Gary Marcus

Rank 19 of 47
|
Score 64
6/23/2025, 2:40:34 AM
X
In reply to:
Gary Marcus
@GaryMarcus
·
18d
@justanotherlaw yes 2, 3, 4 at once with research incentives for 4 (in addition to what you mention).
Lawrence Chan
@justanotherlaw
·
18d
@GaryMarcus Of course, these aren't mutually exclusive or jointly exhaustive -- you might hope for 2 3 4 at once (which is what I'd imagine you'd push for).
Lawrence Chan
@justanotherlaw
·
18d
@GaryMarcus What sort of engagement do you want here?

I agree with the four possible paths forward that you outlined. Paraphrasing a bit, to prevent catastrophic risk from AI, we can:
1. Hope for 'alignment-by-default', or at least that alignment magically becomes easy in the immediate
Gary Marcus
@GaryMarcus
·
19d
Would love to see the AI Safety community engage with this article, which raises challenges about the work that has been done so far, while suggesting that we might need to follow different avenues to make progress.
Saud Hashimi
@saudhashimi
·
19d
A super article by @GaryMarcus that attempted to answer the question of the AI epoch: how do we control behaviour to stop bad AI outcomes. Worth a read.

https://garymarcus.substack.com/…

The statement engages in public discourse by addressing the feasibility of achieving full mechanistic interpretability for large language models (LLMs), a topic relevant to AI safety and ethics. The tone is skeptical, suggesting that such interpretability is unrealistic given the current understanding of LLMs.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not directly cause harm but expresses skepticism about a technical goal, which could be seen as a cautionary stance.
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The statement does not promote understanding or empathy but rather challenges a technical possibility, which could stimulate further discussion and exploration.
  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 by questioning the feasibility of a goal, without resorting to personal attacks. [+1]