Gary Marcus

Rank 10 of 47
|
Score 117
7/21/2025, 2:53:45 PM
X
In reply to:
Chris
@chatgpt21
·
1d
@GaryMarcus @krapstarr Well unfortunately with pre train scaling a 10x will yield 10-20% improvement. However 4.5 compared to 4 saw a pretty massive jump in benchmarks. Just not compared to 4o
Gary Marcus
@GaryMarcus
·
1d
@chatgpt21 @krapstarr that ain’t no whale
Chris
@chatgpt21
·
1d
@krapstarr @GaryMarcus 10-20%
Naveesh /wtf
@krapstarr
·
1d
@chatgpt21 @GaryMarcus how much better tho?
Chris
@chatgpt21
·
1d
@GaryMarcus No simple trick errors

Must have a non reasoning model that’s better than our current 4.5

And a reasoning end that’s better than o3 pro

Would like to see anything that indicates continuous learning but I know that’s far away. (Reasoning well outside of its data would be a
Gary Marcus
@GaryMarcus
·
1d
I love this.

What’s your list of “10 things GPT-5 must absolutely be able to do so [you] don’t end up coping later and calling it good when it’s actually mid”?

If there were no “memeably stupid” errors [#10] I will be shocked.

The statement critiques the promises made about AI advancements, suggesting initial claims were exaggerated. It engages in public discourse by addressing expectations versus reality in AI development.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement is critical but does not appear to cause harm. It challenges the initial promises made about AI advancements.
  2. 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 pointing out discrepancies between promises and outcomes without resorting to personal attacks. [+1]