Gary Marcus

Rank 11 of 47
|
Score 83
9/3/2024, 3:06:02 PM
X
In reply to:
Riley Goodside
@goodside
·
518d
@paul_cal It starts messing up somewhere around 7 or 8 really — inherent counting issues still apply here; still much better than the text case though
Paul Calcraft
@paul_cal
·
518d
@goodside How many different counts have you tried btw?

If reliable, would be interesting to see how high it can count in an image. It's obviously an entirely different paradigm to counting tokens, so might not follow this... (& perhaps even points to a solution)
Mor Geva
@megamor2
·
520d
@goodside Transformers have an inherent limitation in solving this task: https://arxiv.org/abs/2407.15160
Riley Goodside
@goodside
·
518d
Weirdly this is an easier problem

The statement is part of a technical discussion about the limitations of transformer models in counting tasks. It is a continuation of a conversation exploring the counting capabilities of AI models, specifically in the context of counting letters and objects in images. The tone is inquisitive and collaborative, aiming to further understand the behavior of AI models in specific scenarios.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does no harm and is focused on a technical inquiry, adhering to the principle of striving to do no harm with words and actions. [+1]
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The statement promotes understanding and empathy by engaging in a collaborative discussion to explore and understand the limitations of AI models. [+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 without personal attacks, focusing on the technical aspects of the discussion. [+1]