Gary Marcus

Rank 14 of 47
|
Score 83
11/19/2025, 3:48:53 PM
X
In reply to:
Amrith Krishna
@krishnamrith12
·
22d
@GaryMarcus @ylecun A similar argument can be made regarding your idea for world models, which is more of symbolic and causal in nature to what I understand. While I am a fan of such approaches and am currently incorporating those in my own appraoches, it need not be the way Lecun looks at it (3/4)
Amrith Krishna
@krishnamrith12
·
22d
@GaryMarcus @ylecun Probabilistic spaces are a special case in EBMs & are fundamentally coming from a physicist's world view. However, schmidhuber's world models still are in a probabilistic space to the best of my understanding. There are other key differences. Hwoever, this was one I noticed (2/4)
Amrith Krishna
@krishnamrith12
·
22d
@GaryMarcus @ylecun My limited point here, countering your argument, would be that world models as put forward by schmidhuber is fundamentally different from that by @ylecun.

Energy based models forms the core of @ylecun's approach to world models, which generalises to non probabilistic spaces ...
Gary Marcus
@GaryMarcus
·
22d
@krishnamrith12 @ylecun schmidhuber was mention models of the world in 1990 in neural networks, as I mentioned in my review. did LeCun happen to acknowledge that?
Amrith Krishna
@krishnamrith12
·
22d
@GaryMarcus @ylecun Low information density addresses pts. 2 & 3 in your article, and this talk was made on 2016 much before your 2019 "tweet" @GaryMarcus
Exponential divergence addresses points 4 and 5, and in fact @ylecun explcitly mentions "models of world" in the talk.

https://www.youtube.com/watch?v=…
Amrith Krishna
@krishnamrith12
·
22d
@GaryMarcus @ylecun I am not sure what exactly is empty in the tweet. For instance his 2016 keynote talks about fundamental limitations of using autoregressive models, where he talks of low information density and exponential divergence

The statement is part of a technical discussion on world models in AI, involving different approaches by notable figures in the field. It reflects a professional disagreement and a defense of one's own perspective.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not cause harm and maintains a professional tone, respecting differing viewpoints. [+1]
  2. 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 dialogue by acknowledging differences in approach without resorting to personal attacks. [+1]