Gary Marcus

Rank 14 of 47
|
Score 83
11/26/2025, 3:32:00 PM
X
In reply to:
David Manheim
@davidmanheim
·
3d
@GaryMarcus @jm_alexia I would be willing to bet that whatever model you eventually will agree achieves AGI uses an LLM-component that is at least as advanced as current frontier models, i.e. that the energy and time on LLMs so far was, in retrospect, necessary.
Gary Marcus
@GaryMarcus
·
3d
@davidmanheim @jm_alexia they (or more likely more efficient replacements) will play a role.

but spending so much money, time, and energy on them has quite likely delayed us.

we need new ideas, not slightly better LLMs.
David Manheim
@davidmanheim
·
3d
@GaryMarcus @jm_alexia Right, but if those hybrids end up leading to AGI, I think it would be a vindication of the claim that we needed the LLM progress to enable other eventually successful approaches.
Gary Marcus
@GaryMarcus
·
3d
@davidmanheim @jm_alexia um, that is the whole point of neurosymbolic hybrids
David Manheim
@davidmanheim
·
3d
@GaryMarcus @jm_alexia For instance, even if you are correct, as seems plausible, that neurosymbolic approaches are going to be what leads to real general intelligence, those will be used in concert with, or be used or developed by, the LLMs.
David Manheim
@davidmanheim
·
167d
@GaryMarcus @ChrisPainterYup And to the extent that solving with code is a victory for neurosymbolic AI, it seems like neurosymbolic AI is emergent from the methods you view as fundamentally limited. So rather than needing a new approach, it seems LLMs are getting there. But that's not your view, I believe.
David Manheim
@davidmanheim
·
3d
@GaryMarcus @jm_alexia I think the more likely future follows the prototype - that is, I expect that the boost from LLMs and their capabilities enables or is even necessary for the future progress, rather than distracting from those other approaches.

The statement is part of a broader discussion on the role of large language models (LLMs) in achieving artificial general intelligence (AGI). It engages with public discourse on AI development, efficiency, and the potential necessity of current models for future advancements.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not cause harm and is a technical opinion on AI models, thus adhering to the principle of doing no harm. [+1]
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The statement is neutral and does not actively promote understanding, empathy, or compassion, but it does contribute to a technical discussion.
  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 dialogue by offering a perspective on the efficiency of LLMs without resorting to personal attacks. [+1]