Gary Marcus

Rank 17 of 47
|
Score 65
9/13/2024, 10:40:31 PM
X
In reply to:
Cameron Williams
@Wasgo
·
345d
@mealreplacer @GaryMarcus @shadbush @colin_fraser Gotcha riddles aren’t benchmarking. They’re intentional edge cases used to explore failure points. I don’t see why you’re trying to equate them to benchmarking.
Julian
@mealreplacer
·
345d
@Wasgo @GaryMarcus @shadbush @colin_fraser I don’t think that the relative weakness of many (though not all; like, SWE-bench seems somewhat okay-ish) existing benchmarks means we can’t make valid measures of AI capabilities. Just that it could be improved, by a lot. In a way that I don’t think gotcha riddles can.
Cameron Williams
@Wasgo
·
345d
@mealreplacer @GaryMarcus @shadbush @colin_fraser I mean, isn’t running an RFP conceding that the benchmarks don’t currently exist and haven’t been tested as a valid measurement?

Is it possible you’ll create one? Sure. Also possible you’ll fail. And the key is it doesn’t currently exist.
Julian
@mealreplacer
·
345d
@Wasgo @GaryMarcus @shadbush @colin_fraser https://www.openphilanthropy.org…

Here is an RFP we ran on the kinds of benchmarks I’m talking about. A well-executed version of these sorts of benchmarks would be very hard to game — different from how you can just train on a multiple choice QA benchmark.
Cameron Williams
@Wasgo
·
345d
@mealreplacer @GaryMarcus @shadbush @colin_fraser I think the challenge here is that any benchmark that can’t be gamed, would be incredibly difficult to make comparisons with.

If the problems don’t change, you write to the benchmark.
Julian
@mealreplacer
·
345d
@GaryMarcus @Wasgo @shadbush @colin_fraser Can you actually answer this question?

“Would you change your mind if there were difficult benchmarks that are testing LLMs on consequential, real-world tasks, and each generation improved upon the previous?”

The statement in question is part of a substantive engagement with public issues, specifically the benchmarking of AI capabilities. It addresses the validity and improvement of benchmarks for AI, which is a significant topic in the field of artificial intelligence and its impact on society.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not appear to cause harm with words or actions. It is a neutral comment that acknowledges a point made in the discussion. [+1]
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The statement does not explicitly promote understanding, empathy, or compassion, but it does contribute to a constructive dialogue by acknowledging a valid point in the discussion. [+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 criticism and dialogue without resorting to personal attacks or ad hominem arguments. It maintains a respectful tone and focuses on the content of the discussion. [+2]