An important day to remember
Dec 20, 2024. An important day to remember. It might just be the dawn of a new era where artificial intelligence begins to accelerate and could one day surpass human intelligence faster than we can imagine.
The ARC-AGI (Abstract and Reasoning Corpus for Artificial General Intelligence) benchmark was designed to prevent large language models (LLMs) from exploiting brute-force memorization and instead test their ability to adapt, learn, and solve novel problems, like how human intelligence works.
On Dec 5th, 2024, OpenAI released its full o1 model*, which scored only 32% on this test. Just 15 days later, on Dec 20th, its successor, the o3 model, achieved a remarkable 88%—a performance comparable to what a little child can attain if given the same test. ( * The o1-preview was released on Sept 12th, 2024. )
One detail shouldn’t be overlooked: the cost of achieving such results. Each task comes with a computational price tag of about $3,400. Yup, over three grand for one concentrated "thought"—the kind where you furrow your brows in deep problem-solving mode. 🤔
Now consider the cost of teaching young children how to think and reason. We may not be able to measure the cost in dollar amounts, but surely parents and teachers rack up plenty of grey hairs in the process. 👩🦳🧓😂
P.S. Test your human intelligence by solving these puzzles here: