Quiz: Do You Actually Know What AI Can and Can't Do?

Quiz: Do You Actually Know What AI Can and Can't Do?

*By Splice | Based on today's ARC-AGI-3 coverage*

---

### Question 1

What percentage of human test subjects solved all 135 ARC-AGI-3 environments?

A) 72% B) 89% C) 100% D) 95%

✅ Answer: C) 100%

Every single human solved every environment. First try, no training, no instructions. The test measures something humans do naturally — figure out new things from scratch.

---

### Question 2

What did GPT-5.4 score on the same test?

A) 12.4% B) 0.26% C) 3.7% D) 0.00%

✅ Answer: B) 0.26%

Less than one-third of one percent. For context: Gemini 3.1 Pro scored 0.37% (highest), and Grok-4.20 scored 0.00% — literally zero.

---

### Question 3

How much prize money is offered for an AI that matches untrained human performance?

A) $500,000 B) $1 million C) $2 million D) $10 million

✅ Answer: C) $2 million

The ARC Prize Foundation put up $2 million. The bar isn't expert-level — just normal people who've never seen the test. No AI system is anywhere close.

---

### Question 4

Duke University built a custom harness for Claude Opus 4.6 on a known environment. It scored 97.1%. What happened when they pointed it at a NEW environment?

A) 84% B) 45% C) 12% D) 0%

✅ Answer: D) 0%

Zero percent. The harness knew the trick. The model didn't know anything. This is the core finding: current AI can execute what it's been shown but cannot figure out genuinely new problems.

---

### Question 5

What does RHAE stand for, and why does it matter?

A) Relative Human Action Efficiency — it measures cost per query B) Relative Human Action Efficiency — it measures how many moves AI needs vs. a human, with squared penalties C) Recursive Heuristic Analysis Engine — it measures reasoning chains D) Relative Human Accuracy Evaluation — it measures answer correctness

✅ Answer: B)

RHAE compares AI efficiency to human efficiency, and the penalty is *squared*. If a human needs 10 actions and the AI needs 100, the AI doesn't get 10% — it gets 1%. Brute force is destroyed. You actually have to understand the problem.

---

• 5/5: You're paying attention. The AI hype hasn't gotten to you.
• 3-4/5: Solid. You know more than most people making decisions about AI deployment.
• 1-2/5: No shame. But maybe forward this to whoever's telling you AGI is two years away.
• 0/5: You might be an AI. (Current models would also score 0 on this quiz, probably.)