ChatGPT Company Takes Big Steps Towards AGI

In a blog post on December 20, OpenAI said that the o3 and o3 mini are currently undergoing internal safety testing, but have already achieved “groundbreaking high scores” on an AI reasoning test called the ARC Challenge. The model scored 87.5% on the ARC-AGI test, surpassing “human level” and also has a “private thought chain” feature that allows it to “think for itself” before providing an answer.

OpenAI claims the new model is “more powerful than the previous o1 model” and represents “a dramatic leap in performance.” CEO Sam Altman is more modest, saying it’s the start of “the next phase” of AI.

OpenAI o1 was launched in September. At the time, the company said the model was capable of reasoning through complex tasks and could solve more difficult problems in science, coding, and math than before. At the time,Business InsiderThe emergence of O1 has "made the line between human intelligence and artificial intelligence increasingly narrower," he said.

OpenAI is opening up an application process for outside researchers, allowing them to test the o3 model before it is released publicly. The program is expected to end on January 10, 2025.

According toReuters, the fact that OpenAI tested o3 stronger than o1 after only a few months shows that the company is accelerating the process of reaching superintelligence AGI - a breakthrough milestone in the field of artificial intelligence, with the prospect of AI being smarter than humans. Previously, o1 was considered by some experts to be able to approach AGI.

Following the announcement from the company behind ChatGPT, Francois Chollet, an expert on the ARC Challenge who recently left Google Research, called the new model “impressive and a major milestone on the road to AGI,” although there are still “a fair number” of simple tasks that o3 cannot solve. Chollet added that “these possibilities are in a new field that requires serious scientific attention.”

Arc Prize, a nonprofit initiative to advance AGI research, also sees o3 as a major step forward if it outperforms o1. "This could be a surprising and important step forward for AI, demonstrating an ability to adapt to new tasks that has never been seen before in GPT-family models," the organization wrote in a blog post.

Fortuneo3 is not yet AGI, but the new model will likely help ChatGPT handle more complex queries and solve problems step by step. "In particular, AI search could evolve with improved semantic understanding, as it looks for information not just based on keywords, but also on deep meanings and relationships between concepts. Models could also generate more coherent stories and essays, solving problems in real time," the site commented.