Watch the video linked below and then answer the questions.
Are Robots Actually Smarter Than Us? Take the AI Intelligence Quiz!
We’ve all seen the headlines claiming that Artificial Intelligence is on the verge of taking over the world. But how close is AI really to matching human-level intelligence?
In a fascinating video by Half as Interesting, we dive into the bizarre tests researchers use to measure artificial general intelligence (AGI). As it turns out, while AI can easily ace graduate-level exams, it still gets completely baffled by simple things a toddler can do.
Want to test your own knowledge of how AI “thinks”? Take our quick quiz below based on the video, and see how you stack up against the machines!
Part 1: The Multiple Choice Challenge
Test your understanding of the major benchmarks used to measure AI intelligence today.
1. What is “Humanity’s Last Exam”?
- A) A psychological exam given to AI developers to test their ethics.
- B) A test of 2,500 expert-level, graduate-school questions across dozens of subjects.
- C) A coding competition where AIs compete to build the best software.
- D) The final exam given to students at Harvard before graduation.
Click to reveal the answer
Correct Answer: B
The Details: “Humanity’s Last Exam” consists of 2,500 highly technical questions curated by experts. While top AI models originally scored around 8% on it, they have recently climbed into the 35% to 50% Range—which is far better than any single human could do across so many diverse fields!
2. Why did GPT-5.5 fail the “Dead Cat in a Box” question?
- A) It didn’t understand the chemistry of the poison.
- B) It was programmed to avoid answering questions about death.
- C) It relied on trained patterns from “Schrödinger’s Cat” and ignored the detail that the cat was already dead.
- D) The question was mathematically impossible to solve.
Click to reveal the answer
Correct Answer: C
The Details: AI models don’t read with the same flexibility we do. Because GPT-5.5 has processed the classic “Schrödinger’s Cat” thought experiment thousands of times, it recognized the pattern and completely ignored the word “dead” at the very beginning of the prompt.
3. What does the “ARC-AGI” exam measure?
- A) How fast an AI can scrape the Internet for Data.
- B) An AI’s ability to Count letters in Complex words.
- C) How well an AI can adapt to new, abstract logic puzzles using Basic Reasoning.
- D) An AI’s capability to generate hyper-realistic art.
Click to reveal the answer
Correct Answer: C
The Details: Unlike trivia tests, ARC-AGI uses simple grid-based puzzles with visual rules to see if an AI can grasp abstract, human-like logic (like understanding “above” vs. “below” or “open” vs. “closed” shapes).
Part 2: Quick-Fire True or False
Test your instincts on these quick statements. Tap below to see if they are fact or fiction!
4. AI models currently score 100% on “Humanity’s Last Exam.”
Click to reveal the answer
Answer: FALSE
While they are improving rapidly, they currently score in the 35% to 50% Range. Still impressive, but definitely not perfect!
5. No AI Model has successfully solved the ARC-AGI 2 closed-versus-open shape puzzles.
Click to reveal the answer
Answer: TRUE
Spatial concepts like “inside vs. outside” are incredibly difficult for models to grasp automatically without physical-world intuition.
6. In the ARC-AGI 3 “box game,” Claude successfully completed the majority of the levels.
Click to reveal the answer
Answer: FALSE
Claude only managed to beat a tiny 1.8% of the box game, and other models couldn’t even get it to run!
