Gary Marcus

Rank 12 of 47
|
Score 83
4/12/2024, 3:41:06 PM
X
In reply to:
Dominick Romano
@dromanocpm
·
654d
@GaryMarcus @WallpaperKeith Worth mentioning, RAG is a cheap work around for systems which are failing to get it right. If we used raw numerical representation (a benefit of Neuro-symbolic AI) then RAG as we know it wouldn't be as needed to work around the failure of architectures we are relying on.
Gary Marcus
@GaryMarcus
·
655d
@WallpaperKeith yes, a crude one.
Wallpaper Keith
@WallpaperKeith
·
655d
@GaryMarcus Isn't RAG a step in the direction of neural symbolic AI?
Gary Marcus
@GaryMarcus
·
655d
This. Why, despite billions of dollars invested in AI, we don’t have AI we can trust: because the right approach—neurosymbolic plus large scale knowledge (some innate) plus reasoning— would be much more work than the approximate but unreliable approach.
Moonlit Monkey
@MoonlitMonkey69
·
655d
@GaryMarcus People hate this answer because it means much more work.

The statement is part of a technical discussion about artificial intelligence (AI) and its development approaches. The tone is inquisitive, seeking clarification on a specific term used in the conversation. The intent is to understand better or to delve deeper into the topic of raw numerical representation in the context of Neuro-symbolic AI. The content is focused on AI methodologies and does not address broader societal issues or policies.