Linear Capital Perspective | GPT Is a Pandora's Box or Aladdin's Lamp, Depending on Human Understanding and Action on Technology

Last week, Harry Wang, founder and CEO of Linear Capital, appeared at the "2023 China Investment Annual Conference" hosted by ChinaVenture and ChinaVenture.com for a discussion titled China Calls for GPT, sharing his views on ChatGPT's development. Fellow panelists included Gao Xue, CEO and managing partner of Baidu Venture; Hu Bin, founding partner of Zhence Capital; Liu Kai, partner at 5Y Capital; and Xiong Weiming, founding partner of China Growth Capital. The session was moderated by Zhang Suyang, founding partner of Volcanics Venture.

When the moderator raised the notion that ChatGPT had opened "Pandora's box" — with some suggesting we pause development for six months until consensus emerges and policy frameworks mature — he asked whether we've indeed opened a Pandora's box, and if so, whether countries might engage in a new arms race around GPT and what methods they might use to constrain each other.

Harry Wang responded: "In terms of effect, it's roughly equivalent to that, but we don't treat it as Pandora's box. Instead, we've gained an Aladdin's lamp that can answer infinite questions.

"From a practical standpoint, it really does dramatically replace all repetitive, formulaic mental labor. We believe it won't take that long — maybe just three years — before much knowledge work, even meetings like this one, could be handled by five AI robots trained on our historical data, having the conversation for us. The gap between these capabilities emerging and becoming everyday, universally accessible products is precisely where the 'AIGC + applications' opportunity lies. So whether we call it Pandora's box or Aladdin's lamp, we need to take its capabilities seriously, especially those of us in knowledge work.

"We also have this concept of Physical AI. Today much of what we're doing is Digital AGI — digitization. We've also invested in many robotics-related projects. This area has entirely new training methods. Following the same playbook, what you train is a Robotics-native Physical AGI. Combine that with the ChatGPT we're discussing today — which is Digital AGI — and from a negative perspective, you get Skynet and the Terminator: robots that are both intelligent and capable of self-replication. Concerned about production capacity? In the future, production lines won't need human workers — just robots. Agile Robots, one of our portfolio companies, has done extensive work in this area, gradually replacing tasks that many humans cannot perform.

"So I believe the transformation is only just beginning. Used poorly, it's 'Pandora's box'; used well, it's 'Aladdin's lamp.' Returning to the second question: competition and cooperation are inevitable. When many new things emerge, their impact on humanity's future and on relations between nations — how this develops depends first on how we think about it, and second on what we do. If everyone treats AGI as an instrument of confrontation, the outcome will certainly be mutual antagonism. If everyone approaches it with reverence, the result might be different. For example, recent US legislation prohibits AI from making final decisions on nuclear weapons — effectively mandating a safety mechanism across all operations. If everyone can do this, what we should worry about is the ending of the movie Terminator."

When asked about the consequences of large-scale AGI deployment and whether humans will be replaced by AGI, Wang noted that two or three months ago, Americans were quite concerned about this — seeing ChatGPT's acceleration from emergence to rapid development changed many people's assumptions about AGI's timeline. "Today's Digital AGI has a major limitation: its inputs and outputs are organized by humans, text-based, formed from relationships between text and text, token and token. This is really just one type. Even what we call multimodal is still between digital data types.

"But humans have an enormous capability that no AGI currently possesses: continuously taking in and responding to information from our surroundings. Today's Digital AGI lacks this, but this capability will necessarily be realized in Physical AGI. When Digital AGI and Physical AGI combine, you'll find robots constantly processing external inputs and outputs, drawing conclusions, determining next steps, and acting on them. In terms of interfaces, this may not be distinguishable from human behavior. But many of the conclusions and decisions they make won't necessarily align with human interests. Why is the final step of ChatGPT called human alignment? They're already concerned about this in the Digital AI domain. I believe once Physical AI is added, this concern expands a hundredfold, a thousandfold. That's what's frightening."

PS: For more coverage of this discussion, click "read more."

About Linear Capital

We are hiring for multiple positions on our investment team. For details, please click here.

Linear Capital is an early-stage investment firm focused on "frontier technology + industry" — frontier technologies such as data intelligence, digital new infrastructure, next-generation robotics, and new technological transformations in traditional sectors (biomedicine, materials, energy, etc.), applied across vertical industries to dramatically improve efficiency, solve pain points, and enable industrial upgrading. We capture excess returns through substantial increases in industrial value. We currently manage ten funds with total AUM of approximately $2 billion.

Our investment stage focuses primarily on leading angel to Series A rounds, with typical check sizes of $3-8 million or RMB equivalent.

To date, we have made early-stage investments in over 120 startup teams including Horizon Robotics, Kujiale, Sensors Data, Tezign, Rokid, Guandata, and Agile Robots. The aggregate valuation of Linear's portfolio companies is approximately $20 billion.

In the near term, Linear Capital is working to become the premier "Data Intelligence Technology Fund." Over the long term, we aim to build the most influential "Frontier Technology Application Fund."