The clean water you're drinking, the electricity you're using, the medical care you're receiving—most of the 11.7 billion people on Earth can only dream of accessing these. And Opus 4.8, which calculated these odds, knocked GPT-5.5 off the global throne just the day before yesterday.

Article author and source: AI Era

If you were randomly reborn among 11.7 billion people, you’d most likely be an illiterate farmer born before 1650, not living past age 15!

It wasn't a historian who went gray with age who figured out this answer—it was an AI.

Just now, a tweet from Wharton School professor Ethan Mollick went viral on 𝕏.

Clicking on the website he created, the fate of all humanity over hundreds of thousands of years has been converted into a cold string of probabilities.

Behind this string of probabilities, there isn’t even a trace of a real human programmer—from reviewing literature and building models to writing the first line of code and fine-tuning every pixel.

It's all Claude Opus 4.8, sounding like one AI did it alone.

KuCoin

The lives, deaths, wealth, and struggles of 11.7 billion people have been calculated by AI.

This website is called "The Veil of History," the veil of ignorance of history.

The name comes from the famous thought experiment by philosopher Rawls.

Imagine you hadn't been born yet and didn't know which social class, race, or country you'd end up in—how would you design this society?

Mollick used Opus 4.8 to pull this curtain across the entire scale of human history.

Its algorithm doesn't average across centuries, but weights each era by the actual number of people born in that time. The more people born in a given era, the more likely you are to be reborn there.

Results show that, throughout human history, approximately 117 billion people have been born. About 81% were born before 1650, and about 94% were born before 1900.

The website compresses these 11.7 billion people into a grid of 100 tiny figures.

Switch to the "Era" column and you'll see that the vast majority of little people are clustered in the boxes before the Industrial Revolution.

As we move closer to modern times, it becomes sparser; when counting them up, only a few can stand at “after 1900.”

The vast majority of "you" are the ones bending over in the fields.

As you scroll further down the website, you’ll see “Ten Most Likely Lives,” with the top few almost entirely consisting of unnamed individuals from the agricultural era.

No name, no title, not even a single line in history, yet they occupy the largest shares of the entire prize pool.

All the emperors and generals you've memorized since childhood can't amount to one petty person.

After weighting, extreme poverty is the norm, literacy is the exception, and surviving to adulthood is itself a matter of probability.

Click to draw a life, and you've lived to 38.

Statistics alone aren't enough. The most incredible thing about this website is the button at the end labeled “Draw a Life.”

Click once, and it will truly randomly sample one outcome from the full probability distribution.

Era, region, social class, what kind of blessings one can enjoy, and the final outcome—all wrapped up in one.

Grab a random card, and it’s most likely to be this one.

Around 765 AD, a girl, a farmer or herder in sub-Saharan Africa. Living in extreme poverty, illiterate, with no electricity or running water.

Lived to about 38 years old.

Just these two short lines, slammed onto the screen.

You can keep pressing, pressing until that abstract "one in 11.7 billion" becomes a real, mortal person.

Click a few more times, and you'll find yourself falling back into roughly the same situation over and over.

If you don’t believe it, click the link at the end of the article, scroll to the “DRAW A LIFE” button, and see if you can draw a lucky sign for modern life.

After you've finished, welcome to leave a comment with the year, the region, and how old you lived to be.

In addition, the website features an interactive world map based on real geography, created using D3 and Natural Earth.

Drag the time slider to scan through each of the 12 eras.

For the most part, this center of gravity remained firmly anchored in Asia. It wasn’t until modern times that it began to shift toward Europe and the Americas.

In other words, the vast majority of chapters in the great book of humanity have been written on the rice paddies of Asia.

In fact, throughout all of human history, the number of people who have had access to clean drinking water, medical care when sick, and a high likelihood of their children growing up safely is barely enough to fill a handful.

Not to mention having electricity, living in the city, and enjoying a modern middle-income lifestyle.

What you have right now is a setup that the vast majority of 11.7 billion people can only dream of.

Ultimately, the way people live today, which seems ordinary to us, is an glaringly abnormal outlier when viewed against the entire timeline of human history.

"Reincarnation Simulator" — an AI portraying four roles

Delivering a product of this high quality is not as simple as generating a few lines of HTML and CSS.

If a traditional team were to handle this, data analysts would clean the population curve, historical researchers would dig through documents for evidence, UX designers would create wireframes, and frontend engineers would painstakingly code D3 line by line. Putting all these pieces together and refining them until launch would take weeks.

Switch to Opus 4.8, where a single AI effortlessly portrays four entirely different high-paying roles, smoothing out every rough edge along the way.

At the first level, it is a rigorous data researcher.

To calculate the number of 11.7 billion and its corresponding historical distribution, it first retrieved the 2022 report from the Population Reference Bureau titled "How Many People Have Ever Lived on Earth," then cross-referenced the Maddison Project Database, the HYDE dataset, and the United Nations World Population Prospects, ultimately synthesizing a matrix of population shares by era and region.

In this process, calculating the total number of people born in a certain region relies on a clean formula—

The total number of births in each era, multiplied by that era's share of the global population in that region, then summed cumulatively across generations.

The second layer consists of mathematical modeling experts.

The tricky part is that no organization has ever tracked the cumulative number of births by region.

It autonomously decided to use the Monte Carlo method to handle uncertainty, running 4,000 simulation rounds, with greater noise assigned to older and less certain time periods, ultimately outputting the median along with a 5% to 95% confidence interval.

Layer three, front-end architect.

Synchronizing smooth scroll-based storytelling with data states on a single page is no easy feat.

To achieve this, Opus designed its own modern aesthetic scrolling structure that fully resolves the rendering of massive data points in the browser and ensures dynamic chart responsiveness, with smooth, lag-free animations.

At the fourth level, it unleashed the mastery of top-tier copywriting and emotional direction.

First, it sets the stage with a grand historical sweep, then abruptly confronts the insignificance of individual fate. It uses visceral, bodily short sentences to carve cold statistics into the hearts of ordinary people.

Writing good code isn't remarkable. Knowing how to evoke human emotion is almost the last remaining territory of carbon-based life.

The mind that came up with this just topped the charts two days ago.

Back to Opus 4.8.

On the Artificial Analysis leaderboard, it surged to the top with a score of 61.4, overtaking GPT-5.5.

This is the first time Claude has reclaimed the top spot since OpenAI took the lead in April.

On Humanity's Last Exam, Opus 4.8 scored 45.7%, leading the second-place entry by a full 1 point.

On the前沿 physical benchmark CritPt, it also surpassed Gemini 3.1 Pro.

Programming is straightforward.

On the more challenging SWE-Bench Pro, it scored 69.2%, compared to GPT-5.5’s 58.6%, a clear ten-point gap.

In the specialized benchmark GDPval-AA that simulates real-world workplace tasks requiring hands-on work, research, and command-line operations.

Opus 4.8 achieved an Elo score of 1890, outpacing second-place GPT-5.5 by approximately 121 points, implying a win rate of about 67% in direct head-to-head matchups.

Moreover, it reduced the average number of interaction rounds needed to complete the same complex task to 38, eliminating the need for repeated human corrections by independently arriving at the optimal solution through a series of trials and reasoning.

This is precisely the underlying logic that enables it to deliver the entire website on its own.

By now in the final, it's about more than just the score.

Last month, GPT-5.5 dominated the AI throne, dethroning Opus 4.7. More than a month later, Opus 4.8 returned the favor.

Since this spring, this chair has changed hands several times, and no one has had time to sit in it long enough to get warm. The generational gap is now being compressed into weekly back-and-forth struggles.

But what's really changing is the criteria used to evaluate the rankings.

Once a model is treated as an independent agent, being smart isn't enough—it also needs to know its own boundaries.

On the same day Opus 4.8 was released, Anthropic also unveiled another move.

A Mythos-level model stronger than Opus will be available to everyone in a few weeks.

Intelligence and honesty have never been a choice between two options; Opus 4.8’s strength lies in pushing both forward together.

That website that drew global attention is merely a breath it casually exhaled.

References:

https://x.com/ArtificialAnlys/status/2060117582120976868

https://x.com/emollick/status/2060165879908749490

https://veil-of-history.netlify.app/#draw