Gary Marcus

Rank 13 of 47
|
Score 83
5/18/2025, 3:14:15 PM
X
In reply to:
There is no other room 🇬🇪
@uhbif19
·
250d
@GaryMarcus @davidmanheim @XRobservatory > i have documented a bunch of world model errors

I think that what we know for sure now, that funny examples of fails at obvious tasks do not predict anything. It was good hypothesis in like 2012, but now we know it does not work like that.
Gary Marcus
@GaryMarcus
·
250d
@uhbif19 @davidmanheim @XRobservatory 1. he wouldn’t pretend to know how to *draw* them, unless he could, whereas LLMs are systematically pretending to be able to draw things that they draw incorrectly. there is a pathology to that needs a name.

2. i have documented a bunch of world model errors (in ChatGPT’s
There is no other room 🇬🇪
@uhbif19
·
250d
@GaryMarcus @davidmanheim @XRobservatory All blind-from-birth people cannot recognize shapes (apart from their topological properties), even once they get vision. Lev Pontryagin is one of them, and he made important advancements in differential topology. Do you thing he did not understood what square is?
Gary Marcus
@GaryMarcus
·
250d
@davidmanheim @uhbif19 @XRobservatory So you are saying that the image generation is not (despite superficial appearances) properly tethered to a world model, correct?

Let’s start there.
David Manheim
@davidmanheim
·
250d
@GaryMarcus @uhbif19 @XRobservatory It's not a test of the world model because image generation is using diffusion to generate something in the latent space of images it was trained on, not something in the language model.

It's like saying hand-eye coordination is part of understanding; it's kind-of true, but, no.
Gary Marcus
@GaryMarcus
·
250d
@uhbif19 @davidmanheim @XRobservatory I documented a multiple similars one for ChatGPT (not o3) in text but curious why everyone thinks image generation is not a legitimate test of a systems “world model”

The statement engages in a discussion about the limitations of AI models, particularly in their ability to follow rules and perform tasks accurately, which is a public issue related to technology and its implications. The tone is critical and skeptical, questioning the reliability of AI in complex domains based on its performance in simpler tasks.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not directly harm others but questions the reliability of AI systems, which is a valid critique in public discourse.
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The statement does not promote understanding, empathy, or compassion, as it focuses on critiquing AI systems rather than fostering a deeper understanding of the issue. [-1]
  3. Principle 4:
    I will engage in constructive criticism and dialogue with those in disagreement and will not engage in personal attacks or ad hominem arguments.
    The statement engages in constructive criticism of AI systems, questioning their reliability without resorting to personal attacks or ad hominem arguments. [+1]