OpenAI Long Report: AI can handle research that would take humans a century to complete, with costs decreasing by 40 times year by year.

OpenAI recently released its latest lengthy report titled “AI Progress and Recommendations,” which comprehensively reviews recent breakthroughs in artificial intelligence, future development pace, and directions for safety governance. The report points out that AI systems are no longer limited to chat Bots or search engine assistance, but are gradually able to surpass the knowledge-based competitions and research tasks of the smartest human minds. OpenAI predicts that by 2026, AI will be able to make micro scientific discoveries, and by 2028, it will have the capability to achieve significant scientific breakthroughs. However, at the same time, the safety of AI must also be controlled concurrently.

AI Capability Accelerates Evolution: From Second-Level Tasks to Century-Level Tasks

OpenAI points out that the advancement of AI far exceeds societal awareness. In just a few years, AI has progressed from being able to perform tasks that humans can do in seconds to completing work that takes hours or even days. In the future, it may even handle research-level tasks that would take humans hundreds of years to accomplish.

At the same time, the cost of intelligence is rapidly decreasing, with an annual decline of about 40 times. The report states that this means the era of obtaining high-level intelligence at extremely low costs is approaching quickly.

Is AI the true Web3? It will achieve the universal benefit of human intelligence.

OpenAI believes that while the evolution of AI will dramatically change production structures and economic models, there will not be severe upheavals in daily life in the short term. Human society has a strong inertia and often only gradually adjusts after new technologies are integrated. The report states: “We expect the future world will present entirely new and more fulfilling ways of living, allowing more people to live prosperous and meaningful lives.” However, OpenAI also warns that this transition will be accompanied by a difficult economic adjustment period and a reshaping of the labor structure, and may even require a redefinition of the social contract.

The report emphasizes that the positive application potential of AI will be particularly significant in data-intensive fields. AI will:

Help humanity gain a deeper understanding of health and disease mechanisms.

Accelerating materials science, drug development, and climate modeling

Provide personalized educational opportunities for students worldwide.

OpenAI pointed out that these specific achievements will help society form a consensus that AI makes life better rather than just more efficient.

Coincidentally, editor Colin Wu also shared his personal insights: “Blockchain is really not Web3; instead, it is AI that truly brings that sense of shock reminiscent of when traditional internet and mobile internet burst onto the scene. It is the real Web3, revolutionizing everything. In the future, even in the poorest areas, people will be able to enjoy consultations with the world's best doctors and educational guidance for free through AI. Blockchain has not truly achieved financial inclusivity; those who hold Bitcoin and possess information asymmetry are becoming richer; but AI has the potential to achieve inclusivity of human intelligence.”

AI Safety and Governance: Preventing Superintelligent Systems from Going Out of Control

OpenAI explicitly stated in its report that while the potential of AI is enormous, its risks could also be catastrophic. The company argues that safety research should be regarded as the foundational engineering to ensure positive impacts of AI, and calls for the establishment of shared standards and research sharing mechanisms globally.

Recommended measures include:

Frontier laboratories share security standards and research findings: The establishment process of building fire prevention regulations to reduce racing risks (race dynamics).

Establish a public oversight and accountability system: Ensure that AI technology aligns with the public interest during its promotion.

Building an AI Resilient Ecosystem: Just as the internet era developed cybersecurity nets, establish encryption, monitoring, and emergency response systems.

Continuous Monitoring and Reporting of AI Impact: Observing real societal effects through data to facilitate adjustments in policy and business decision-making.

Ensure personal empowerment and accessibility: OpenAI suggests that AI should become a fundamental public resource, like electricity, clean water, and food, allowing everyone to use it under their own control.

The report presents two AI development scenarios:

Normal development: AI will gradually permeate society like printing or the Internet, and policy tools can be adjusted according to existing models.

Superintelligent Development: If AI reaches self-improvement capabilities at an unprecedented speed, traditional regulation will be insufficient to respond. At this point, reliance on multinational coordination, strengthening security agencies, and preventing risks of bioterrorism and military misuse will be necessary.

OpenAI pointed out: “The ultimate high-level principle must be: the progress of AI must be accountable to public institutions,” but the path to achieving this goal may be completely different from all previous technological revolutions. This argument demonstrates a pragmatic optimism towards AI: on one hand, it anticipates that AI will fundamentally change the way knowledge is produced within a decade; on the other hand, it emphasizes that safety, consensus, and accountability should be at the core, establishing a systemic foundation for coexistence between humans and AI.

The report concluded: “AI will become the infrastructure for a new generation of humanity, and the key is to ensure it makes the world a better place, not a more dangerous one.”

This article OpenAI Long Report: AI can handle research that would take humans a hundred years to complete, with costs decreasing by 40 times each year, first appeared on Chain News ABMedia.