Gary Marcus

Rank 13 of 47
|
Score 84
6/27/2025, 4:22:39 PM
X
In reply to:
David Manheim
@davidmanheim
·
228d
@GaryMarcus @WesRothMoney On #2, I think the bigger crux is that I don't think that symbols will help, we still don't have any idea how to do safety that is robust to human level, much less smarter than human, AI systems.
Gary Marcus
@GaryMarcus
·
228d
@davidmanheim @WesRothMoney 1. the actual work quoted above is exactly showing LLMs NOT developing symbolic understanding, but instead leveraging it outside tool. that’s STRONG in my direction and strongly away from yours.

2. crux is that we need symbols in the loop if we are to have any hope at safety, as
David Manheim
@davidmanheim
·
228d
@GaryMarcus @WesRothMoney I have claimed it was plausible that LLMs would develop that symbolic understanding without the tools, and this is slight further evidence in your direction, but either way it's not a crux on whether we need to focus on safety, and ensure we do not build dangerous AI.
Gary Marcus
@GaryMarcus
·
228d
@davidmanheim @WesRothMoney if you use an LLM plus
symbolic tools that vindicates all the arguments i have been making for 30 years that symbols and neural nets both need to be in the loop.

period.
David Manheim
@davidmanheim
·
228d
@GaryMarcus @WesRothMoney I don't understand why this favors neurosymbolic approaches.
I would argue that it is showing that those approaches are unnecessary because the non-neurosymbolic (but symbolic) tools can be used by non-neurosymbolic agent-based systems.
Gary Marcus
@GaryMarcus
·
228d
@WesRothMoney or: the models on their own can’t reason. they need external tools, typically built with (often derided) classic symbolic techniques, in order to reason reliably.

huge vindication for those of favoring neurosymbolic approaches.

The statement is part of a complex discussion on AI safety and the role of symbolic laws in alignment, engaging with public concerns about AI development and safety.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not use harmful language and maintains a neutral tone, 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 contributes to understanding by engaging in a technical discussion about AI safety, promoting a deeper exploration of the topic. [+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 dialogue, responding to previous points without resorting to personal attacks or ad hominem arguments. [+1]
  4. Principle 7:
    I will uphold the principles of free speech and use my platform responsibly and with integrity.
    The statement upholds free speech principles by contributing responsibly to a public debate on AI safety. [+1]