Have you noticed that the entire tech world is gripped by a kind of "speed mania"?
Writing code, searching literature, drafting proposals... with just a single prompt, AI will serve you the most precise, most perfect answer in 0.5 seconds. Everyone is frantically optimizing their workflows, surgically removing every bit of waiting, every hint of friction between "asking" and "getting an answer."
Yet Terence Tao, widely regarded as one of the greatest minds alive and a Fields Medalist, poured a bucket of ice-cold water on humanity's efficiency狂欢 in a recent in-depth interview:
"AI may be making us too efficient." "Life needs a little distraction—it brings enough randomness and 'high temperature.'"
This statement—which sounds like an excuse for slacking off—comes from a prodigy who has dominated mathematics since age 10 and proved that prime numbers contain arbitrarily long arithmetic progressions. It has utterly shaken the global research community.
Why would absolute efficiency become poison for innovation? Why would eliminating every "detour" potentially put an end to humanity's original discoveries?
It all begins in a dusty old library.
The Disappearing "Adjacent Shelf": Precision Search Is Killing "Serendipity"
Tao offered a highly evocative example.
Back in his graduate school days, if you wanted to find a paper in an academic journal, there was no shortcut. You had to walk down long corridors, enter a dimly lit library, and flip through rows of massive wooden shelves one by one.
After finally pulling out that hefty volume, just as you read your target paper, your eyes would often casually drift to the next article in the same journal, or to an entirely unrelated book on the adjacent shelf.
"Sometimes it was completely useless, but many times, you would suddenly stumble upon stunning inspiration," Tao said.
▲ In his Dwarkesh interview, Tao pointed out that leaving some distraction in life brings randomness and "high temperature"
In the history of science, this wonderful phenomenon is called Serendipity. Fleming returned from vacation to find bacteria-free zones around a moldy petri dish—and penicillin was born. An engineer melted a chocolate bar in his pocket with microwave radar—and the microwave oven was invented. Even 3M's "too weak" adhesive accidentally became the world-famous Post-it Note.
The miracles in scientific history almost always spring from such "unplanned collisions."
But what about AI-powered search today?
You enter a highly precise prompt, the algorithm performs vector approximation across billions of parameters, and within fractions of a second it precisely hits the exact point you wanted.
You get what you wanted—and may miss the entire universe you hadn't thought to look for.
This concern is backed by research. A major 2021 study published in the prestigious journal PNAS (analyzing 241 disciplines, 90 million papers, and 1.8 billion citations) delivered brutal data: as the number of digital papers explodes and search becomes increasingly efficient, newly published papers in science are becoming more and more fixated on a small set of "highly cited canon".
There are too many papers, and they can be found too quickly; everyone heads straight for the standard answer. It's like pouring sand too fast and too forcefully—individual grains never get the chance to trigger local avalanches. Disruptive innovation is being suffocated to death by the torrent of hyper-efficiency.
The Ghost of Princeton: When the Environment Is "Absolutely Focused," the Brain Dies
Many people believe the ultimate state for innovation is this: give a genius a quiet cottage far from the world, block out chores, administrative tasks, and meetings, and let them focus absolutely for 24 hours a day.
Tao once believed this too.
He spent an entire year at the Institute for Advanced Study (IAS) in Princeton—the holy land of the world's top scholars. It was where Einstein, von Neumann, and Gödel once worked. Teaching duties and trivialities were kept outside the gates, and everyone's sole goal was pure thinking.
▲ The Institute for Advanced Study (IAS): a friction-free paradise scholars dream of, yet hiding the paradox of "inspiration burnout"
The first few weeks were like paradise. Tao finished papers that had been backlogged for years in one go, and could think through complex mathematical problems for hours without interruption.
But after a few months, something very strange happened.
Tao found his inspiration had completely dried up. Faced with a perfectly blank whiteboard, his brain went on strike—he even found himself mindlessly refreshing the internet out of sheer idleness.
This echoed the scathing remarks physicist Richard Feynman made about the IAS decades earlier: Feynman observed that those top minds shut away in beautiful woodland cottages often fell into profound guilt and depression, because they were cut off from the messy, hands-on work of the laboratory. No young students came running to them with those "absurd yet stimulating" lay questions, and the gears of their brains ground to a halt.
Tao explained this with an exquisite term borrowed from statistical physics and machine learning: "The brain needs high temperature."
In the world of algorithms, "temperature" controls the randomness of sampling. If you set the temperature to 0, the algorithm becomes mechanically rigid, forever circling in a deep pit of "local optima." Moderate interference, the noise of chores, and the wandering of a distracted mind are precisely the "thermal perturbations" that heat the brain up—they can violently jolt you out of a dead end and hurl you into dimensions you've never set foot in before!
The Algorithm's Open Secret: Only "Exploit," No "Explore"
In reinforcement learning, there is a classic problem called Explore vs. Exploit.
- Exploit: frantically harvest along the most profitable, safest known path;
- Explore: risk failure to fumble around in unknown, dark territory.
An X user @PAClearning captured the essence of the AI era with a single, terse phrase in Indonesian: "Exploit tanpa explore (exploit without explore)."
▲ In-depth analysis in the comments: over-optimized search eliminates the random friction needed for cross-domain pattern recognition
As commenter Ajit put it: "Over-optimized search erases random friction, and cross-domain pattern recognition happens precisely within that friction. When you use algorithms to strip away all the unexpectedly adjacent 'noise,' you are trading major discoveries for rapid convergence to expected answers."
If every knowledge worker in our entire civilization spends every day using AI to converge in seconds to "the highest-probability known answer," then society's cognitive bandwidth will be locked into the mean of existing knowledge forever.
The Fatal Question: Will AI Make World-Changing Geniuses Extinct?
This concern is grounded in reality.
Jarek Duda, the inventor of Asymmetric Numeral Systems (ANS)—the mathematical foundation underlying Apple's operating system (LZFSE compression), Facebook's backbone network (Zstandard compression), and the next-generation image format JPEG XL—saw Tao's remarks and immediately posed a soul-searching question:
▲ ANS inventor Jarek Duda asks: In an era where AI automatically selects from known optimal solutions, will disruptive algorithms ever be discovered again?
Before ANS existed, "arithmetic coding" had been repeatedly and independently invented in the computer science world, and was revered as an unassailable standard answer. The reason Duda was able to invent ANS—which crushed all predecessors—was precisely that he maintained a kind of "productive ignorance" outside the mainstream paradigm and veered onto a side path that no one had ever explored.
Duda posed a sharp rhetorical question: "If it were today, when AI automatically selects from the best known methods for you, would this kind of disruptive thing ever be discovered by a human?"
The answer is chilling.
AI is the ultimate "highway builder"—it can pave existing roads smoother than light itself. But civilizational leaps often come from someone aimlessly catching butterflies in the roadside grass, and accidentally falling into a mysterious wormhole.
Reclaiming the "Wasted" Friction
In the discussion thread on X, one highly liked reply struck a nerve with countless people:
▲ User Helgrim on the blank space between question and answer: "We always treat everything between a question and its answer as wasted time, but sometimes that's exactly where learning happens."
"We have always treated everything between 'asking' and 'getting an answer' as wasted time; but sometimes, those seemingly inefficient detours, struggles, and accidents are precisely where learning and epiphany happen."
AI is indeed the greatest tool humanity has ever created. It can produce reports for you in seconds, write scripts for you in seconds, and clear away the obstacles of mechanical labor.
But always remain vigilant: never let your own brain degenerate into a prompt-responding machine that only pursues instant rewards.
Go to a physical bookstore and aimlessly flip through two obscure, even terrible books. Let your thoughts wander freely during a walk. Allow yourself to take a few seemingly ridiculous detours when solving problems...
Because in an era where cold algorithms rule everything, those seemingly careless distractions, wanderings, and frictions are precisely the last refuge of human spirit and creativity.