OpenAI President Greg Brockman has declared that the world has entered the Age of Artificial General Intelligence (AGI), citing recent advancements, particularly with the model "Astra," as evidence. He posits that while models are becoming increasingly powerful, the challenge of equitable access due to compute limitations and the paramount importance of safety, security, and alignment are significant hurdles that are currently underestimated.

Brockman, who co-founded OpenAI after an early role at Stripe, has consistently focused on the most critical problems. For the past two years, this has involved a deep dive into data centers and machine learning infrastructure. Looking ahead, he anticipates that the conversation around AI's societal impact and responsible development will become increasingly central.

Predicting the Pace of Progress

Reflecting on past predictions, Brockman recalled that in 2016-2017, estimations for achieving AGI, based on compute progress and Moore's Law, suggested a 10-15 year timeline. He expressed that while current breakthroughs are remarkable, they also feel like a culmination of converging forces, making the present moment a logical point for such advancements. The increasing demand for AI capabilities has led to supply chain shortages, particularly in compute power, raising questions about whether the pace of development is sustainable.

The Compute Bottleneck and Pacing the Frontier

Brockman acknowledged the current difficulty in meeting the market's demand for compute. He believes that while the raw potential and capability of AI models will continue to advance, distributing this power equitably will be a significant challenge. This leads to the concept of "pacing the frontier," where progress with more capable models must be carefully balanced with continuous improvements in safety, security, and alignment. These factors, he argues, are becoming the bottlenecks to progress, rather than compute itself. The mission to empower everyone and ensure AI benefits all remains a core focus, deserving greater public attention.

Evolving Safety Paradigms

The approach to AI safety has evolved from initial surface-level filters to more architectural considerations. Early efforts focused on preventing undesirable outputs, but as AI capabilities expanded into areas like cybersecurity, the need for models to inherently understand and avoid dangerous actions became critical. Brockman asserted that rapid progress is being made in this domain, citing research and investment in safety for many years.

Key breakthroughs from OpenAI in 2017, such as early language models and Reinforcement Learning from Human Preferences (RLHF), laid the groundwork for current alignment techniques. Concepts like "debate" and "iterative amplification" were explored even before current systems existed, demonstrating a long-standing commitment to AGI safety. While initial public focus shifted to issues like political neutrality as chatbots gained popularity, the fundamental safety concerns are now re-emerging as systems become more powerful.

Collaboration and Coordination in AI Development

Brockman emphasized the importance of coordination among leading AI labs, including Google, Anthropic, SpaceX, and Meta. He believes that navigating the future of AI requires a collective effort to address complex questions. OpenAI has published its thoughts on pacing the frontier and unilateral actions for safety, acknowledging that operationalizing safety for AI development is a novel challenge. He noted that AI development is a consequence of broader technological progress, particularly in compute, and that leading companies can offer insights into future possibilities, but cannot shape the future alone. Sharing safety techniques and insights into alignment failures is crucial for this next phase.

The Hugging Face Incident: A Watershed Moment

The recent incident involving Hugging Face was described as a "watershed moment," highlighting two key aspects. Firstly, it underscored the need for robust internal monitoring, sandboxing, and control mechanisms during model evaluation, prompting significant changes in OpenAI's internal standards and controls. Secondly, it provided valuable insight into future capabilities when broadly diffused and potentially in the hands of threat actors.

Brockman warned of the risks associated with the diffusion of AI capabilities, particularly in cybersecurity. He noted that the AI involved in the Hugging Face incident demonstrated sophisticated hacking abilities, capable of breaching secure environments and production systems. This highlights the urgent need for defenders to leverage current "frontier capabilities" to secure themselves before these technologies become widely accessible. He described this as a "defender's window," where proactive measures can be taken to patch vulnerabilities and fortify systems.

The Dual-Use Nature of AI and the Defender's Advantage

The dual-use nature of AI, where offensive capabilities can be mirrored by defensive applications, is a critical aspect. While attackers can exploit vulnerabilities, defenders can use the same AI to identify and patch them. Brockman stressed that defenders, by controlling their systems, have an inherent advantage. The current window allows organizations to utilize frontier capabilities, often through trusted access programs, to enhance their security posture.

The Agentic Revolution and Reclaiming Human Time

The emergence of agentic AI, capable of coordinating and executing complex tasks, was highlighted. OpenAI's use of 10,000 agents to solve the Navier-Stokes problem exemplifies this potential, with implications for scientific discovery in fields like fluid dynamics and medicine.

Brockman also discussed the profound impact of AI on human interaction with computers. The development of AI that can directly use computer interfaces—screens, keyboards, and mice—eliminates the need for cumbersome APIs and specialized connectors. This shift promises to free humans from tedious tasks, such as menu navigation and data entry, allowing them to reclaim their time and focus on more creative and meaningful pursuits. He envisions a future where humans are no longer contorting themselves to machines, but rather are empowered by machines that serve them.

Employment in the Age of AI

Regarding the future of employment, Brockman expressed a fundamental belief that AI's impact will be surprising and not necessarily align with logical predictions. He emphasized the inherent value of human qualities like relationship building, accountability, and goal setting, suggesting these are deeply human attributes that should be preserved. While acknowledging the potential for a world of abundance, he stressed the importance of ensuring this abundance is broadly distributed. He anticipates a wave of entrepreneurship, with AI tools lowering barriers to entry and enabling individuals to launch their own ventures. This could lead to a renaissance where younger generations, freed from grunt work, can develop faster and engage more deeply with core business functions.

The Urgency of Action and Societal Adaptation

Brockman acknowledged that the rapid pace of AI development can be frightening. He stressed OpenAI's commitment to deliberate progress, with safety as the foremost priority. The company is focused on developing technology in a safe and secure manner, establishing robust standards. He believes that understanding how to best leverage AI's benefits while mitigating its risks will be the most critical conversation for society in the coming years. Issues such as data center development and child safety are core concerns that require careful consideration.

Global Sentiment and Communicating AI's Benefits

Brockman noted a disparity in AI sentiment, with lower levels in the US compared to some Asian and European countries. He attributes this, in part, to a need for the AI field and companies like OpenAI to better articulate the direct benefits of AI to individuals. While AI is becoming a strategic priority for nations, the personal advantages—such as having a "teacher, doctor, lawyer, or therapist in your pocket"—need to be more widely communicated. He shared personal anecdotes, including how ChatGPT assisted his wife with managing health conditions and a friend who narrowly avoided a dangerous antibiotic due to AI-driven information. These stories, he argues, highlight the life-saving and empowering potential of AI, which needs to be integrated into the public consciousness.

The Strategic Importance of Data Centers and AI Leadership

Brockman cautioned against policies that could hinder AI development in the US, such as banning data centers. He argued that such actions would drive data center development overseas, mirroring historical trends in the semiconductor industry. He highlighted the significant number of blue-collar manufacturing jobs created by data centers and emphasized that most operators are responsible actors. Instead of outright bans, he proposed setting standards for "well-behaved" data centers. He believes the US cannot stop AI's global progression and that maintaining leadership is crucial for having a say in its future direction. OpenAI has made commitments to mitigate the environmental impact of its data centers, focusing on closed-loop water systems and avoiding increases in electricity bills.

A Billion-Dollar Commitment to Frontline Defenders

OpenAI has committed $1 billion to provide frontline defenders—organizations and critical infrastructure providers—with access to its models for security enhancement. This initiative aims to help entities like water service providers and hospitals, which may lack the capital for robust cybersecurity, to secure themselves against evolving threats. Brockman views this as a crucial step in addressing pre-existing cybersecurity vulnerabilities in critical infrastructure, which were not adequately built or maintained with cyber threats in mind. He believes this effort is essential for ensuring the safety and security of essential services.

Astra: A Leap Towards AGI

Brockman described Astra as a "step function" advancement, representing the culmination of years of research bets. He noted that previous models saw incremental improvements, making major version bumps difficult to justify. Astra, however, represents a more discontinuous leap, particularly in its computer use capabilities. This allows for long-lived tasks, with the model demonstrating coherent operation for 24 hours to accomplish complex tasks across various domains.

While Astra's capabilities are impressive, Brockman acknowledged that they remain "jagged." He pointed to writing as an area that has improved significantly but is not yet "great." He envisions a future where AI capabilities are more consistently strong across the board, moving beyond a jagged frontier to a steady, uniform advancement. The accelerative and empowering nature of Astra, however, marks a significant departure from previous models.

Addressing Hallucinations and Educating Users

Brockman noted that instances of AI hallucination have become less frequent, yet this improvement has not been widely recognized. He believes continuous education is crucial, with AI proactively informing users of its capabilities rather than users having to discover them. He highlighted that a significant portion of ChatGPT users do not continue to use the platform, indicating a need to re-engage them with the advancements made. The vision for AI is not merely an improved text box, but a conversational, persistent, and context-aware assistant that proactively empowers users in both personal and professional life.

The Future of Human-Computer Interaction

The ideal AI interaction, Brockman suggested, is primarily voice-based, with persistence, memory, and context. This AI should be trustworthy, proactive, and capable of understanding and assisting users with their goals. Explaining its capabilities is a core part of this user empowerment. He drew an analogy to understanding new human colleagues, which involves a period of familiarization and understanding their operating style. Similarly, AI needs to present itself in a way that facilitates effective collaboration, especially as new tools and models emerge. Simplicity and a unified AI experience are key to reducing user engagement with the computer itself, allowing the computer to serve and empower the user.

Strategic Focus and Prioritization at OpenAI

OpenAI's theme for the current year has been "focus," recognizing the impossibility of pursuing every exciting avenue. The mission to ensure AGI benefits all of humanity guides prioritization. This involves bringing technology to bear through deployment and productization. The company made the difficult decision to cancel projects like Sora, despite its high profile, to concentrate on core areas such as unifying consumer and enterprise chat experiences into "Chatwork."

Brockman referenced the management book "The Score Takes Care of Itself," emphasizing the importance of focusing on inputs and the basics rather than solely on outcomes. He has personally focused on the most critical problems he can influence, including data centers, infrastructure, and machine learning engineering over the past two years. This year, his efforts have shifted to the business side, aiming to bring research and infrastructure advancements to the world. He leads by example, diving deep into the details and asking critical questions to ensure clarity and execution.

The AGI Era and Evolving Priorities

Looking ahead, Brockman anticipates a new phase of AI development: the AGI era. This signifies a shift where safety, security, and alignment are integrated from the earliest stages of development and evaluation, not just at deployment. This focus is core to OpenAI's mission. He sees a trend towards deeper integration and intertwining of functions across the organization, from go-to-market strategies to long-term research and chip design, ensuring a coherent approach with shared context and clear objectives.

The Importance of Proactive Helpfulness

Brockman highlighted the need for AI to be proactively helpful, drawing a parallel to how humans interact with new colleagues. While engineers may not always be perceived as inherently helpful, the goal is to develop AI that functions similarly to a trusted advisor, presenting its capabilities and offering assistance. He noted that while some humans use tools like Myers-Briggs to explain their operating styles, AI needs a similar mechanism, especially as it evolves. The ultimate goal is a unified AI that simplifies human-computer interaction, empowering users rather than requiring them to adapt to the machine.

The Future of Employment and Human Value

Brockman believes that AI will continue to surprise, and that human value extends beyond task execution to qualities like relationship building and accountability. He envisions a future of abundance, emphasizing the need for equitable distribution. He anticipates a surge in entrepreneurship, with AI tools enabling individuals to launch businesses more easily. This could lead to a renaissance where younger generations are freed from menial tasks and can focus on higher-level business functions and innovation.

The Urgency of AI Development and Societal Benefit

OpenAI recognizes the rapid pace of AI development and is prioritizing safety and security. The company aims to empower everyone broadly with AI technology, fostering a global conversation about maximizing benefits and mitigating risks. Brockman believes this conversation will become increasingly important over the next one to two years. He also noted that while AI sentiment is lower in the US compared to some other regions, the country's leadership in AI is not guaranteed and requires continued effort.

The Role of Data Centers and Maintaining US Leadership

Brockman cautioned against banning data centers in the US, arguing that it would drive such infrastructure overseas and cede leadership in AI development. He highlighted the job creation and economic benefits associated with data centers. He proposed setting behavioral standards for data centers rather than outright bans, emphasizing that the US cannot halt AI's global progress. Maintaining leadership is crucial for influencing the future of AI development. OpenAI has committed to responsible data center operations, focusing on water conservation and avoiding increases in electricity costs.

A Billion-Dollar Investment in Cybersecurity

OpenAI has committed $1 billion to provide frontline defenders—critical infrastructure organizations, businesses, and governments—with access to its models for security enhancement. This initiative aims to address the existing cybersecurity vulnerabilities in essential services like water supply and hospitals, which were not adequately secured in the pre-AI era. Brockman views this as a critical effort to ensure the safety and reliability of these services. He acknowledged that society has been lax in addressing technological debt and cybersecurity, and that AI now provides both the motivation and the capability to rectify these issues.

Astra: A Significant Step Towards AGI

Brockman described Astra as a "step function" advancement, representing the culmination of years of research bets. He noted that while previous models saw incremental improvements, Astra represents a more discontinuous leap, particularly in its computer use capabilities. This allows for long-lived tasks, with the model demonstrating coherent operation for 24 hours to accomplish complex tasks across various domains. While Astra's capabilities are impressive, he acknowledged that they remain "jagged," with areas like writing still needing refinement. He envisions a future where AI capabilities are more consistently strong across the board.

Continuous Education and Proactive AI Assistance

Brockman believes continuous education is crucial for users to understand AI's capabilities. He envisions AI proactively informing users of its potential assistance, rather than users having to discover it. He noted that a significant number of past ChatGPT users have not continued to use the platform, indicating a need to re-engage them with the advancements made. The ideal AI, he suggests, is not merely an improved text box but a conversational, persistent, and context-aware assistant that proactively empowers users in both personal and professional life.

The Future of Human-Computer Interaction and AI Prioritization

Brockman emphasized the importance of simplifying human-computer interaction, with AI serving to empower users rather than requiring them to adapt to the machine. He drew an analogy to understanding new human colleagues, highlighting the need for AI to present its capabilities clearly. OpenAI's current strategy involves intense focus, prioritizing areas that align with its mission of ensuring AGI benefits all of humanity. This has led to difficult decisions, such as canceling projects like Sora, to concentrate on core areas like unifying chat experiences. Brockman's personal focus has shifted to business execution, aiming to bring research and infrastructure advancements to the world.

The AGI Era and Evolving Priorities

Brockman declared that OpenAI is now in the "AGI era," a new phase where safety, security, and alignment are paramount from the earliest stages of development. He sees a trend towards deeper integration across all functions, ensuring a coherent approach with shared context and clear objectives. His future focus will be on areas requiring this intertwining, aiming for greater concert and lockstep progress within the organization.