Human-Level AI, OpenAI o3 Ignites AGI Debate

- OpenAI o3 Model Scores 87.5% on Human-Level AI Benchmark - Controversy Over Achievement of Artificial General Intelligence (AGI) [Unblock Media] OpenAI's latest AI model, the o3 model, has sparked intense controversy over artificial general intelligence (AGI) after scoring an unprecedented 87.5% on the "Thinking Like a Human" benchmark. This score was obtained from the Autonomous Research Collaborative Artificial General Intelligence (ARC-AGI) benchmark and is considered to be nearing human-level performance. San Francisco-based AI research company OpenAI announced the o3 and o3-mini models as part of the "12 days of OpenAI" campaign, suggesting higher goals have been achieved in AI following Google's release of their competing o1 model. Unlike other large language models that rely on pattern matching, the o3 model is designed to use a "program synthesis" approach, generating and applying new algorithms to solve problems. ARC Prize co-founder François Chollet commented in a blog post that "o3 is a system capable of adapting to tasks it has not encountered before, approaching human-level performance in the ARC-AGI benchmark domain." The ARC Prize reported that the average human performance score ranged from 73.3% to 77.2%. However, Chollet stated that "passing the ARC-AGI is not the same as achieving AGI, and personally, I do not yet consider o3 to be AGI." He added that the new ARC-AGI-2 benchmark addresses existing limitations and could potentially lower o3 model's performance to below 30%. With the introduction of new tests where humans can achieve over 95% scores without training, it is anticipated that the model's performance will be evaluated more accurately. Some experts have challenged whether the ARC-AGI benchmark test itself is the best indicator of whether the model truly approaches human-level problem-solving ability, as it may not fully reflect the model's genuine reasoning capabilities. Meanwhile, OpenAI researcher Vahidi Kazemi claims "this is AGI" and states, "In my opinion, we have already achieved AGI." This has sparked active discussions about what constitutes the criteria for AGI. However, OpenAI CEO Sam Altman has not taken a definitive stance on the achievement of AGI, instead describing o3 as a "very smart model." He suggested that intelligence alone may not be a sufficient condition for AGI and emphasized the need to focus on the next stages of AI development.