Google Dipmand CEO Demis Hasbis considers a rece confusion about achieving doctoral artificial ielligence as “meaningless” and says modern systems lack importa capabilities such as “coinuous learning”. He also predicted that we are still far from 5 to 5 years with the real AGI.
“You often hear from some of our competitors who refer to modern modern systems as” doctoral ielligence “, according to Windows Ceral, in a new speech, referring to Openai’s claims indirectly. I think this is meaningless. “
He explained that real ielligence should be able to operate at all doctoral levels, while today’s models are not. They may shine in a sophisticated job at a mome, but at the next mome they fail in a simple math issue. In his view, such systems should not be iegrated to be AGI, because real ielligence should not be made such a misguided mistake.
New statemes by Google Dipmind CEO about artificial ielligence
Demis Hasbis poied to the vulnerability of chats against “Pram Engineering”; The fact that the answer of a model strongly depends on how the question is asking, not the advanced ielligence. This is reminisce of users’ complais of Microsoft’s coupolot, which led to the launch of “Copilot Academy” to teach artificial ielligence.
One of the most importa compones in today’s systems is the capability of “coinuous learning”. This feature allows artificial ielligence systems to be coinuously smarter by learning through new online coe, modifying their knowledge and correcting their behaviors. Without this ability, the chattes stay in the same limited knowledge they had during their training and cannot learn something new.
Of course, the confusion about the definition of AGI has made the statemes of the specialists coradictory. The classic definition of comprehensive artificial ielligence is that its cognitive abilities are level or beyond human; However, in the Microsoft Partnership and Openai coract, the term refers to a $ 2 billion profit system.
