OpenAI’s ChatGPT-o1 is considered by some experts to be a model that is close to AGI.
On December 5, OpenAI o1 was officially released after nearly three months of public testing. As part of the Shipmas campaign, OpenAI said the model is “capable of reasoning, designed to think more carefully before responding.”
Some experts have experienced and evaluated that o1 is gradually reaching the “realm” of AGI, meaning artificial intelligence can be “on par” with, or even surpass, human intelligence.
“Models like o1 show that humans often ignore near-AGI systems that are already capable of outperforming humans on most intellectual tasks, but are not autonomous or self-directed,” Wharton professor and AI expert Ethan Mollick wrote on the social network X.
Mollick has categorized the levels of AGI according to their development on One Useful Thing, with the highest level (level 1) being a system that can perform any task better than a human. Level 2, or “weak AGI,” is a machine that outperforms human experts in all tasks, although no such system currently exists. Level 3, or “Artificial Central Intelligence,” is AI that outperforms average human experts in tasks that require high intelligence. While level 4, “Joint Intelligence,” is the result of human-AI collaboration.
“In my opinion, we have achieved AGI. While not ‘better than any human at any task,’ AI is ‘better than most humans at most tasks,’” OpenAI technical team member Vahid Kazemi wrote on X on December 7.
Others, however, argue that AGI is a long way off. “The idea popularized by science fiction and Hollywood is that someone will discover the secret of AGI, or human-level AI, or AMI, whatever you call it. Turn on a machine, and voila, AGI. That’s not going to happen. The emergence of AGI is not a specific event, but a gradual process over time,” Yann LeCun, Meta’s chief AI officer, said in March.
