It's the first time in the history of the world.
AI has exploded directly into human intelligence testing.
Mensa Norwegian IQ test, 35 difficult to decipher graphical reasoning puzzles, limited to 25 minutes.
Claude Fable 5.1, in addition to the GPT-6 Astra, was shot dead in the field, and the result was a direct blow-out of the scroll.The theoretical limit of 151 points。
It was also the last seven re-exams and the second full-time.They have only rolled up their ceilings, and no one knows where their ceilings are.
Putting it in humans,Almost 98% of the world's population can't even touch a 130-point line.You and I are probably among the ones who are being squeezed.
If you don't believe in evil, try the 33rd question of Hell's Hardness in this volume, and the answer is at the end of the first paragraph.
Test 33 of Mensa NorwayDon't be discouraged. This is not just a problem for people, the vast majority of AI is involved here too.
This scroll is scrolling without examining any knowledge reserves, and it gives you a bunch of mindless broken graphs to see if you can figure out the rules yourself in a messy mess.
Psychologists call this "fluid intelligence", which is the last line of defense against self-deprecating machines.
As a result, this line of defense is now being blocked.AIIt's not like I'm going to be able to do anything.
Humans measured only 145, and AI tested 151.
The first thing you see is a clock-shaped curve.
That's the distribution of human intelligence, and the vast majority of people are crammed into a mountain range near 100 points.
In the past, AI images were scattered on the lower left-hand corner of the curve.
Now, they are squeezed one by one to the solitary zero tail at the far right of the curve, with the heads all over one another.
Tracking AI's MSA Norway scores for each AI, with the full line on the right side of this scroll.The list was compiled by American journalist Maxim Lott.
He's like a tough supervising teacher who tests more than 30 mainstream AI's every week, with some models listening to the text and some looking directly at the original answer.
He used the Mensa Norway test, which was intended for humans to do. The system only scored a maximum of 145 points.
Tracking AI scores to AI, and the number of questions to be answered continues to go up and down, and the total of 35 pairs is 151.It has gone beyond what Mensa has set for humans.。
However, 145 caps are the limitations of this scroll, and the IQ of the crowd is not capped.
We estimated this by the standard distribution of intellectual property.It's about one out of 3,000 people with 151 points.。
Of the more than 8 billion people in the world, there are only about 2.7 million people with 151 points or more, and there are about 3,000 in a city of 10 million people.
The entrance line for the Mensa Club is 130 points, and the two AI players have scored 21 points.
And 151 is just the ceiling of this scroll, and if the scroll can go upwards, it will reach 160 points, leaving just over 20,000 people worldwide.
AIThis is a small group of people standing at the top of the pyramid of human intelligence.
And it's more than just these monsters that can squeeze into this little bunch.
On TrackingAI's list, GPT-5.6 Sol and Gemini 3.1 Pro scored 145 points, and models that broke 140 points in the public domain have been ranked in the long list, while domestic models have also finished first.
So many models are crammed into high segments, and the gap is drawn into the last few stress-axis problems.
Tracking AI has calculated that the first 10 questions can be sorted almost by an AI, and at the end of the roll, only five of the 35 questions are correct, and only two of the 33 questions are left.
According to Lott's calculation, there are only 148 points wrong in one of the 35 rounds, so do the two for the 33rd, which are basically the two players with the full score.
The first question to be asked is to publish the answer here.The correct option is E.。
The correct number of AI for each question, orange is the MSA Norwegian volume, with only two remaining for the 33rd question.It's scary enough to know that a few years ago, AI was still a highly biased student.
From 64 points to 151 points, AI only took two and a half years
In March 2023, a Finnish psychological evaluation expert gave ChatGPT a formal Wei's adult intelligence test, which examined only vocabulary and common language questions.
ChatGPT directly excavatedLanguage IQ of 155It killed 99.9% of humanity in seconds.
However, when he encounters "What's Sebastian's father's name", his brain turns and he dies instantly.
In the words of this expert, ChatGPT is "without eyes, ears and hands".
As it turns out, language and knowledge have always been the most hands-on subjects in AI.Graphic reasoning is its biggest soft rib.。
This is exactly what Menssa Norway is trying to figure out.
The matrix problem was first devised by the British psychologist John Raven in 1936, and is widely considered by psychologists to be the best way to measure a person's overall intelligence.
This hard bone has been dormant for 60 years, starting with MIT's first program of geometrical analogies in 1962.
We're working on a timeline of IQ questions based on publicly available AI.Until the beginning of 2024, Gemini Advanced will be the simplest 1st question in the Mensa volume and will be able to correctly formulate a series of "A+B=C" equations without any calculations.
In February 2024, Gemini Advanced solves a number of equations in a vacuum.Lott translated one of the lines into text for the AI, and the result was no good. Some AI were only mistaken for six lines, scored a disgraceful 64 points, and blinked the answer card with their eyes closed, and the best Claude-3 had just crossed the human average.
So Lott at the time asserted that AI did not yet have the general intelligence of humans.
In September 2024, OpenAI's o1 learned how to think and move forward.Before you open the door, think internally, step by step, test and check.More than half a year ago, Lott said AI didn't work, and this time it was titled "A huge breakthrough in AI intelligence".
Since then, the number has risen steadily, and in April this year, the first 151 points have been dropped from the list.
We compiled data from TrackingAI and Maxim Lott, each of which was the record for most updates at the time.So, in just two and a half years, a 60-year-old college student scored 151 points from 64 blind and visually impaired students.
According to Lott's statistics, from May 2024 to October 2025, the IQ scores of the top AIAn average of 2.5 points per month。
In contrast, there is evidence in psychology of how slow human growth is.
Throughout the 20th century, with improvements in nutrition and education, human IQ scores have increased.That's about three points a decade.This is the famous Flynn effect.
This comparison.It takes more than 30 years for humans to climb 10 points.AIIt's only four months away.。
So quickly, the first reaction of any normal person is, did he stealthily come up with the answer?
It's the same as changing the rolls that didn't go through the net, so it's close to full.
Lott was also suspicious that year.
After all, the MSA Norway test has been online for years, and the topics and answers have long been cleaned up by netizens.
Even Tracking AI itself has published the text version of 35 questions along with the correct answers, so it's likely to have been printed during AI training.
TrackingAI opens the Norwegian version of the Mensa theme, and the one on the right is the correct answer.In order to rule out this possibility, in May 2024, he specifically sought out a new set of questions for a member of Mensa from scratch.
This is a question for readers who have tried a round of scoring criteria and have never been exposed on the Internet and cannot be found in any AI training data.
After the change of the scroll, the scroll effect did come to light. To this day, some models have replaced the secret scroll with a twenty-fold scroll instead of a public scroll.
But the head model changed a scroll that I'd never seen before.So close to completion.。
The encrypted volume contains only 16 questions, with a maximum score of about 136, and OpenAI's GPT-5.6 Terra view graph has killed 135.
Fable 5.1 and Astra also have 130 points, alongside a state-of-the-art model.
The results of the secretive scroll are distributed, the scroll never crossed the net, and the thread is its full line.Last October, Lott also raised a bamboo flag, estimating that AI will have to wait at least two more years to get a full score on the graphic version of this scroll and say that Halloween 2027 will return for approval.
In less than a year, the GPT-5.6 Terra hit the front of the line.More than a year earlier than he predicted。
Since there is no explanation, there is only one conclusion left.I was blind two and a half years ago.AIIt's a really long story.
It's hard to solve the problem of AI, and humanity is almost done.
The scrolls are filled, and the people with the questions can only change more difficult.
Therefore, tracking AI is marked next to the full score line as "highest possible score (temporary)".
At the end of last year, Lott calculated that this year it would be more difficult to fill in the gaps of more than 130 points to replenish the gaps of high-level AI.
Of course, there are those who do not obey.
François Chollet, the "Father of Keras", designed an ARC-AGI reasoning test specifically designed for AI.
He recently argued that if we define intelligence as the ability to transform experience into efficiency, that is, to see the same number of examples of people who learn faster, AI today is about six times the number of human beings, about a million times less.
Chollet's own end-line for AGI is also in question.
In February, he revealed that ARC-4 was on the rise, launching early next year, with the 6th and 7th generations to follow.
In his own words, the key is always to come up with new questions.I'm not going to say anything about what humans can do.AIIt's just that the title doesn't stop.。
At the time, he estimated the time for AGI to be after 2030.
When someone asked him earlier this month if the verdict of 2030 was still countless, he changed his mind.It will be sooner.He's going faster than he anticipated.
Someone asked Chollet if the judgments for achieving AGI in 2030 are still uncalculated.
The IQ test was born in 1905. For 120 years, from school divisions to military recruits, humans have been able to use this measurement for the same class of seats.
It's a premise behind the scale that almost no human has ever doubted.Only a man can be measured, and only a man can stand on the highest scale.
Now the scale is still there, but the presupposition has collapsed. A root is not a man's thing; it has climbed in succession from head to tail, and the root line at the top has not been able to stop it.
Lott can also make more rigid rolls, and Chollet can make the next generation of ARCs.
But as AI has gone from blindfolded to full-blown for two-and-a-half years, how many unresolvable questions can be solved in the human brain, no one dares to book a ticket.
In the days leading up to the complete exhaustion of the human database, what we need to rethink is not just how intelligent AI is, but what is left of it after we remove the "intelligent" layer.
RoboInd editorial channelResearch, robotics companies and market intelligence for cross-border sourcing.Explore the channel ↗
Share this
End of editorialExplore more from Global Robotics News ↗