AI Expert Argues Large Language Models Obstruct Path to True Artificial General Intelligence

François Chollet, co-founder of the ARC Prize, argues that the AI industry's heavy investment in large language models represents a fundamental misdirection that has delayed genuine progress toward artificial general intelligence by as much as a decade. According to Chollet, while LLMs can become more powerful and useful for business applications, they are architecturally incapable of achieving human-level reasoning and adaptability required for true AGI. His critique suggests that funding flowing toward LLM development comes at the expense of exploring alternative approaches that might actually lead to achieving AGI.

François Chollet established the ARC Prize as a standardized measurement tool to evaluate genuine progress toward artificial general intelligence, recognizing that existing AI benchmarks may not accurately capture what true AGI requires. His central argument distinguishes between incremental improvements in large language model capabilities—which he acknowledges can yield practical business benefits—and architectural advancement toward systems capable of human-level reasoning and learning from novel situations outside their training parameters.

Chollet's critique addresses a resource allocation problem within the AI industry, suggesting that substantial investment concentration in large language model development may be crowding out funding for alternative research approaches that could more directly address AGI's fundamental challenges. His position reflects concern that the current industry trajectory, while producing commercially valuable tools, represents a strategic misdirection that could extend the timeline to achieving genuine artificial general intelligence.