Wed Dec 25 17:17:01 UTC 2024: ## OpenAI’s New AI Model Achieves Human-Level Performance on General Intelligence Test

**Canberra, Australia/Stanford, CA** – A groundbreaking new artificial intelligence (AI) model, OpenAI’s o3, has achieved a remarkable milestone, scoring 85% on the ARC-AGI benchmark – a test designed to measure general intelligence. This surpasses the previous best AI score of 55% and matches the average human score.

The achievement, announced on December 20th, has sent ripples through the AI research community. The ARC-AGI benchmark assesses an AI’s “sample efficiency”—its ability to learn from a small number of examples and apply that knowledge to new, unseen problems. This capacity for generalization is considered crucial for true artificial general intelligence (AGI).

Unlike previous models like ChatGPT, which rely on massive datasets, o3 demonstrates significantly improved sample efficiency. The test involves solving grid-based pattern recognition problems, requiring the AI to identify rules based on only three examples before applying them to a fourth. Researchers believe this success stems from o3’s ability to search through different “chains of thought” to solve problems, similar to the approach used by Google’s AlphaGo.

While OpenAI has released limited information about o3’s inner workings, researchers like Francois Chollet, creator of the benchmark, suggest the model’s success hinges on its capacity to identify “weakest rules”—the simplest, most generalizable rules that solve the problem.

The implications of this breakthrough are significant. If o3’s capabilities are truly representative of a leap towards AGI, it could revolutionize numerous sectors, ushering in an era of self-improving AI. However, experts caution that further evaluation is needed to fully understand o3’s capabilities and limitations. A more comprehensive understanding of its success rate and failure modes is crucial before drawing definitive conclusions. The development also raises important questions about the governance and ethical implications of increasingly sophisticated AI systems.

Read More