Meta AI chief says ChatGPT will never reach human intelligence...
Meta’s top AI scientist discussed the limitations of large language models like ChatGPT in achieving human-level intelligence. According to Yann LeCun, these models lack a deep understanding of logic and various essential cognitive functions. He emphasized that such models do not comprehend the physical world, lack continuous memory, struggle with reasoning, and are unable to plan hierarchically.
LeCun pointed out that relying on Large Language Models (LLMs) to attain human-like intelligence is flawed due to their dependency on specific training data to respond accurately. This reliance on training data makes them inherently unsafe for broader applications.
New AI systems with human-level intelligence
Despite these limitations, LeCun is actively working on developing a new generation of AI systems that aim to imbue machines with intelligence comparable to humans. However, he estimates that achieving this goal may take up to a decade.
This revelation coincides with Meta's recent challenges, as the company's value plummeted by nearly $200 billion following Mark Zuckerberg's announcement of increased investments to position Meta as a leading AI corporation globally.
Advancements in the AI industry
Meanwhile, AI company Scale successfully secured $1 billion in a Series F funding round, valuing the startup at approximately $14 billion. Additionally, French startup "H" disclosed a funding raise of $220 million, reflecting the ongoing investments and innovations in the AI sector.
Despite these advancements, experts raise doubts about AI's capacity to replicate human thought processes. Akli Adjaoute, an AI veteran with three decades of experience, highlighted the fundamental disparity between AI's pattern recognition capabilities and human cognitive understanding.
Adjaoute explained that while AI excels at recognizing patterns based on algorithms, human cognition involves a more nuanced comprehension of objects and concepts. Unlike AI, humans effortlessly perceive entities like numbers without relying solely on predefined algorithms or patterns.
