AGI Debate Ignited: Are Today's LLMs Already Here?

Groundbreaking Claim in AGI Debates: Have LLMs Passed the Fundamental Tests?

The most ambitious goal of the artificial intelligence world, Artificial General Intelligence (AGI), aims to develop systems with human-like learning, comprehension, and reasoning capacities. The year 2024 was shaken by an unexpected claim in this field. Four academics from the University of California, San Diego, argued in a commentary published in the prestigious science journal Nature that current Large Language Models (LLMs) meet the basic criteria for AGI and demonstrate human-level performance. This statement created deep reverberations in technology and academic circles, bringing the question "Has AGI arrived?" back to the center of the agenda.

Academics' Arguments and Criticisms

According to the academics' thesis, next-generation LLMs like GPT-4, Claude, and similar models have reached a level where they can compete with humans in areas such as complex language understanding, logical inference, code writing, and multi-step problem solving. These models' ability to perform a wide range of tasks reflecting the versatility of human intelligence suggests they are not merely systems specialized in a narrow field, but rather show the first signs of a general intelligence capacity. However, this view has not gained broad acceptance in the scientific community. Many experts emphasize that the mechanisms underlying LLMs' impressive language skills are fundamentally different from how humans learn, establish causality, and experience the physical world.

Critics highlight that these models are based on statistical correlations rather than genuine understanding and still have serious limitations in areas like creativity, common sense, and social intelligence. Another important criticism is that LLMs lack independent agency or goal-setting ability. That is, they cannot set their own objectives and plan accordingly like a human. This situation keeps them positioned as systems that are still "standing with human support."

Technical and Social Barriers on the Path to AGI

Experts agree that there are still significant barriers to overcome to achieve truly meaningful AGI beyond current LLMs. From a technical perspective, radical progress is needed in areas such as causal reasoning, long-term planning, interaction with the physical world, and learning from scratch (few-shot/zero-shot learning). As noted in web sources, although current AI systems show superior performance in certain tests, this does not make them AGI; they are more accurately described as "narrow geniuses."

The social and ethical dimension is another critical area. It is predicted that AGI could potentially permeate all engineering and technology fields by the 2030s and enable the creation of advanced robotic devices. However, control of this power, its impacts on the labor market, the establishment of security protocols, and ethical frameworks emerge as discussion topics as important as technical developments. Although AGI is seen as a "Holy Grail" in humanity's technological evolution, the path to this goal is filled with both hope and risks.

Conclusion: An Ongoing Journey

The claim by University of California academics has become an important spark that reveals the rapid progress in the artificial intelligence field and pushes us to reconsider standards. The achievements of LLMs can be considered a significant milestone on the path to AGI. However, fully capturing the profound and multidimensional nature of human intelligence will likely require new architectures beyond language models, perhaps hybrid systems. As of 2024, while it may be possible to say we have opened the door to AGI, the question of whether we have stepped through that door appears likely to continue being shaped by intensive research and philosophical debates. Although technological breakthroughs are exciting, the path to AGI is both technical...