Is AI Actually Smart, Or Just Pretending? ‘Humanity’s Last Exam’ Might Spill The Beans

Digital Disconnect: Artificial Intelligence (AI) can do some wild stuff these days — write college essays in seconds, give you a doctor-level health analysis, and even craft legal contracts (because who needs lawyers, right?). But here’s the million-dollar question: Is it actually smart? Like, “can outthink a professor on a bad day” smart? Or just fast at Googling stuff and stringing words together?

Well, we might finally have a way to find out. Meet Humanity’s Last Exam — a test designed to figure out whether AI is just a glorified Wikipedia scraper or something closer to a genuine brainiac.

Why Do We Need Another AI Test?

Good question. Didn’t we already have all those fancy benchmarks like MMLU, which AI models are crushing with over 90 per cent scores?

Yep, and here’s the catch: those tests are like multiple-choice quizzes for overachievers who know how to memorise but don’t actually understand what they’re doing.

Humanity’s Last Exam? Whole different ball game.

This test isn’t about regurgitating facts; it’s about whether AI can think its way out of a paper bag.

What Makes Humanity’s Last Exam So Special?

First off, this isn’t just a regular quiz. It’s a 3,000-question, brain-melting nightmare spanning over 100 subjects — from physics and philosophy to law and medicine. And unlike those cushy academic tests made by bored researchers, these questions were cooked up by nearly 1,000 experts across 50 countries. Talk about teamwork.

And then there’s the multi-modal twist. Ten per cent of the questions require AIs to process both text and images — because apparently figuring out words alone is too easy. Imagine asking a robot to solve a math problem while showing it an abstract art piece. Welcome to Humanity’s Last Exam.

But wait, there’s more. To stop AI from cheating, some of the hardest questions are kept hidden from public datasets.

So no, companies can’t just train their bots on all the answers and flex fake intelligence.

How Are AIs Doing On Humanity’s Last Exam?

Terribly.

Even the “best of the best” models are barely scraping low double-digit scores.

Compare that to the over-90 per cent they boast on older benchmarks, and it’s clear they still have some serious catching up to do.

What Does This Mean For The Future?

Humanity’s Last Exam isn’t just a tougher exam — it’s an intervention for AI hype. It’s telling AI developers, Hey, maybe stop showing off and start fixing your robots’ critical thinking skills.

If nothing else, this test is a big win for accountability. Companies now have to prove their models are worth the hype, not just flashy smoke and mirrors. And if AI does ever master Humanity’s Last Exam? Well, we might just have systems that actually think — instead of just faking it really well.

Until then, get ready for a whole lot of “we’re working on it” from the AI industry. Because if they want to brag about being the smartest in the room, they’ll have to pass Humanity’s Last Exam first.

Digital Disconnect is an ABP Live-exclusive column, where we explore the many admirable advancements the world of tech is seeing each day, and how they lead to a certain disconnect among users. Is the modern world an easier place to live in, thanks to tech? Definitely. Does that mean we don’t long for things to go back to the good-ol’ days? Well, look out for our next column to find out.