Gary Marcus

Rank 11 of 47
|
Score 122
4/16/2025, 9:28:02 PM
X
In reply to:
Daniel Litt
@littmath
·
38d
@GaryMarcus @AvpElk I don't think this is the definition used by many people, e.g. by ARC-AGI. But FWIW I think o3 is (with notable exceptions) substantially better at math than people who never never studied math.
Gary Marcus
@GaryMarcus
·
38d
@littmath @AvpElk It should be able to do what *trained* humans can do in any domain. Not just beat untrained humans in a given domain - especially given training on the full web. (Amount of training is a relevant variable).

Last year, Elon: AI will be able to beat any human at any task
This
Daniel Litt
@littmath
·
38d
@AvpElk @GaryMarcus Right, it's kind of odd to give a definition of AGI that humans don't meet!
AVP
@AvpElk
·
38d
@GaryMarcus @littmath My definition of AGI has never required more than median human performance.

Probably less than that, really. Pretty much *all* humans have general intelligence.
Gary Marcus
@GaryMarcus
·
38d
@littmath I actually disagree with your reconstruction. That would be like saying a chess computer is good if it can beat the average high school student. None of the historical definitions of AGI ever invoked median performance, esp not of untrained people.

That’s a major lowering of
Daniel Litt
@littmath
·
38d
I don't think o3 is AGI, but FWIW I think a model could be AGI but not superhuman at math research. As I understand it, AGI means "as good as the median human at almost all intellectual tasks," and I don't think surpassing median human mathematical performance is a very high bar.

The statement is part of a broader discussion on the definition and evaluation of Artificial General Intelligence (AGI), engaging with public issues related to AI capabilities and benchmarks. The tone is analytical and critical, focusing on the appropriateness of current measures of AGI.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not cause harm and maintains a neutral tone, focusing on the critique of a concept rather than individuals. [+1]
  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 questioning the validity of a measure without resorting to personal attacks. [+1]
  3. Principle 7:
    I will uphold the principles of free speech and use my platform responsibly and with integrity.
    The statement contributes to the discourse on AGI responsibly, questioning existing measures and definitions, which is integral to the evolution of AI understanding. [+1]