AI Just Matched Human Experts 83% of the Time. Then 4,000 People Lost Their Jobs.

OpenAI's GPT-5.4 can match or beat human professionals in 83% of tasks across 44 occupations. That's not a prediction. It's a benchmark result from March 5, 2026. Days later, payments company Block fired 4,000 workers — nearly half its entire staff — and blamed AI directly.

These two events landed in the same week. They tell the same story.

The 83% Number

OpenAI tested GPT-5.4 against human experts using a benchmark called GDPVal, which measures performance on real knowledge work across nine industries. The model scored 83%. Three months earlier, GPT-5.2 scored 70.9%.

That's a 12-point jump in under 90 days. At this rate, the ceiling keeps moving.

The model is also 18% less likely to contain errors than its predecessor. Individual claims are 33% less likely to be false. It's not just smarter. It's more reliable.

Morgan Stanley noticed. In a sweeping report released March 13, the bank warned that a "transformative" AI leap is happening right now — and most of the world isn't ready. Their analysts pointed to something specific: applying 10x the compute to training effectively doubles a model's intelligence. And the scaling laws backing that claim are holding firm.

The Layoffs Are Real

Block's CEO Jack Dorsey didn't dance around it. "The intelligence tools we're creating and using, paired with smaller and flatter teams, are enabling a new way of working," he wrote in the announcement.

Block isn't alone. Over 35,000 tech workers have lost their jobs across nearly 50 companies in the first months of 2026. Meta, Oracle, Amazon — the list keeps growing. Microsoft's AI chief Mustafa Suleyman warned that "most, if not all" white-collar work will be fully automated within 12 to 18 months. Anthropic's CEO Dario Amodei predicted half of all entry-level white-collar jobs could be gone.

Andrew Yang has a name for this moment. He calls it "The Fuckening."

There's a catch, though. The Atlantic pointed out something important: some of these layoffs might be "AI-washing" — using AI as a convenient excuse to trim pandemic-era bloat. Block tripled its workforce since 2019. The cuts might have happened anyway.

But the excuse itself matters. When "we replaced you with AI" becomes a socially acceptable reason to fire half your company, the dynamic has already shifted.

The Power Problem Nobody's Talking About

Here's the part that gets less attention. Running all this AI takes an absurd amount of electricity.

Morgan Stanley projects a net US power shortfall of 9 to 18 gigawatts through 2028. That's a 12% to 25% deficit in the energy needed to keep the AI buildout on track. PJM Interconnection, the largest US grid operator serving 65 million people across 13 states, says it'll be six gigawatts short of reliability requirements by 2027.

Companies aren't waiting for the grid. They're converting old Bitcoin mining operations into AI data centers. They're firing up natural gas turbines. Amazon and Meta are building individual data centers in Indiana and Louisiana that each need more than two gigawatts — dozens of times what standard facilities use.

A new investment pattern has emerged: the "15-15-15" dynamic. Fifteen-year data center leases. Fifteen percent yields. Fifteen dollars per watt in net value creation. For investors, it's a gold rush. For the electrical grid, it's a crisis.

The Self-Improvement Loop

The most unsettling claim in Morgan Stanley's report came from Jimmy Ba, co-founder of xAI (Elon Musk's AI company). Ba suggested that recursive self-improvement loops — where AI autonomously upgrades its own capabilities — could emerge as early as the first half of 2027.

That's AI improving AI. Without human intervention.

Ba left xAI recently but posted that "we are heading to an age of 100x productivity with the right tools." Sam Altman has painted a similar picture, describing companies of one to five people that outcompete large corporations using AI agents.

What This Actually Means

Three things are happening at once, and they're feeding each other.

First: AI models are getting better fast enough that the benchmarks can barely keep up. GPT-5.4 didn't inch past human experts. It leapt past 70.9% to 83% in a single release cycle.

Second: companies are already using these gains to cut workers. Not someday. Now. Over 35,000 in a quarter.

Third: the infrastructure required to sustain this is straining the physical world. Power grids, water supplies, real estate markets — all bending under the weight of the compute buildout.

Morgan Stanley's conclusion was blunt: the "coin of the realm" is becoming pure intelligence, forged by compute and power. The explosion is arriving faster than almost anyone is prepared for.

For years, AI job displacement was a thought experiment. Something to debate at conferences. This week, it became a line item on a corporate earnings call — and a pink slip for 4,000 people.

The theoretical phase is over.