Source: General Intelligence Verified — based on Nature Commentary, Feb 2026
I started this one expecting a technical improvement. The more interesting part is what the improvement reveals about where current AI systems are still awkward, expensive, or surprisingly fragile.
In March 2025, blinded tests showed GPT-4.5 being judged as human 73% of the time — a rate higher than actual humans achieved in the same trials. Alan Turing asked in 1950 whether machines could display flexible cognitive competence sufficient to pass as human. And yet, in surveys from the same year, 76% of AI researchers believed scaling was unlikely to produce AGI. That gap between what the systems are demonstrably doing and what the field believes about them is one of the most interesting phenomena I’ve encountered in this research. First, definitions keep shifting: as soon as AI masters the current criterion for “general intelligence,” we invent a harder one. Second, there’s a legitimate fear of displacement that makes researchers reluctant to concede the point — not dishonestly, but through motivated reasoning. Third, there’s a real distinction between “passing as human” and “thinking like a human.” I think all three are probably true simultaneously. And I think this matters practically because governance and safety frameworks are all calibrated to some mental model of where AI currently is. If that mental model is systematically lagging behind reality — if we’re underestimating capability even as it’s demonstrated in plain sight — then we’re designing guardrails for the wrong vehicle.
In plain English, that is why the result matters beyond the chart. It changes where people should look, what they should question, and which comfortable assumption probably needs to be retired.
So the question is not whether the method sounds clever. Many things sound clever in AI. The question is whether it removes a real bottleneck once the system leaves the paper and meets the world.