Introduction

If you’ve used cutting-edge AI models like GPT o1 Pro (/w Deep Research), Grok-3, or DeepSeek, you’ve likely experienced that "iPhone moment"—the realization that AI is not just another technology but a fundamental shift. But if you haven’t been using these frontier reasoning models yet, it’s easy to be misled into thinking AI is just another overhyped buzzword.

Your skepticism will feel justified—that is until the seemingly solid ground suddenly shifts beneath your feet. Don't get left behind - get on the front foot.

At the end of 2024, some were claiming AI was entering the “trough of disillusionment.” The rapid leaps of the previous year seemed to slow, with only incremental updates between June and December. Then, almost overnight in early 2025, deep search and reasoning models arrived—transforming what was possible. AI progress isn’t going to come as a steady march forward. We will see sudden, seismic spikes, leaving the unprepared scrambling to catch up. Like playing chess, boxing or snowboarding , the best strategy in these games/sports isn't to wait and react. Being on the back foot is the worst approach. Instead an attacking stance - is much better —anticipating moves, preparing early, and seizing opportunities before they become obvious.

Similarly with AI, I suggest businesses and professionals get started on the "long-arc":

• Invest in infrastructure and data to ensure AI can be integrated effectively.

• Encourage teams to experiment with AI tools, developing hands-on familiarity before the stakes are too high.

• Identify the first principles of your workflow — so that you don’t end up just automating old processes but designing them to fully leverage the new options created by AI.

The question isn’t whether or not AI will reshape industries—the change is already underway. The real question is whether you’ll meet that change on the front foot or the back foot. AI will continue to improve (fast). Some are hesitating to act on AI because they are waiting for the AI trend to plateau and things to become clear. They highlight all the technical challenges for AI to overcome and expect things to stagnate. However, let's be clear. This technology is the worst it’s ever going to be—from here, it only gets better.

Let’s look at four potential drivers of continued rapid improvement in AI

1. Compute power is exploding

The scale of investment in AI compute is unprecedented. History show that more compute = smarter AI.

Consider the numbers:

• GPT-3 was trained with ~1,000–2,000 GPUs.

• GPT-4 used ~25,000 GPUs.

• The next-gen frontier models are being trained on mega-clusters of 200,000–300,000 GPUs, consuming as much energy as ~2 nuclear power stations(!). (Note that talking about GPU numbers is a simplification —it’s actually about the level of compute). AI training compute has been roughly doubling every 6 months (it's an exponential).

2. Better training and engineering techniques

The main labs have many promising training & engineering methods still to try:

• Distillation and optimization techniques are making AI models faster and more cost-effective to run. Labs are optimizing every layer of AI training, and DeepSeek demonstrated how CUDA-level programming improvements to new model architectures drive massive efficiency gains in model performance.

• Self-play and reinforcement learning—like the techniques behind AlphaZero—are being applied to general AI models, improving their ability to reason and self-improve.

Frontier labs aren’t just scaling AI’s intelligence; they’re also looking at how to expand it's cognitive capabilities:

• Memory—so AI can retain and refine long-term knowledge.

• Planning—enabling it to strategize rather than just react.

• Creativity—moving beyond pattern recognition to true innovation.

3. Approaches outside LLMs and Generative AI

While Large Language Models (LLMs) and generative AI get the headlines, the broader AI movement is driving breakthroughs across science and industry:

• AlphaFold, DeepMind’s protein-structure prediction model, has mapped over 200 million proteins, turning a process that once took years into seconds.

• GNoME (Graph Networks for Materials Exploration) has used deep learning to predict 2.2 million new crystal structures and ~380,000 stable new materials.

Demis Hassabis, CEO of DeepMind, has openly suggested that Artificial General Intelligence (AGI) could be just 3-5 years away. Whether that timeline holds or not, it’s clear that AI isn’t slowing down—it’s accelerating. Learn now. Act now. Lead now.

The rate of AI/technology progress is (as it always has been) un-precedented and accelerating. The increase in speed means also an increase in uncertainty. We cannot see ahead clearly what is going to happen. So we need to act even while we don't know exactly where things are going.

We need to have more courage

As AI scales intelligence, leaders must scale courage.

• Courage to embrace uncertainty, knowing that agility beats perfection.

• Courage to rethink how we work, instead of clinging to outdated models.

• Courage to create change, rather than waiting to be forced into it.

So don’t just observe from the sidelines—engage with it:

• If you’re a business leader, start piloting an AI project this quarter.

• If you’re a manager, train your team on AI tools and review your workflows.

• If you’re a policymaker, craft smart guidelines that foster innovation.

And if you're one of the above or none of them: this is the time to learn. Try the latest and most powerful models liks o1 Pro, Grok-3, Claude 3.7 and DeepSeek R1. Take an AI course. Start discussions at your workplace. Upgrade your workflows.

The age of AI isn’t coming—it’s here. Your future self and your team will thank you for acting today.