"Jim, How Fast Is AI Really Advancing?"

Why I Believe Most People Underestimate the Curve, and Why That Matters

I’m often asked a deceptively simple question... “Jim, how fast do you believe AI is actually advancing?”

It’s usually asked with genuine curiosity... sometimes excitement, sometimes skepticism, and often by people who work around technology but not deep inside it. Given that I’ve spent over 30 years in IT, alongside many years following developments in quantum physics, quantum computing, and artificial intelligence, I try to answer honestly. Yet I often walk away with the sense that my answer didn’t quite land. Not because it was wrong, but because the way humans intuitively understand progress is fundamentally mismatched with how AI is evolving.

The Problem Isn’t AI. It’s Human Intuition.

Humans are very good at reasoning about linear change.

• If something improves by 10% a year, we can visualize that.

• If a system gets “a little better” each month, we understand it.

What we’re bad at is understanding exponential growth, especially when it starts slowly and then accelerates beyond intuition. This isn’t a flaw, it’s just biology. And it's important to understand that AI is not progressing linearly.

The Old “Penny That Doubles” Problem

There’s an old thought experiment that many of you may remember... "Would you rather have one million dollars today, or a penny that doubles in value every day for 30 days?"

Most people pick the million dollars. In terms of which will provide you the most money or income in the end, they’re wrong. The penny wins, dramatically, but only in the last few days. For most of the month, it looks unimpressive, even trivial. Then suddenly, it’s not.

AI advancement feels exactly like that penny.

For weeks, months, even years, progress looks incremental...

• Slightly better text

• Slightly better images

• Slightly better automation

Then one day, entire workflows quietly disappear.

My “Guess-timate”... AI Advancements Are Doubling Every 2–3 Months

When people press me for a number, I often say something like this, “I’d estimate that AI’s effective computing intelligence is doubling every two to three months.”

That statement often triggers disbelief.

So let me be precise... I am not saying that AI is becoming “twice as smart” in a human sense every few months. What is doubling, or compounding at a similar rate, is something far more impactful. The amount of useful cognitive work AI can perform per dollar, per unit of time.

That distinction matters.

AI Isn’t One Curve... It’s Several, All Compounding

AI progress is not a single metric. It’s the interaction of several accelerating curves:

1. Compute Scale

The amount of compute used to train frontier models has been doubling far faster than Moore’s Law ever did, often measured in months, not years.

2. Algorithmic Efficiency

Models today do more with the same compute than models from even a year ago. This means capability increases even without more hardware.

3. Data and Feedback Loops

AI systems now:

• Generate their own synthetic training data

• Learn from real-world deployment

• Improve through continuous feedback

This creates recursive improvement... systems that get better at getting better.

4. Cost Collapse

What required massive budgets two years ago is now available to individuals, startups, and mid-sized companies.

When cost drops, adoption accelerates. When adoption accelerates, feedback increases. When feedback increases, improvement compounds.

Why Progress Sometimes Feels “Sudden” To Some

The most important changes don’t announce themselves, they emerge quietly, then become unavoidable. AI does not replace “jobs” all at once.

It replaces:

• Task fragments

• Decision latency

• Junior-to-mid expertise

• Coordination overhead

• Cognitive friction

Organizations don’t wake up one morning replaced by AI. They wake up one morning less competitive, and can’t quite explain why.

Why People Think We’re Still “Early”

I do feel they are correct, but for the wrong reason.

People think we’re early because -->

• AI still makes mistakes

• Outputs don’t feel “human”

• Systems need supervision

We’re actually early because:

• The curve hasn’t gone vertical yet

• The compounding hasn’t reached its visible phase

• The economic impact hasn’t fully surfaced

We are not at the end of the story, we are near the end of the introduction.

So, Is There A Better Way to Frame It?

When I want this to land with executives, engineers, or skeptics, I reframe the question, “AI may not be doubling in intelligence every few months, but the amount of work it can do per dollar almost certainly is.”

That framing avoids science fiction and anchors AI where it belongs:

• In economics

• In productivity

• In leverage

That’s where disruption actually happens.

Every major technological shift looks slow... until it doesn’t. AI is no exception.

What makes this moment different is not just the speed of progress, but the self-reinforcing nature of the systems we’re building. Tools that learn, improve, and scale themselves are fundamentally different from anything that came before. The people who struggle to grasp this aren’t unintelligent, they’re simply applying linear intuition to an exponential world. And history has shown us, repeatedly, how that ends.