The rapid evolution of o3-class AI models is nothing short of extraordinary. In just two months, we’ve gone from AI systems operating at a college level to those demonstrating PhD-level proficiency. The road the Artificial General Intelligence (AGI) has never been clearer and so close than now. Maybe we don’t understand yet the impact of OpenAI’s model o1, but o6 will be AGI.
Such leaps are emblematic of the power of compute scaling and optimization, but even for those of us deeply immersed in AI’s development, the speed of these advancements has been shocking. Humans thrive on change, but this pace isn’t just exhilarating…it’s disorienting. We’re being propelled into a future that demands adaptation at an unprecedented rate.
Short-term impacts
In the immediate future, AI models will display “spiky” performance. They’ll excel in domains where success can be clearly defined and measured, such as math, coding, and logical reasoning. These areas are perfect candidates for well-defined reward functions that reinforcement learning (RL) thrives on. However, tasks requiring more subjective evaluation, such as creative writing or nuanced storytelling, will lag behind. Fiction writing, for instance, lacks the clear metrics that RL relies on, making it a harder nut to crack.
This unevenness will define the next year or two. We’ll see AI models performing at near-superhuman levels in some areas while failing spectacularly in others. Their blind spots will be glaring and frustrating, but they’re also opportunities for refinement. Over the next 1–3 years, I expect these gaps to narrow as we incorporate broader data sets and RL strategies into training. Emotional and sensory data will play a crucial role, bridging existing limitations and enabling these systems to engage more holistically with human tasks.
Once these models achieve a more balanced intelligence, we’ll be staring at AGI-level systems. At that point, the distinctions between human intelligence and machine intelligence will blur, at least for practical purposes. For many, including skeptics, this will be the moment when the reality of AGI becomes undeniable.
Agents and Automation: The 2025 Milestone
In 2025, I predict that AI agents will become an integral part of how we interact with technology. These systems will automate complex workflows, handling tasks that require navigating apps, websites, and various platforms. Imagine instructing your computer to collect data from multiple sources, organize it into a report, and send it to the relevant stakeholders – all without lifting a finger. The market potential for such tools is enormous, and the incentives for AI labs to develop them are clear. The automation of routine computer work is not just a convenience; it’s a revolution in productivity.
This shift will also redefine many industries. Knowledge workers, for instance, will experience varying levels of disruption. Mathematicians, whose work exists almost entirely in the symbolic realm, are particularly vulnerable. AI models excel in symbolic reasoning, and within three years, they will surpass human mathematicians in nearly every measurable way. Tasks that once required years of study and expertise will be accomplished by AI in moments. This transition will challenge traditional notions of intellectual supremacy and force us to reconsider what it means to be an expert in the age of AI.
Of course, the unsolved problems in mathematics will be solved: e.g. Riemann hypothesis, the Hodge conjecture, The Birch and Swinnerton-Dyer Conjecture, Navier Stokes Equations and other.
The changing role of software engineers
Software engineering, in contrast, will undergo a more nuanced transformation. In the short term, engineers will find themselves entering a golden age of productivity. AI will act as a force multiplier, enabling developers to complete tasks faster and with greater accuracy. This will lead also for many non-technical workers to enter this field, hence the “entry” now is knowing how to use prompts.
By the end of 2025, coding will feel less like writing lines of code and more like orchestrating workflows for a suite of intelligent agents. Engineers will specify outcomes, and AI systems will handle the execution.
This new paradigm doesn’t spell the end for software engineers. Coding is only one part of the profession. Engineers bring context, collaboration, and creativity to the table- elements that AI struggles to replicate. For instance, designing software architecture or working closely with stakeholders to understand their needs requires human insight. Engineers who embrace AI tools will become more effective, not obsolete, using these systems to amplify their capabilities.
Over the longer term, however, the profession will evolve in ways that are harder to predict. The difference now is that we’re moving from Python to natural language. While this will democratize coding, making it accessible to more people, the best engineers will still be those who can navigate multiple layers of abstraction.
Robotics: Slower progress, Bigger impact
While AI is transforming knowledge work at breakneck speed, robotics is progressing more slowly. This isn’t due to a lack of intelligence but because of the physical constraints that robots must operate within. Gravity, friction, and hardware limitations all present challenges that are far more complex than symbolic reasoning. Even so, the impact of smarter AI systems on robotics cannot be overstated. These models will play a critical role in training robotic systems, improving their ability to perceive and act in the real world.
Once these barriers are overcome, robotics will revolutionize industries like manufacturing, logistics, and construction. Imagine autonomous robots capable of replicating themselves and mining resources, laying the groundwork for a new era of industrial automation. Although this timeline is longer, likely measured in years rather than months, it’s not decades away. The combination of smarter AI systems and advanced robotics will fundamentally reshape our relationship with the physical world.
The compute arms race
At the heart of AI’s rapid evolution lies the race for compute. o3-class models are a testament to the power of massive compute clusters, and the competition to scale these resources is intensifying. Companies like OpenAI, Meta, and others are investing heavily in supercomputers, knowing that each additional magnitude of compute unlocks exponential gains in model performance.
This arms race is creating a fascinating dynamic. On one hand, some labs are focused on optimizing algorithms to make better use of limited resources. On the other, companies with access to vast compute power are pushing the boundaries of what’s possible. This competition is driving innovation at a pace we’ve never seen before. Interestingly, open-source efforts may also play a significant role, as distributed compute networks challenge the dominance of centralized, closed systems.
AI and Scientific Discovery
Perhaps the most exciting frontier for AI lies in scientific research. Fields like theoretical physics, chemistry, and biology are poised to benefit immensely from AI’s ability to synthesize decades of research into actionable insights. Theoretical disciplines align well with AI’s strengths in symbolic reasoning. If AI can surpass human mathematicians, why not theoretical physicists? The implications are profound. Breakthroughs that once seemed decades away could become reality in just a few years.
This raises important questions about the role of human expertise. As AI takes on increasingly complex tasks, we’ll need to reconsider what it means to contribute meaningfully to scientific discovery. AI has the potential to democratize access to knowledge, but it also challenges our traditional notions of intellectual achievement.
Risks
With all this potential comes significant risk. The most immediate threat isn’t rogue AI but human misuse. From AI-enhanced propaganda to autonomous weapons, the ways in which these technologies can be abused are deeply concerning. Societal instability, fueled by job displacement and misinformation, is another pressing issue.
Regulation will be crucial in addressing these risks. Governments must strike a delicate balance between fostering innovation and mitigating harm. Public perception will also play a critical role. If society isn’t ready to embrace these changes, progress could be slowed or derailed entirely. The biggest bottleneck to AI’s advancement may not be technical, it may be us.
The Future in AI times
Despite these challenges, I remain optimistic about the future. The possibilities that AI presents are extraordinary. Imagine autonomous robots building habitats on other planets, AI uncovering new physical laws, and advanced healthcare systems eliminating diseases. These aren’t distant dreams; they’re plausible outcomes within the next decade.
But achieving this vision will require collective effort. We must ensure that AI development aligns with societal needs and values. This means fostering ethical research, supporting international cooperation, and creating systems that distribute AI’s benefits equitably. The choices we make now will shape the future for generations to come.
We are living through one of the most pivotal moments in human history. The rise of AGI-level systems marks the beginning of a new era, one filled with opportunities and uncertainties.
Our challenge is not just to adapt to this new world but to actively participate in creating it. Whether through innovation, education, or advocacy, each of us has a role to play. The future of AI is not predetermined; it is being written by us.
