Gary Marcus

Rank 15 of 47
|
Score 76
6/15/2024, 3:07:01 PM
X
In reply to:
Jonathan Marcus
@fastkoder
·
517d
@GaryMarcus I disagree, but you could be right! The science is definitely far from settled, and that's why we are studying how these models perform cognition, from the bottom up.

I look forward to reading your thoughts on this and on our paper.
Gary Marcus
@GaryMarcus
·
517d
@fastkoder Don’t doubt this, but the system is still essentially computing something akin to correlations over an n-dimensional language space; don’t think they are real concepts. I am hoping to write about this (and Anthropic’s recent interpretability paper) over the next week or two.
Jonathan Marcus
@fastkoder
·
517d
@GaryMarcus That's a totally valid concern. We've also done causal experiments where we clamp features to high values and see the model behavior change predictably (Golden Gate Claude was one such example).
Gary Marcus
@GaryMarcus
·
518d
Correlation is not causation.
Tsarathustra
@tsarnick
·
518d
Jonathan Marcus of Anthropic says AI models are not just repeating words, they are discovering semantic connections between concepts in unexpected and mind-blowing ways

The statement by @fastkoder constitutes public discourse as it engages in a substantive discussion about the nature of AI models and their cognitive capabilities, which is a significant public issue in the field of technology and science.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does no harm with words and actions. It is respectful and acknowledges the value of the paper, even though there are disagreements. [+2]
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The statement promotes understanding and empathy by expressing interest in the paper and acknowledging its value, despite disagreements. [+2]
  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 and dialogue without resorting to personal attacks or ad hominem arguments. [+2]