisturiser would you like?” The consumer might say, “Something more hydrating.” The AI might prompt further: “For hydrating, would you prefer something dewy or something lighter?” That can draw out a lot of information.

At the same time, AI tools can be combined with operations.

Take the game-skin example again. Could the brand build a model that generates character skins for players to play with? Players share their creations in the community and vote, and the brand releases the most popular design. The player who designed it could get a set of SSR skins or similar. That gives operations a much more varied toolkit.

There’s also user-generated derivative work. Many users may not have creative skills themselves, but with AI tools they can create.

Simon Meng,

AI-generated video of the evolution of cyberpunk sea creatures

  • Understanding what the metaverse fundamentally is
  • Advice for brands
  • On building the metaverse

What is the metaverse, fundamentally? It’s that more and more of the activities that used to happen offline are moving online.

It started with shopping and text chat; in future, many gatherings, and many meetings like today’s conversation, will move online too. But for now, what we have is fragmented. We chat in Tencent Meeting or WeChat Channels, watch videos on another site, and play games somewhere else. In future we may need a platform that brings all of this together.

As for brands, it’s actually quite hard to answer, because the structure of the metaverse hasn’t taken shape. Brand marketing today starts with capturing traffic, either offline, in shopping districts and commercial streets, or online, by finding KOLs in each field to reach their audiences. That strategy works because we know where offline and Web 2.0 traffic is and how people consume. But the metaverse has no unified platform yet, and the form its activities will take is still vague, so the question is hard to answer: its new social structure hasn’t formed.

Advice for brands:

I’d advise brands to keep their noses to the wind. When online shopping platforms rose, those who got in first won the dividend of organic traffic. So brands should keep sensing: which metaverse platform will take off? Which form will become mainstream, blockchain games or NFTs? Which form might come to represent the metaverse? Nothing has yet changed the rules of the game, so keep experimenting and keep sensing.

On building metaverse platforms:

There’s currently no global metaverse platform (though some underlying technologies are widely recognised). China’s big tech companies have the initial conditions, in money, technology and people, to develop one. But the road from development to success is long, there’s a great deal to do, and since it’s something new, its development may not fully follow existing trends. We have to keep watching.

On how future Web3 platforms will take shape, we raised two key points in the livestream: building the infrastructure, and migrating users.

(For more discussion of building metaverse platforms, watch the livestream replay.)

Conclusion

AI technology will keep developing, and our imagination about technology is no longer limited. To humans, AI is a tool, a force that drives change in our times.

AIGC has greatly increased the efficiency of content creation, but the “soul” of good content still needs to be infused with emotion and observation of life, and AI tools can’t replace that.

We’ll close with a line Simon shared in the livestream, in the spirit of the Tao Te Ching:

Technique without the Way stops at technique;

with the Way but without technique, technique can still be sought.

We’ll keep following developments in AI and AIGC.

For more on the livestream,

scan the code and tap “Livestream replay” on our Channels homepage.

Previously in the Sanxiangshi column

Content planning: Jianjian

Editing: Jianjian

Visual editing: Bessie & Shanshan

Search