Gary Marcus

Rank 12 of 47
|
Score 83
4/12/2024, 2:52:03 PM
X
In reply to:
Roland Fernandez
@rolandalong
·
656d
@GaryMarcus By neurosymbolic, are you envisioning a small set of rules that guide a neural model, or is the interaction more deep/complex?
Gary Marcus
@GaryMarcus
·
656d
@rolandalong when i have free time i will explain why the bitter lesson has been misleading
Roland Fernandez
@rolandalong
·
656d
@GaryMarcus Beware- the bitter lesson lies in waiting!
Gary Marcus
@GaryMarcus
·
656d
The only plausible approach to trustworthy AI I know:
• Neurosymbolic hybrid AI
• Large-scale knowledge, with an innate core
• Better reasoning
• Explicit, dynamically updatable cognitive models
Vishnu
@vishivishx
·
656d
@GaryMarcus .@GaryMarcus what do you think is the most promising alternative approach that will allow AI to scale to true intelligence, instead of this Transformer-based Autoregressive Next-token prediction, which seems more akin to just curve-fitting over all the text on the internet?

The statement is part of a technical discussion on artificial intelligence (AI) approaches, specifically neurosymbolic hybrid AI. It is a response to a question about the complexity of the interaction between rules and neural models in AI systems. The content is focused on sharing academic references that provide a deeper understanding of the topic.