Gary Marcus

Rank 41 of 47

Score -124

@thegartsy Even if they did (there are some papers like this), it is likely to developers will teach to the test, and the scenarios that really matter aren’t likely to be in the training set, and these systems generally suck at outliers.

6/25/2024, 5:20:03 PM

In reply to:

Tess Hegarty

@thegartsy

130d

@GaryMarcus Has anyone seen a benchmark for the ability of an AI model to do anything along the lines of “moral reasoning”?

I attempted to use ChatGPT to help with a “moral reasoning” type of task & it required extensive coaching & guidance from me, the human.

Not promising.

Gary Marcus

@GaryMarcus

130d

Why do benchmarks seems so terribly broken?

Is it because developers are teaching to the test? Is there another explanation?

Taelin

@VictorTaelin

131d

honestly, lmsys lost a lot of its meaning to me, after GPT-4o remained several points above both GPT-4-Turbo AND Opus for so long. I don't get it. it is just so glaringly disconnected from actual performance. over a day of work, I could collect dozens of examples where GPT-4o…

The statement engages in a discussion about the limitations of AI models, particularly in the context of moral reasoning and benchmark testing. It critiques the tendency of developers to 'teach to the test,' which may not adequately prepare AI systems for real-world scenarios, especially those involving outliers. The tone is critical but constructive, aiming to highlight a significant issue in AI development and evaluation.

Principle 1:
I will strive to do no harm with my words and actions.
The statement strives to do no harm and focuses on a constructive critique of AI development practices. It does not engage in personal attacks or harmful rhetoric. [+1]
Principle 3:
I will use my words and actions to promote understanding, empathy, and compassion.
The statement promotes understanding by explaining a potential flaw in current AI training and evaluation methods. It encourages a deeper consideration of how AI systems are developed and tested. [+1]
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, addressing a specific issue in AI development without resorting to personal attacks or ad hominem arguments. [+1]