Shanaka Anslem Perera ⚡@shanaka86
Elon Musk said the biggest mistake he ever made in hiring was overweighting intelligence.
The man who built a neural network that learned physics from 9 billion miles of driving data. Who taped out AI5 on April 15. Who runs the Colossus supercluster training a 1 trillion parameter model. Who is building Terafab to produce 1 terawatt per year of AI compute. That man looked into the camera and said: “I think goodness of heart is important. I underweighted that at one point.”
Now look at what his own machines just did to the argument for pure intellect.
Frontier AI models scored above 94 percent on GPQA Diamond in April 2026. This is a PhD-expert-level science benchmark so hard that questions are only included if non-expert PhDs with internet access and 30 minutes cannot answer them. Human domain experts score 65 percent. Some re-benchmarks pushed that to 69.7.
The machines beat the smartest humans alive by over 24 points on questions designed to require deep doctoral expertise.
Intelligence is being commoditized in real time. Not in theory. In reproducible benchmark scores that improve with every training run. Grok 4.3 is finishing its 1 trillion parameter checkpoint this week. Claude scores 94.2. Gemini 94.1. The cost of PhD-level reasoning is collapsing toward the price of an API call.
Musk saw this before the benchmarks confirmed it. His hiring process at Tesla and xAI now requires no resume and no cover letter. Three bullet points describing the toughest technical problems you have ever solved. A 20-minute conversation where, as he put it, “if the conversation is not wow, believe the conversation, not the paper.” And then the filter that cannot be faked: talent, drive, trustworthiness, and goodness of heart. He called those traits “fundamental” and “unchangeable.”
This is not soft management philosophy. This is resource allocation under scarcity.
When cognitive ability was scarce and expensive, you optimized hiring for IQ. You hired the credential. The degree. The prestige signal. The college wage premium held at 62 percent because intelligence was hard to find and impossible to replicate at scale.
That constraint just broke. An API call costing cents now outperforms a PhD on the hardest science questions humanity has ever constructed. The 62 percent wage premium has stagnated for two decades while AI capabilities doubled annually. The crossover is not coming. It arrived.
So what is still scarce?
Not computation. Terafab will produce that at planetary scale. Not knowledge. Every frontier model contains more factual information than any human who has ever lived. Not reasoning. GPQA Diamond proved that.
What is scarce is the thing Musk identified: whether someone will do the right thing when nobody is watching, when the deadline is impossible, when the shortcut is invisible, and when the cost of integrity is personal. That cannot be trained into a neural network. It cannot be fine-tuned. It cannot be distilled from data. It emerges from a life lived with a specific set of values that no architecture can replicate.
The man building infinite intelligence just told you the only thing it cannot produce.
Character is the last advantage that cannot be automated. And the person who understands that best is the one making intelligence cheapest.