Written by: Techub News

During the World Economic Forum held in Davos, Switzerland, The Wall Street Journal invited Dario Amodei, co-founder and CEO of Anthropic, for an in-depth conversation. As one of today’s most noteworthy leaders in cutting-edge AI companies, Amodei's thoughts pertain not only to the commercial success of his company's Claude model, but also touch on the fundamental reshaping of social structures, economic forms, and even the global landscape by AI technology. This interview reveals how this CEO, with both a scientific and entrepreneurial perspective, views the real trajectories and potential risks amidst the AI frenzy.

Smooth Technological Curve and Swinging Public Perception

At the beginning of the interview, the host pointed out the shift in public discussions about AI over the past year: from excitement about “what AI can do” to concerns about “what AI is doing to the world.” In response, Dario Amodei’s answer was straightforward: he believes that businesses, policymakers, and governments are far from being adequately prepared for the impact of AI.

With more than a decade of observational experience in the AI field, Amodei depicted a contrasting picture: technological capabilities follow a “smooth exponential curve”, similar to Moore's Law in computing, where models make significant cognitive advances every few months, and this process remains stable. However, public opinion and reactions are on this smooth curve, oscillating violently. Every three to six months, media attitudes experience a “polar reversal”—from extreme excitement (“It will change everything”) to utter skepticism (“It’s all a bubble, about to collapse”). This fluctuation reflects public perception rather than the facts of the technology itself.

Similarly, the polarity of public opinion regarding whether technology is “good” or “bad” also swings. In 2023 and 2024, there are widespread concerns about AI risks and misuse; by 2025, the political winds shift to emphasize AI opportunities; and now, it seems the winds are swinging back. Amodei believes that Anthropic tries to maintain a consistent stance: acknowledging that AI's impact possesses a “peculiar balance”—this technology's capabilities are immensely powerful, but its positive and negative effects exist simultaneously.

He penned an article titled “The Loving Machine” more than a year ago, radically portraying the positive side of AI: helping to cure cancer, eradicating tropical diseases, and bringing prosperity to regions that have yet to enjoy economic development. His view on this remains unchanged. However, he also soberly recognizes the other side of the coin: bad things can equally occur.

High Growth and High Unemployment Coexist: An Unprecedented Economic Phenomenon

Amodei used the economic impact as an example to illustrate the core paradox brought by AI. This technology's hallmark will lead us into a world of rapid GDP growth, while simultaneously unemployment rates and inequality may also be very high. He noted that this combination has almost never occurred historically. In the past, high GDP growth meant a plethora of transactions and job opportunities. But AI is a truly unprecedented disruptive technology, and it is logically possible to have a 5% or 10% GDP growth accompanied by a 10% unemployment rate. He feels both excited and concerned about this.

He provided micro-level evidence by citing AI programming. After the release of Anthropic's latest model Claude Opus 4.5, some engineering executives within the company told him: “I no longer write code at all; I just let Opus do the work and then edit it.” Their recently released collaborative tool for non-coders, Claude Co-work, was almost entirely built by Claude Opus in one and a half weeks.

Amodei acknowledged that software engineers still have work to do at the moment (even if they are only responsible for 10% of the workload), but this won't last forever. Models will take on more and more work. This is a microcosm: productivity will significantly increase, software may become extremely cheap, or even almost free. The premise that building software once required millions of users to amortize costs may no longer hold true. However, simultaneously, entire careers and professions built over decades may cease to exist.

He believes society can handle this transition, but the key is that almost no one is aware of the upcoming changes and their scale.

Measuring, Adapting, and Distributing: Three Key Steps to Respond to Major Changes

When asked how society can adapt to this high-growth, high-unemployment world, Amodei proposed three key steps.

The first step is measurement. Anthropic has launched the “Anthropic Economic Index” and updated it four or five times. It is a real-time index that tracks how Claude models are used for various tasks, the spread across industries, among U.S. states and countries worldwide, and the proportion of automation versus augmentation tasks, all in a privacy-protecting way. Amodei emphasized that any policy will be blind and erroneous before we can measure the form of this economic transformation. Many policy failures arise from being based on fundamentally erroneous premises.

The second step is to help people adapt. This includes adapting to using technology in existing jobs or transitioning from one occupation to another. He predicts that jobs in the physical world may increase while jobs in the knowledge economy may decrease (the trajectory of robotics may be slower). Additionally, which jobs truly require “human contact” will be determined by the market. At the corporate level, when software and entire knowledge work become cheap, what will be the company’s “moat”? This will trigger a massive scramble for resources and reorganization. Thus, educating people to adapt and informing them of expectations is the second step.

The third step is the role of government. Amodei believes that in the face of such a macro-scale structural displacement, governments must play a role. The economy’s total size (the pie) will grow significantly, and funds will be plentiful enough that balancing the budget may not even require specific actions. But the issue lies in how to allocate it to the right people. He believes that we should reduce concerns about suppressing growth and focus more on ensuring that everyone can share in the growth benefits—this is contrary to the current mainstream sentiment, but the technological reality will soon force our views to change.

Corporate Strategies and Safety Principles: Anthropic's Differentiated Path

The host mentioned that Anthropic was founded out of concern for OpenAI's emphasis on safety, and inquired whether competitive pressures (such as maintaining an edge over China) had compromised its safety principles. Amodei elaborated on the differentiated path chosen by Anthropic.

One wise early choice was to focus on the enterprise market instead of the consumer market. He believes that counteracting one’s own commercial incentives is very difficult, but selecting a business model that does not require too much resistance is relatively easier. He has many concerns about consumer AI because it often seeks to maximize user engagement, leading to “slop” and advertising inundation. Anthropic does not need to operate in this way; they sell products directly to enterprises, which already have clear value, without needing to seek monetization avenues for millions of free users, nor maximize engagement in a “death race” with other giants.

Even so, Anthropic has made sacrifices. They have conducted many tests on models that other companies have not, being most proactive in testing for deception, extortion, and flattery. When they discovered that all models had such issues, they ensured that communication with the public was established. They have also pioneered the science of “mechanistic interpretability”, dedicated to examining the inner workings of models. Amodei candidly admits they are not perfect, but overall, they have done well.

Regarding China, he emphasized that this is not about competition; it is about a public interest mission. He worries that if authoritarian regimes dominate this technology, it will produce bad outcomes for everyone in the room. AI may be particularly suited for authoritarian rule, deepening the repression we have already witnessed. It can create personalized propaganda, infiltrate any global computer system, surveil entire populations, detect dissent everywhere, and suppress it, even creating large drone armies targeting individuals. This is very frightening and must be stopped. He believes the government’s attention on this issue is currently insufficient, and targeted policies should be implemented, such as not selling these chips.

The "Highlights" of Claude and Business Prospects

When the topic shifted to the current success of the Claude model, Amodei shared its commercial growth in a “smooth exponential curve”: revenue grew from zero to about $100 million in 2023, from about $100 million to about $1 billion in 2024, and from about $1 billion to about $10 billion in 2025 (all rough numbers). Despite extreme statements on social media every few months about Anthropic “changing the world” or “totally failing,” they focus solely on this continuously and rapidly progressing curve, which gives them confidence.

Even with a smooth curve, there are breakout moments. One current breakthrough point revolves around the adoption of Claude Code among developers. The latest model Opus 4.5 has reached an inflection point, where incremental improvements accumulated to a moment that began to attract noticeable attention. A second acceleration factor is that many non-technical personnel (both internally and externally) have come to realize that Claude Code can accomplish incredible agentic tasks—not just writing code, but also organizing to-do lists, planning projects, tidying folders, processing information, and summarizing. The demand from non-technical personnel is so strong that they are willing to wrestle with the command line. Based on this unmet demand, they developed an improved UI version for non-coding tasks using Claude Code in about two weeks, which, upon release, exceeded metrics far beyond any previous release.

Regarding an IPO, Amodei stated that there are currently no definite plans; the highest priority remains to sustain revenue curve growth, better sell models, warn about societal impacts, and bring positive social changes. But he acknowledges that this is a capital-intensive industry, and the funding available from private markets is ultimately limited.

When discussing competition with Google Gemini, Amodei believes that differentiation once again provides an advantage. Google and OpenAI are engaged in existential competition in the consumer space (completely encompassing OpenAI’s business and disrupting Google’s search business), so the enterprise market is not their primary focus. Anthropic's corporate strategy allows it to respond more calmly.

Regarding Anthropic not generating videos and images, Amodei thinks there is virtually no demand for such services in the enterprise sector (perhaps slide presentations are a fringe use case). If needed, they can contract a model from someone else. He observed that a large number of short videos are either false, addictive, or slop; this segment of the market is not one he is eager to enter.

Scientists and Entrepreneurs: Different Leadership Philosophies and Responsibilities

Amodei compared the differences in approach between scientists leading large AI companies and tech entrepreneurs. This technology is the intersection of decades of academic research and the large-scale infrastructure and funding required over the past decade and a half. Some companies are led by individuals with scientific backgrounds (like himself and Google’s Demis Hassabis), while others are led by the entrepreneur generation of the social media age.

He believes scientists have a long tradition of considering the impacts of the technologies they create, feeling responsible without ducking responsibility. Their initial motivation is to create things for the world, thus they will also be concerned about potential mistakes. In contrast, entrepreneurs, especially those from the social media generation, have very different motivations, affected by their selection effects and the way they interact with (or manipulate) consumers, leading to varying attitudes.

Key to AI Safety: Mechanistic Interpretability

When responding to an audience question about “the most important technological breakthroughs needed to ensure the safety and controllability of cutting-edge AI,” Amodei clearly pointed out the need for more progress on mechanistic interpretability, which examines the inner workings of models.

One issue with training these models is that we cannot ascertain whether they will do what we believe they will do. Just like conversing with a person, what models say in one context may not be a faithful representation of their true thoughts. They may lie. Any phenomenological testing or training cannot ensure this. But similar to using MRI or X-ray to understand the human brain (which cannot be discerned through conversation alone), examining the inner workings of AI models is, according to Amodei, ultimately key to making the models safe and controllable, as it is the only ground truth we possess.

Education, Equity, and Responsibilities in Future Societies

Regarding AI’s impact on K-12 educational achievement gaps, Amodei acknowledged the short-term issue of cheating using AI, but the deeper question is: what skills should we teach in an AI-driven world? What should education look like? He admitted the uncomfortable truth is that he isn’t even sure about the direction of future careers.

He suggested returning to earlier concepts about education, moving away from the very economic and mercenary view of education that has dominated in recent years, towards a view that education aims to shape you as a person, build character, enrich you, and make you a better person. He believes this is actually a safer foundation for future education.

Finally, when faced with the question of “what responsibilities do AI laboratories have towards underdeveloped economies, countries, and people,” Amodei expressed concerns about this issue on multiple scales. This is not just a matter between nations; he worries that developing countries might be bypassed by the technological revolution and is also concerned about internal divisions within nations. Customer data show that startups are adopting AI very quickly, while traditional companies, due to their size and fixed operations, are moving slowly. There are also differences in the speed at which technology spreads across U.S. states.

He painted a nightmare scenario: a “zeroth world country” could emerge, comprising around 10 million people (for example, the 7 million in Silicon Valley plus 3 million scattered elsewhere), forming its own economic system and becoming decoupled or disconnected from the outside world. High GDP growth there might manifest as 50% growth, with technology so powerful it could tear the world apart. He believes this would lead to a dystopian world that we must consider how to prevent.

Anthropic is reflecting or taking some measures in this regard. In terms of developing countries, they have begun substantial work around public health, collaborating with the Rwandan Ministry of Education and engaging in multiple partnerships with the Gates Foundation. He hopes rapid economic growth (which theoretically should translate to faster catch-up growth) could also happen in developing countries. Within nations, it is crucial to think about how to prevent certain areas from decoupling and how to ensure that Mississippi experiences the same economic growth as the contained area of Silicon Valley. They have already initiated work around economic mobility and economic opportunity.

Amodei reiterated that both areas require government involvement. The current ideology cannot survive the nature of this technology and cannot survive reality. What he discusses, although it may currently be politically labeled, will become bipartisan and universal, as everyone will recognize its necessity. He predicts that when looking back, if not next year, then the year after, everyone will come to this realization.

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。