In other news justifying neurosymbolic AI and global models, after Strong statements from Demis Hassabis yesterdayYann LeCun, historically hostile to symbolic approaches, has just joined what looks like a neurosymbolic AI company focused on reasoning and models of the world, apparently built on roughly the same type of plan as presented in 2020. Here’s the announcement:
And my 2020 project that the company mission above reminds me intensely:
where the hybrid was cashed as a hybrid of neural networks on logic/symbolic based approaches.
Given the grief LeCun has caused me over the years, this is a stunning development, and yet another sign that Silicon Valley is desperate for alternatives to pure LLMs – and is finally open to a reorientation around the mix of neurosymbolic AI, reasoning, and world models that researchers like me have long recommended.
It is also further vindication for Judea Pearl and his tireless promotion of causal reasoning. Perhaps this would be a good day to reread his Book of Whyand mine Restart AI (with Ernest Davis), both of whom anticipated aspects of the present moment years in advance.
