Can ChatGPT match human intelligence?

WEB DESK: In a recent interview with the Financial Times, Yann LeCun, the head of artificial intelligence at Meta, expressed skepticism regarding the capabilities of large language models (LLMs) to attain human levels of planning and reasoning.

LLMs, which power popular generative AI tools such as OpenAI’s ChatGPT, Meta’s Llama, and Google’s Gemini, are trained using extensive datasets. However, LeCun highlighted their inherent limitations, emphasizing that their accuracy in responding to prompts is contingent upon the specificity of the training data provided.

According to LeCun, LLMs possess a “limited understanding of logic,” lack persistent memory, and struggle to comprehend the physical world. Furthermore, they are unable to engage in hierarchical planning or reasoning, as per LeCun’s assessment, rendering them incapable of reaching human-like cognitive abilities.

Attributing their shortcomings to their reliance on narrowly tailored training data, LeCun cautioned that LLMs are inherently “unsafe” and advocated for exploration into alternative AI models for achieving human-level intelligence.

As one of the influential figures in AI, often regarded as one of the “AI godfathers” for his foundational contributions to the field, LeCun outlined Meta’s endeavor to pioneer a new generation of AI systems through a paradigm known as “world modeling.”

Read More: Vanishing stars: new study uncovers mystery of missing massive stars

The world modeling approach, as described by LeCun, involves the development of AI systems that emulate human cognition by building an understanding of the surrounding environment and predicting potential outcomes based on changes within it. However, LeCun cautioned that achieving human-level AI using this approach could take up to a decade.

With LeCun and his team of approximately 500 researchers at Meta’s Fundamental AI Research lab spearheading efforts in this direction, the quest for human-like artificial intelligence takes a new trajectory, challenging the conventional reliance on LLMs and propelling the exploration of innovative AI methodologies.