AI chatbots like ChatGPT will never be as intelligent as humans ...
Meta’s chief AI scientist Yann LeCun has expressed his views on large language models (LLMs), which are the driving force behind AI chatbots like ChatGPT and Gemini. According to LeCun, these models will never be able to match human intelligence in reasoning and planning.
Limited Understanding of Logic
In an interview with the Financial Times, LeCun mentioned that large language models currently lack a profound understanding of logic. He emphasized that they do not possess a grasp of the physical world, lack persistent memory, cannot engage in reasoning in any meaningful sense, or plan hierarchically.
Challenges with Accuracy
LeCun further stated that large language models like ChatGPT are considered "intrinsically unsafe" because their ability to provide correct responses is heavily reliant on the accuracy of the data they have been trained on. He highlighted that these models essentially utilize accumulated knowledge from extensive training data rather than true reasoning.
The Quest for Human-Level Intelligence
When discussing the path to achieving human-level intelligence in AI, LeCun mentioned Meta’s Fundamental AI Research (Fair) lab, which is diligently working on a new AI system with the goal of developing common sense and understanding the functioning of the world around us. This approach, known as ‘world modelling’, poses potential risks for Meta as investors typically seek quick returns on AI investments.
Striving for Artificial General Intelligence (AGI)
LeCun believes that the pursuit of Artificial General Intelligence (AGI) is not merely a design or technological challenge but a scientific one. This highlights the complexity and depth of the advancements needed to reach a level of intelligence comparable to human reasoning and planning capabilities.
In conclusion, while AI chatbots like ChatGPT and Gemini have their inherent limitations in terms of cognitive abilities compared to humans, they continue to be valuable tools in various applications.
