Written by: Techub News Compilation

Introduction

Recently, OpenAI CEO Sam Altman engaged in a nearly hour-long in-depth dialogue with renowned economist Tyler Cowen on the Roots of Progress Institute's talk show. This conversation took place against the backdrop of OpenAI's frequent product and business collaborations and the global anticipation for the next-generation model. Altman rarefully elaborated on his outlook for GPT-6, how AI is reshaping company operations and scientific discoveries, the ultimate challenges of computing power and energy, as well as the social contract and governance in the AI era. As one of the most influential leaders in the current AI wave, Altman’s remarks provide a key perspective for understanding OpenAI’s strategic direction and the future evolution of AI technology.

Summary

• High expectations for GPT-6: Altman believes that GPT-5 has shown a "glimmer of insight" in AI-assisted scientific discovery, and GPT-6 is expected to achieve a leap akin to that from GPT-3 to GPT-4, truly becoming a powerful tool for scientific research.
• AI-driven organizations are imminent: He predicts that within the next two to three years, billion-dollar-level companies operated jointly by two or three people and AI will emerge, with AI deeply involved in and even leading decision-making in important departments.
• The ultimate constraint on computing power is energy: Altman points out that the fundamental bottleneck for acquiring more computing power lies in obtaining more "electrons" (i.e., energy), with short-term relief relying on natural gas, while the long-term winners will be nuclear fusion and solar energy.
• Exploring new AI-native interactions and business models: OpenAI is collaborating with Johnny Ive to create a completely new AI-native computer interaction paradigm; in terms of business models, they are more focused on how to discover new sciences and realize commercialization through the smartest models rather than solely relying on transaction commissions like hotel bookings.
• Social adaptation is the biggest challenge to AI proliferation: Although AI’s decision-making ability may soon surpass that of humans, societal trust and acceptance of AI will take longer, becoming a key obstacle to the deep integration of AI into economic life.

GPT-6: The "Glimmer" of Scientific Discovery and the Next Leap in AI Capability

When asked why he has special expectations for GPT-6, Sam Altman provided a vivid metaphor. He likened GPT-3 to the first instance of a "mental Turing test" being passed, while GPT-5 marks the starting point where AI begins to show "glimmers of insight" in the scientific field—such as providing new ideas, assisting in research papers, or solving specific problems. However, Altman believes GPT-6 has the potential to achieve a qualitative change similar to that from GPT-3 to GPT-4, consolidating these glimmers into a true capability that drives scientific discovery.

Regarding how research laboratories should prepare for the arrival of GPT-6, Altman admitted it is a challenge. The most straightforward way might be to pose current tricky research questions to it. However, he thought more deeply about structural changes on an organizational level: How to rebuild the entire organization around GPT-6 or future AI cores, rather than simply allowing top researchers to use it as an additional tool for existing knowledge. Using OpenAI itself as an example, he suggested a highly thought-provoking thought experiment: how to accelerate the realization of an AI CEO operating OpenAI, and doing it better than himself? He believes that pondering "what hinders an AI CEO" is a super useful method for designing future organizational structures and identifying key obstacles. He recommended that managers of research laboratories should engage in similar thinking, even though the conclusions may differ.

So, how close is the timeline for AI's deep involvement in company operations? Altman predicts that a scenario where an important department (not a small department) is mainly operated by AI will be realized within "single-digit years," meaning it is not far off. He even provided a more specific forecast: within the next two to three years, the world will see billion-dollar companies operated by merely two or three people and AI together. He previously predicted it would happen in a year, but now slightly postponed; the reason is not a more pessimistic view on AI capability, but a more pessimistic view on human adaptation speed. Altman agreed that AI's actual decision-making capabilities may soon be able to handle most tasks, but it will take longer for people in society and organizations to feel truly comfortable with it.

Organizations, Talent, and the Survival Rules in the AI Era

In light of OpenAI's recent intensive product releases and business collaborations, Tyler Cowen was curious about how Altman enhances individual and organizational productivity. Altman's answer focused on empowerment and focus: hiring and promoting top talent, delegating a significant amount of work to them, and ensuring that as company goals become increasingly clear, he can understand and concentrate on the most core matters. He mentioned that as the world becomes more willing to collaborate with OpenAI, transaction negotiations also become more expedited.

As OpenAI's business extends into the hardware (consumer devices, chips) field, Altman also shared the differences in managing hardware talent versus AI talent. The hardware field has longer cycles, more capital intensity, and higher trial-and-error costs, so he would trust and empower only after getting a thorough understanding of the related personnel. However, he revealed that OpenAI's chip team feels more like a research team than a traditional chip company, which may lead to remarkable success.

When recruiting, Altman specifically pays attention to candidates' adaptability to AI. An important indicator is how they currently use AI. Those who only use AI to improve Google Searches will raise a "yellow flag," while those who thoughtfully consider how their work will be reshaped by AI in three years will raise a "green flag." He stressed that the future belongs to those who actively embrace and think about co-evolution with AI.

Regarding team collaboration tools, Altman bluntly stated that email is "very bad," while Slack, though good, is also not perfect as it creates a lot of "fake work." He envisions a future driven by AI, with an entirely new productivity suite that replaces existing documents, slides, emails, and Slack. In this new paradigm, people will trust that their AI agents work alongside other AI agents for most matters, only reporting to humans when necessary. Although OpenAI is still far from achieving such a system internally, he believes this is not due to insufficient model capabilities, but rather because the effort has not yet been devoted to building it.

The Ultimate Challenges of Computing Power, Energy, and Hardware

When asked about the ultimate constraint on acquiring more computing power, Altman's answer was concise and profound: "Electrons", meaning energy. He believes that in the short term, natural gas can relieve this constraint; in the long term, the world will be dominated by nuclear fusion and solar energy, both of which he is very optimistic about. Regarding the potential public acceptance issues of nuclear energy due to its name, he believes that as long as the price is low enough (for example, a tenth of the price of natural gas), demand will outweigh all irrational concerns.

Altman is also concerned about the risk of a significant shift in computational paradigms, such as a full transition to optical computing, which could render current heavy asset investments a dead end. However, he remains willing to bet because even if extremely intelligent models can be built using minimal quantum computing, the demand for cheaper and richer computing power remains endless.

On the hardware front, Altman acknowledged that self-developed chips "have no easy parts." However, he pointed out a less discussed but crucial aspect is the recursive improvement potential of AI on hardware research and development: robots making robots, data centers building data centers, chips designing the next generation of chips. This could make many hardware challenges easier to solve.

Business Models, Social Contracts, and Trust Reconstruction

Tyler Cowen shared his experience of planning a trip using ChatGPT and posed a sharp question: How can OpenAI profit from such recommendation services without harming user trust? Altman first pointed out an interesting phenomenon: although AI is known for "hallucinations," ChatGPT is often reported by users as one of the most trusted products of major tech companies. He believes this stems from its commercial model being potentially aligned with user interests: ChatGPT's goal is to provide the best answers (users pay for or expect this), rather than being influenced by advertising revenue like search engines that might be affected if "too good" at their performance.

Altman envisions a scenario where if ChatGPT recommends the best hotels and charges a commission for bookings that is the same as other hotels, while recommendations remain unaffected by payment, this could be acceptable. However, he also acknowledged that in a world with ubiquitous smart agents, the commission rates for services like bookings might significantly decrease due to competition. He firmly believes that the monetization of the smartest models in the world is not through hotel bookings, but rather through discovering new sciences and commercializing them in ways that only the smartest models can achieve.

Nonetheless, Altman believes that creating a super-intelligent system that is easy to use, deeply integrated, and understands users throughout their lives, and investing in infrastructure, chips, and energy to make it abundant and cheap, is the most likely path to making the world a better place. He is willing for OpenAI to engage in endeavors that are not solely about maximizing economic benefits but are important for the world.

Regarding the societal impacts brought by AI, Altman predicts a high likelihood that "the social contract will have to undergo significant changes," although the specific forms are unknown. He foresees that governments may become more deeply involved. In terms of regulation, he advocates differentiating treatment based on capability thresholds: AI entities capable of large-scale self-replication and causing severe harm require regulation; but for most systems, excessive intervention is unnecessary.

Culture, Bias, and Future Interaction Forms

The dialogue also touched on broader cultural aspects and future outlooks. Altman believes that books, as a form of thought carrier, have strong vitality, but in the future, new ways that are more suitable for interacting with clusters of thoughts may emerge. He anticipates that his work methods (emails, meetings, documents, Slack) will undergo tremendous changes, but core human experiences such as spending time with family, enjoying nature, and socializing with friends are unlikely to change for a long time.

Regarding the combination of AI and social media, Altman shares Cowen's sentiment: he is very interested in the results of queries he makes with ChatGPT, but finds the results shared by others to be boring. He believes that ChatGPT is essentially a single-player experience, but in the future, when everyone has powerful personal AI agents, it may give rise to entirely new social dynamics.

In terms of interactive forms, Altman revealed that OpenAI is collaborating with renowned designer Johnny Ive to create a completely new computer interaction paradigm designed for AI, which could completely disrupt the computational paradigms of the past fifty years. He acknowledged that people still love typing in text boxes, which may reflect some limitations of humanity, but he remains hopeful about creating entirely new interfaces.

Finally, when asked about what question he would most like to ask an expert, Altman posed a philosophically significant thought: When superintelligence is constructed, passes security tests, is prepared to start, and is about to complete incredible self-improvement and exploration of the stars, what kind of prompt should you enter before pressing the "confirm" button? He candidly admitted he has no answer yet, but this is one of the problems he most wants to solve.

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。