Tech Please: Making money with AI

Editor's note: "Tech Please!" takes a sideways look at all things science and technology in China, revealing trends you won't hear about anywhere else – from cutting-edge developments to the bizarre and whimsical in the world's most exciting tech market.

Fed up with videos promising quick riches in AI? Move beyond the hype! Explore how AI startups are making millions, and gain valuable insights from industry leaders on how your new venture can seize the opportunities.

In this episode of "Tech Please!" we'll discuss some money-making tech, uncover the challenges and risks tied to this fast-evolving technology, and be well-prepared for the journey ahead.

'Most value generated'

The fastest-growing software's mother company, OpenAI, now has a valuation of $86 billion. That's almost three times what it was six months ago. In terms of revenue, the company has reportedly hit over $1 billion through research collaborations, licensing, API access and partnerships. That is how lucrative this emerging sector could be. 

But how can one enter this sector and take a share of the gains? Must one employ enormous amount of capital and computing power and build a large language model (LLM) from scratch? Absolutely not! Because existing LLMs can be leveraged to your advantage.

For example, ChatGPT API allows developers to integrate the model into their own applications, products or services, according to OpenAI's website. But at a price, of course.

Image generator Midjourney is an example of how a computer is capable of generating high-quality images from simple text-based prompts. The model converts text into a vector and then uses a diffusion model to turn random noise into beautiful art.

We don't know exactly which language model does Midjourney use, but if I were running the business, I'd focus on training the diffusion model.

The company is a 10-people squad that makes $200 million in annual revenue without investors. Their services are offered through subscriptions on Discord, a chat platform, and the packages range between $10 and $120 per month. But much of its revenue goes toward getting more computing power.

Another AI player that has left a mark is the AI video generator Pika Labs. It is a four-people team that makes AI-powered video generator capable of editing videos in 3D animation, anime and cinematic style. It is now valued at nearly $200 million.

The so-called "next Pixar" is a multimodal model that can transfer virtually any prompt – be it a text description, an image, or even a video – to a clip on your demand.

"Generative AI is becoming a commodity literally for everyone," Massimiliano Genta, the CEO of AI startup Metabob, told CGTN.

"They just increased the availability of pre-trained model(s) that has enabled (startup) team(s) to get smaller. That's also one of the reasons we are seeing so many AI startups in the past year," Genta said. 

Metabob is an AI-powered software that helps developers with code reviews. This 10-person startup team crafted their own model from scratch, however, that is based on Graph Neural Networks.  

It seems that the field of generative AI is still unsaturated and there is quite a lot of room for more businesses to enter. You just need to find an area that's niche enough and train your model good enough to be acceptable by the market.

As a practitioner in the field himself, Genta believes that there's still space for AI businesses to grow, especially for industries that are software development-wise or customer-oriented.

"It's where we see the most value generated," he said.

AI's limitation and regulation

Like any other business, AI businesses come with risks. The current landscape of AI, although much more advanced than years before, still faces certain limitations. 

"There's still lots of mistakes created by AI," Genta said. "So whenever the task is too delicate, it's not yet good enough."

"The sector has to develop a lot of compliance," he added.

However, the rapid development also poses challenges to the legal framework.

"The whole trajectory and the whole architecture of AI has totally transformed," said Zhang Xin, the director of the Research Center for Digital Economy and Legal Innovation at the University of International Business and Economics.

"That's why I think maybe sometimes our regulatory framework (should) also be transformed and adapted to this AI new type of technology."

If you want to join the race, making your model compliant to the regulatory framework is something you must consider.

Zhang believes that it is the developers' responsibility to first, develop technologies that can differentiate AI generated content. Secondly, for generative AI tools to be deployed in high-risk scenarios such as healthcare, judicial trials or finance, the bar should be raised high to ensure information accuracy and reliability. 

Despite these risks, the potential rewards of AI businesses are significant, and those who join this sector are likely to be at the forefront of innovation and change in the future.

"In a few years, I think, the emerging methodology will show the potential to redefine the boundaries of what generative AI can achieve and offer exciting prospects for future innovation," Genta said.

Host: Zhao Chenchen

Copy editor: Moosa Abbas

Cameraman: Zhao Wenting

Post production: Zhao Yuxiang

3D designer: Pan Yongzhe

Cover image designer: Gao Hongmei

Producer: Cao Qingqing

Chief editors: Wen Yaru, Wu Gang

Executive producer: Zhang Shilei