Artificial general intelligence (AGI) is the next level for artificial intelligence (AI) where it will exceed all human capabilities. Researchers at DeepMind say there are five levels of AGI and the first level has already been achieved by ChatGPT. But Yann LeCun, the head of AI at Meta issued a statement that the large language model (LLM) will not be able to reach the level of AGI.
In an interview with the Financial Times, LeCun said that LLM only has the intelligence of a cat. The disadvantage of LLM is that it is highly dependent on the data it is trained on. If the trained data has errors, the AI will simply regurgitate the wrong answer.
LeCun added, AI currently does not have the ability to think using logic which limits its ability to reach the AGI level. He is also less sure when AGI will be achieved and how it will not be done. For sure according to him, it cannot be through training using LLM alone.
The problem with AI using LLM has become increasingly apparent lately. Going viral on social media today, the answer was given by Google SGE who suggested adding glue to the cheese to make a thicker pizza. This nonsensical answer is given because Google uses Reddit data and on that site a joke about adding glue to cheese is indistinguishable from a serious answer.