Gary Marcus

Rank 13 of 47
|
Score 83
12/23/2024, 4:26:02 PM
X
In reply to:
Jon Barron
@jon_barron
·
397d
@blurrylogic @shlomifruchter Flood fill is like 20% of it. Morphological operators would get you a long way. Even some sort of cellular automaton thing should be helpful. Throw them all in and brute force it, it'll be cheaper than o3.
Blurry Logic
@blurrylogic
·
397d
@jon_barron @shlomifruchter hmm none of the kaggle top teams tried toolformer

but what image procssing algos would solve these ?????( take a look at other ques aswel)
Jon Barron
@jon_barron
·
397d
@shlomifruchter Have the LLM turn the input text into a 2D image, then do a breadth first search where the successor function is picking a 1980s image processing algo. Train the LLM on those searches. Does this make sense? Seems like it should be reasonable
Shlomi Fruchter
@shlomifruchter
·
397d
@jon_barron ARC prize is still ongoing if you have ideas how to beat it :)
Jon Barron
@jon_barron
·
397d
I don't follow LLM benchmarks so pardon my ignorance, but can't you just give the LLM access to old-school 1980s image processing algos via tool use? Seems like a lot of these could be brute forced by chaining Bresenham line drawing with flood fills or morphological operators.

The statement critiques the effectiveness of benchmarks in evaluating AGI solutions, suggesting that they may not capture the desired outcomes. It implies skepticism about the depth of understanding gained from such tests.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not cause harm and is a neutral critique of benchmarks.
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    It promotes understanding by questioning the effectiveness of benchmarks, encouraging deeper consideration of AGI evaluation methods. [+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 by providing a thoughtful critique without personal attacks. [+1]