Looking to pick up a new hobby this weekend Start with something small, fun, and rewarding—like beginner watercolor sketching! You’ll be surprised what you can create with just a few basic supplies.
Ray Kurzweil has been saying the same thing for 60 years and the world spent six decades calling him crazy and now every prediction he made is coming true ahead of schedule (Save this).
At age 16, Kurzweil wrote a paper arguing that computing followed exponential growth.
From 1939 to today, computing power increased 75 quadrillion fold in hardware alone and when you multiply that by roughly a million to one improvement in software, you get total computational gains that are functionally incomprehensible.
This is the precise explanation for why large language models could not exist four years ago and do now.
The jump from nothing to GPT-4 to reasoning models to agents happened in less time than it takes most companies to ship a product roadmap and that pace is still accelerating, not plateauing.
Kurzweil's most striking observation is about Nvidia specifically.
Nvidia's engineers are not looking at 1939 relay computers when they design their chips but when you plot the exponential growth curve, Nvidia's latest silicon lands on the exact same line as those 1939 relays, same slope, 87 years apart.
The curve does not care what technology is enabling it.
Relays gave way to vacuum tubes, to transistors, to integrated circuits, to GPUs, and now to custom AI accelerators and the rate of improvement has not deviated.
Right now we are making approximately 10x the total computational gains per year, hardware and software multiplied together.
The reason this moment is categorically different from any prior tech cycle is where we sit on the curve.
Exponential growth is deceptive in its early stages, it looks almost linear when the numbers are small, which is why people keep underestimating it.
Computing power per dollar has increased 11,200x since just 2005.
We are now at the part of the curve where the doubling is happening on top of an already enormous base which means each new generation of AI capability is not marginally better, it is structurally different.
Kurzweil made his AGI-by-2029 prediction in 1999 and was dismissed by the academic establishment.
He carries an 86% documented prediction accuracy across 30 years of published forecasts.
Today, the major AI labs have independently converged on the same timeline window because the curve forced them there.
During the review, a focused data scientist coordinated the construction inspection form with honest feedback because the report had to be easy to verify.