Now is the Artificial Intelligence Singularity

By Brian WAng

In 2015, Waitbutwhy.com had some articles and diagrams explaining how the artificial intelligence revolution of a technological singularity would play out. The diagram above shows how there would be a series of improvement rates that would get steeper and steeper and faster and faster.

My article from yesterday explains how 12-20 years ago we shifted from Moore's law to GPU scaling to AI-LLM scaling and now to XAI scaling speed.

Moore law. -double every 2 years, 1000X compute in 20 years (40-60 year duration)

GPU-AI-LLM scaling - 5X every year in AI compute, 15000 times in 6 years (12 years of duration)

XAI scaling since April 2024 to today and projected through at least 2026 and out to 2030 (11-15X every 6-9 months, 4 million-200 million in 6 years.

A debatable AGI will be in 2025. This will be Grok 4, Grok video and Grok voice. Second half of 2025, unsupervised general robotaxi, thousands (~10K) teslabots in factories doing useful work.

Artificial General Intelligence (AGI): Sometimes referred to as Strong AI, or Human-Level AI, Artificial General Intelligence refers to a computer that is as smart as a human across the board—a machine that can perform any intellectual task that a human being can.

Clear AGI and debatable ASI (superintelligence in 2026).

Clear superintelligence in 2027. It will no longer be reasonably debatable that superintelligence has arrived in 2027. In 2028-2029, those still saying superintelligence is not here will be viewed as just in denial.

Artificial Superintelligence (ASI): Oxford philosopher and leading AI thinker Nick Bostrom defines superintelligence as “an intellect that is much smarter than the best human brains in practically every field, including scientific creativity, general wisdom and social skills.” Artificial Superintelligence ranges from a computer that’s just a little smarter than a human to one that’s trillions of times smarter—across the board.

The above chart from 2015 has all human brain 10**25 flops per second happening about 2050. The new rate of progress could see 2 × 10²² FLOPS by the end of 2025 and used for training in 2026. 10**25 could be 2027-2028 with 10 million chip data centers and using two more major chip upgrades. Rubin in 2026, Ultra in 2027, new architecture in 2028. Tesla will make Dojo 2 in late 2025, Dojo 3 in 2026, Dojo 4 in 2027 etc... OVERLAY the TOP GRAPH of corrected increasing improvement rates over this graphic.