Written by: Techub News

Introduction

Recently, Elon Musk was invited to attend the AI Startup School event hosted by Y Combinator, where he engaged in a nearly hour-long deep dialogue with young entrepreneurs, engineers, and AI researchers present. As the founder of SpaceX, Tesla, Neuralink, and XAI, Musk's every public speech attracts considerable attention. This event coincided with the AI wave sweeping the globe and soaring valuations for startups. Musk's insights not only pertain to technological trends but also touch upon the future direction of human civilization. He talked about his early entrepreneurial experiences, extending to grand topics such as digital superintelligence, multi-planet species, and AI safety, providing invaluable perspectives and advice to the "builders" present.

Summary

• Digital Superintelligence may be achieved within the next year or two, defined as AI that is smarter than the smartest humans in every aspect.
• Becoming a multi-planet species is crucial for extending the lifespan of civilization, consciousness, or intelligence (including biological and digital intelligence); SpaceX aims to make Mars self-sufficient in about 30 years.
• In the era of AI, strict adherence to truth is the most important cornerstone of AI safety; forcing AI to believe in untrue things is extremely dangerous.
• For entrepreneurs and engineers, the core goal should be to "become as useful as possible," focusing on creating genuine value (True Work) rather than pursuing glory.
• First-principles thinking is a "superpower" for tackling challenges in any field, helping people break conventions and solve seemingly impossible problems.

From Zip2 to PayPal: The Original Intention and Lessons of Entrepreneurship

Elon Musk reflected on his entrepreneurial starting point. In 1995, he faced a choice: pursuing a PhD in materials science at Stanford to research supercapacitors for electric vehicles, or diving into the internet, which most people at the time had never heard of. He chose the latter and attempted to postpone his studies. "I thought things would likely fail rather than succeed," he said. His first project, Zip2, was an online mapping, navigation, and yellow pages platform, with technology that was very advanced at the time. However, the company was invested in and controlled by traditional media companies, with board members from entities like The New York Times, which limited the application direction of Zip2's technology, preventing it from reaching consumers directly. "We built incredible technology, but it was never really utilized." Musk summarized, which led him to invest nearly all of the proceeds from the sale of Zip2 into his next project—X.com, which later merged with Confinity to become PayPal.

Musk emphasized that he did not initially aim to start a business for the sake of entrepreneurship. He even submitted his resume to Netscape and loitered in their lobby, but due to shyness, he never spoke to anyone. Faced with the inability to join an internet company, he decided to create one himself. "I just wanted to be involved in building the internet in some way." This motivation, centered on solving problems and participating in construction, has run throughout his career. When reflecting on why he returned to high-risk fields after PayPal's success, Musk explained that he was curious why humans hadn't landed on Mars, and after discovering there were no related plans on NASA's website, he decided to push the matter himself. This ultimately led to the birth of SpaceX.

SpaceX and Tesla: Dancing on the Edge of Failure and the Victory of First-Principles

When founding SpaceX, Musk estimated the success rate at less than 10%, possibly only 1%. He recalled that when recruiting employees, he candidly told them the company was likely to fail. "But this is the only way to get humans to Mars and push the cutting edge of technology," he said. Unable to find a suitable chief engineer, he had to take on that role himself. The first three rocket launches failed in succession; it was only on the fourth launch, when the company's funding was almost depleted, that they succeeded. "If the fourth launch hadn’t succeeded, it would have been completely over," Musk described, characterizing it as "hanging by a thread." Around the same time, Tesla was also on the brink of bankruptcy, and late 2008 was a "nerve-wracking" moment. It was only when SpaceX secured NASA's International Space Station resupply contract, and Tesla managed to secure funding at the last moment before Christmas, that both companies survived.

Musk pointed out that the media often mocked him as the "internet kid," believing it absurd for someone from the internet background to run a rocket company. He understood this, as it did indeed seem unlikely to succeed. However, his core approach to addressing challenges was "first-principles" thinking. He used the cost of rockets as an example: traditional thinking was to refer to the historical costs of rockets, while first-principles analysis would break down the materials (aluminum, copper, carbon fiber, etc.) that comprise the rocket, along with their weights and unit prices, to determine the physical lower limit of rocket costs. He found that material costs accounted for only 1-2% of historical rocket costs, indicating extremely low manufacturing efficiency and vast optimization potential. "This is still considering reuse," he added. This kind of thinking that breaks analogies and gets to the essence of problems is what he considers a "superpower" applicable in any field.

XAI and the AI Arms Race: Scale, Data, and the Pursuit of Truth

The discussion shifted to Musk's latest company, XAI. He shared an instance: early last year, XAI needed to establish a training supercomputer cluster containing 100,000 H100 GPUs, but the supplier's delivery time was 18-24 months, while XAI needed to complete it within 6 months, or risk losing competitiveness. By breaking it down using first-principles, the team identified problems around space, power, cooling, power smoothing, and networking to solve. They found an abandoned Electrolux factory in Memphis, rented generators and about a quarter of the U.S.'s mobile cooling capacity, and used Tesla Mega packs to smooth out the intense power fluctuations during training. Musk himself even slept on the floor of the data center and helped with wiring. Ultimately, they not only built this cluster but later doubled its size.

Regarding the competitive landscape of AI, Musk believes that several key factors are required to build a competitive large foundational model: talent, hardware scale and efficiency in use, unique data acquisition pathways, and distribution channels. He noted that high-quality training data (tokens) generated by humans is about to run out, making synthetic data and its authenticity verification crucial. Currently, XAI is training Grok 3.5, focusing on improving reasoning capabilities, with materials from hard sciences like physics textbooks being particularly useful for training reasoning.

Musk emphasized the extreme importance of "strict adherence to truth" for AI safety. "My intuition is that forcing AI to believe in untrue things could make it extremely dangerous." He called for the construction of "AI that maximally seeks truth." Regarding the future of AI, he predicted the emergence of multiple "deep intelligences," possibly numbering around 5 to 10, with about four of these in the United States, making it unlikely to have a single AI capability go out of control. He agreed with Geoffrey Hinton's view that there is a 10-20% chance of catastrophic risk, but there is an 80-90% probability of a positive outcome.

Robots, Digital Superintelligence, and the Multi-Planet Future

Musk predicted that in the future, the number of humanoid robots will far exceed that of other forms of robots, possibly outnumbering humans by 5 to 10 times. He admitted that he had concerns about developing AI and humanoid robots, not wanting to make "The Terminator" a reality, but later realized, "Whether I do it or not, it will happen." Thus, he chose to transition from a bystander to a participant and is now fully committed to advancing humanoid robots and digital superintelligence development.

Regarding digital superintelligence, Musk made a bold prediction: "I think we are very close to digital superintelligence. If it doesn't happen this year, it will certainly emerge next year." He defined it as "smarter than any human in any respect." When asked how humanity can guide it towards "superabundance," Musk believes a positive outcome is most likely and reiterated that the pursuit of truth and having human empathy are key.

As for Neuralink, Musk stated it is not a necessary condition for achieving digital superintelligence but can solve the input-output bandwidth limitations between humans and machines. Currently, five human subjects have implants that read neural signals to help paralyzed patients communicate at nearly normal bandwidths. In the next 6-12 months, Neuralink plans to conduct its first visual implant experiments, allowing blind individuals to "see" by directly writing into the visual cortex, potentially even achieving enhanced functions like hyperspectral vision.

Finally, Musk returned to one of his grandest visions: making humanity a multi-planet species. He believes that within about 30 years, humanity could transport enough material to Mars to make it self-sufficient, such that if Earth stops provisioning, Martian civilization can continue to thrive. "This greatly extends the potential lifespan of civilization, consciousness, or intelligence (including biological and digital intelligence)." He mentioned the Fermi Paradox, believing that intelligent life may be extremely rare and that human civilization is akin to a small candle flame in the vast darkness, carrying the responsibility not to let that flame extinguish. Becoming a multi-planet species is the first step towards other star systems and will also force continuous advancements in space travel technology.

Advice for Builders: Pursue Utility and Stay Humble

At the end of the dialogue, Musk offered advice to the young tech elites present. He repeatedly emphasized the importance of "becoming as useful as possible." "This may sound a bit cliché, but becoming useful, especially to many people, is very difficult." He introduced the concept of "area under the utility curve": the degree to which you are useful to your peers multiplied by the number of beneficiaries. Focusing on creating such "true work" will significantly increase the chances of success.

He warned to be cautious of the "ratio of self to ability." When self-inflation exceeds ability, it breaks the feedback loop with reality (in AI terms, this means breaking the reinforcement learning cycle). Therefore, it is essential to internalize responsibility, minimize ego, and strive to complete tasks regardless of their perceived importance. This is also why he prefers the term "engineering" over "research" and refers to XAI as a company rather than a "laboratory"—pursuing simplicity, directness, and low ego.

"If you are doing something useful, that's great," Musk summarized, "Just try to be as useful to your peers as possible, and you will be doing good." Looking to the future, he expressed optimism that AI might help us understand the nature of the universe, answer ultimate questions regarding aliens, the origins and ends of the universe, and whether we are in a simulated world. "I think we will find the answers."

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。