Author: Techub News Compilation

In a recent public dialogue held in New York, Anthropic co-founder and CEO Dario Amodei engaged in an in-depth conversation with New York Times columnist Andrew Ross Sorkin. As an early core member of OpenAI, a key leader in the development of ChatGPT-2/3, and now the helm of the rapidly valued Anthropic, Amodei offers insights that blend the foresight of a technological pioneer with the realistic considerations of an industry leader. During the dialogue, he not only addressed skepticism about the overheated investment in AI and the "bubble theory," but also candidly shared Anthropic's survival strategies in the "cone of uncertainty," and expressed his views on sharp issues such as national security, regulation, and employment impacts related to AI.

Technological Optimism and Financial Caution: Survival Rules in the AI Frenzy

When asked whether the current massive investments in the AI field, often amounting to hundreds of billions of dollars, represent a bubble, Dario Amodei provided a dialectical answer. He broke down the question into two layers: technological prospects and economic realities.

On the technological front, Amodei claims to be "one of the most optimistic people." His confidence stems from twelve years of observation and practice regarding "Scaling Laws." This principle, initially articulated by him and his colleagues, indicates that as long as more computational power and data are continuously invested, the performance of AI models will steadily improve across almost all tasks—from programming, science, and biomedicine to law, finance, materials, and manufacturing. "This almost covers all areas of value creation in the modern economy," he summarized. Using Anthropic's own growth as an example: the company's revenue escalated from zero to $100 million in 2023, reaching $1 billion in 2024, and is expected to lie between $8 billion and $10 billion in 2025. This tenfold annual growth rate, in his view, is direct evidence that technological value is being realized.

However, switching to an economic and financial perspective, Amodei expressed clear concerns. The core issue lies in a fundamental "uncertainty": the trajectory of future revenue growth is extremely hard to predict, yet building the computational infrastructure (data centers) required to support these revenues takes one to two years of lead time. Companies must make decisions now regarding how much computational power to procure by early 2024 to service expected business sizes in early 2027.

"This creates a 'cone of uncertainty,'" Amodei explained, "I cannot predict whether revenue a year from now will be $20 billion, $50 billion, or another number." He attempted to make conservative plans, but risks abound: if computational power is under-procured, the company will fail to meet customer demands and relinquish the market to competitors; if over-procured, it may face financial crisis or even bankruptcy due to revenue not covering high fixed costs.

Amodei pointed out that the ability of businesses to withstand this risk depends on their profit margins. High profit margins provide a buffer. He implied that Anthropic, due to its focus on the enterprise market (rather than consumer-level), has a superior business model and higher profit margins, and thus can manage this risk in a "relatively responsible" manner. However, he pivoted to state clearly that not all players are so cautious: "What worries me is that some participants in the ecosystem, if they simply make a timing error or deviate just a little, could experience terrible consequences." When pressed on which companies are engaging in "YOLO" (reckless risk-taking), he chose to evade the question but hinted that the answer was obvious.

In response to OpenAI CEO Sam Altman's goal of shifting from significant losses to profitability within two years, Amodei declined to comment on the financial situation of other companies while reiterating Anthropic's cautious calculation principles based on the "cone of uncertainty": procuring enough computational power to ensure that even in the worst-case scenarios (such as the tenth percentile), it can remain competitive.

Regarding the concerns about "circular trading" (i.e., chip suppliers like NVIDIA investing in AI companies, which then use the funds to purchase their chips), Amodei defended its rationale. He illustrated with a simplified model: assuming building a 1-gigawatt data center requires about $50 billion in capital expenditures, amortized over a 5-year lifespan, the annual cost would be about $10 billion. A company rapidly growing with nearly $10 billion in annual revenue might not have $50 billion cash on hand, but collaborating with capital-rich giants willing to sell chips to obtain partial investment for initial year expenses, then paying back with revenues subsequently, is logically sound. "In principle, there is nothing wrong with that." He emphasized that the risk only arises when the scale of such transactions becomes excessively large and revenue projections become overly aggressive, leading to overexpansion.

On the debate about the chip depreciation cycle, Amodei again showed his conservative stance. He believes the issue is not the physical lifespan of the chips, but the rapid iteration of new chips at faster speeds and lower costs, leading to a swift depreciation of the value of old chips. "This might happen just a year after you purchase the chips." Therefore, Anthropic has assumed a very aggressive chip efficiency improvement curve in its planning and made conservative financial forecasts based on that. "We believe we can cope in almost all scenarios." However, he reiterated that he cannot speak for other companies and hinted that there might be some overly optimistic assumptions.

Enterprise Path vs. Consumer War: Anthropic's Differentiated Moat

Recently, the industry shock caused by Google's new model release, along with the "red alert" issued internally by OpenAI, highlights the intense competition in the consumer AI market. Dario Amodei expressed that he is "very grateful" for the differentiated path that Anthropic has chosen.

Amodei pointed out that the core battlefield for OpenAI and Google lies in the consumer market. Google needs to defend its monopoly in search, while OpenAI's focus is also on consumer business. Serving enterprise customers is secondary to them. In contrast, Anthropic has consistently focused on the enterprise market. Over time, its models have increasingly been optimized for enterprise needs—coding capabilities are improving rapidly and are expanding into various areas such as finance, biomedicine, retail, energy, and manufacturing.

"In these model wars, although our models are also very impressive... we are somewhat developing in a different direction or on a different dimension." Amodei believes this allows Anthropic to be less affected by the fierce struggles of the consumer market, occupying a "privileged position" to focus on continued growth and model development without sounding a "red alert."

So, what is the moat for AI companies? When model capabilities eventually converge, will users frequently switch solely based on who has the latest and strongest model? Amodei provided a negative answer from the enterprise market perspective.

First, he emphasized the fundamental differences between building models for enterprises and for consumers. Enterprise-level models are more focused on coding, high-intelligence activities, and scientific capabilities, rather than user engagement. This differing optimization goal results in significant differences in the "personality" and focus of the models' capabilities.

Second, even if general artificial intelligence (AGI) arrives, Amodei does not think all models will converge at a single point. "Specialization exists alongside general intelligence." Just as humans possess general intelligence but excel in different fields, AI models can and will specialize.

Finally, he pointed out the inherent stickiness of the enterprise market: companies establish relationships with vendors and become accustomed to using specific models. Even seemingly standardized API businesses make it difficult for enterprise customers to switch between different models because downstream clients have already adapted to the interaction styles and "personalities" of the existing models.

When asked if merely stacking existing Transformer architectures and computational power can lead to AGI, Amodei gave an affirmative answer. "I believe scaling will get us there." He reiterated the scaling laws, predicting that only slight improvements may occasionally be needed in the future. He compared the advances in AI capability to the exponential curve of Moore's Law: models will become increasingly powerful in all aspects, with new models released every few months achieving breakthroughs in coding, science, and mathematics. He revealed that some researchers at Anthropic have indicated they no longer personally write code but instead let Claude generate drafts, with themselves only handling edits. "This process will continue... what we will see in the future is merely a continuation of the past, but to a deeper extent."

National Security, Regulation, and Employment: The Public Responsibility of AI Leaders

Dario Amodei is known for his frankness on AI policy issues, especially regarding the export of advanced chips to China. Despite Anthropic forming a partnership with NVIDIA, and NVIDIA CEO Jensen Huang expressing dissatisfaction with his remarks, Amodei clearly stated that his views have not changed.

He characterizes this as a national security issue rather than an economic one. Amodei painted a picture: as model capabilities increase exponentially, a "genius nation in data centers" will eventually form. It is crucial where this "nation" is located geographically. "If it lands in an authoritarian state, I believe they will surpass us in every aspect: intelligence, national defense, economic value, research and development... I worry they will be able to oppress their own people, establishing a perfect surveillance state." Thus, he believes that democratic countries must gain this advantage first, which is an "absolute necessity." Selling advanced chips to China "only increases the likelihood that they will reach their destination first; it's common sense."

When the conversation shifted to the risks of surveillance within democratic countries, Amodei attempted to elevate the discussion to policy principles rather than targeting specific individuals or governments. The core principle he proposed was: "We should actively use these models in all possible ways, except those that would make us more like our authoritarian adversaries. We need to defeat them but should not take actions that would lead us to become like them."

In response to White House AI affairs chief David Sacks accusing Anthropic of "fear marketing," engaging in "complex regulatory capture," and harming the startup ecosystem, Amodei refuted this. He pointed out that he had written papers on AI risks long before founding the company in 2016, and the major AI bills supported by Anthropic (such as SB 53) include exemptions for small companies with annual revenues below $500 million. "These accusations are completely unfounded."

He further explained the differing positions on regulatory issues. He believes that some people compare the AI revolution to the internet or telecom revolution, believing the market will resolve everything on its own; this viewpoint may have been reasonable in the past, but "those closest to AI do not think so." Real AI researchers are excited about its potential but also concerned about national security risks, model alignment issues, and economic impacts. He described proposals advocating a ten-year pause on all state-level regulations (in the absence of a federal framework) as dangerously akin to "driving while removing the steering wheel because you won’t need to turn for the next ten years."

Regarding the impact of AI on employment, Amodei previously predicted that up to half of entry-level jobs could be affected. In this discussion, he systematically elaborated three levels of response to this challenge.

First Level (Private Sector Driven): Enterprise clients face a trade-off when using AI. They can automate existing processes (e.g., claims processing, "know your customer" procedures) with AI, significantly improving efficiency and reducing costs, but the required manpower will sharply decline. At the same time, they can leverage AI to create substantial new value. Even with AI completing 90% of the work, the productivity of human employees could be leveraged tenfold; sometimes, even more employees may be needed to handle workloads a hundred times greater than before. Society should encourage enterprises to engage more in the second form of value creation.

Second Level (Government Intervention): Amodei believes that retraining programs are not a panacea, but the government ultimately needs to intervene financially. The rapid growth driven by AI will create a massive economic "cake," and the government needs to ensure that wealth does not become overly concentrated and support workers in transition through measures like tax policies.

Third Level (Restructuring Social Structures): In the long run, a society with powerful AI must be different. He referenced Keynes' vision in "the economic possibilities of our grandchildren" where people may only need to work 15 to 20 hours a week. Work may shift in core meaning for many from economic survival to self-actualization. "Society is flexible... We need to jointly find out how to operate in the age of general artificial intelligence (AGI)."

Amodei concluded that his warnings are not meant to disseminate pessimism but because "early warnings are the first step in addressing problems." Only by recognizing the "landmines" ahead can society avoid them rather than blindly stepping into them. This leader in the AI field, while exhibiting tremendous optimism about technological prospects, also shoulders the responsibility to call on society to prepare for the future.

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。