Gary Marcus

Rank 17 of 47
|
Score 68
9/13/2024, 10:31:21 PM
X
In reply to:
Julian
@mealreplacer
·
344d
@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
·
344d
@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
·
344d
@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?”
Gary Marcus
@GaryMarcus
·
344d
@mealreplacer @Wasgo @shadbush @colin_fraser I feel like a couple of us here are trying hard to explain the problem is—reliable generalization to novel problems—and you just don’t want to get it. I can’t help with that.
Gary Marcus
@GaryMarcus
·
344d
@mealreplacer @Wasgo @shadbush @colin_fraser Dude if a system can’t reliably follow the rules of chess what do you think happens on more complex problems?!

You can’t have AGI in a system that frequently and unpredictably says ridiculous things even in relatively easy problems.
Julian
@mealreplacer
·
344d
@Wasgo @GaryMarcus @shadbush @colin_fraser You really can’t just merely assert that failing on weird little problems is something that is strictly necessary to be solve before we get “real AGI”

The statement constitutes public discourse as it engages substantively with the issue of benchmarking in AI, a topic of public concern within the field of artificial intelligence. The tone is analytical and aims to contribute to the ongoing debate about the validity and reliability of AI benchmarks.

  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 reasoned critique of the difficulty in creating valid benchmarks for AI, which is a constructive contribution to the discourse. [+1]
  2. Principle 2:
    I will respect the privacy and dignity of others and will not engage in cyberbullying, harassment, or hate speech.
    The statement respects the privacy and dignity of others. It does not engage in cyberbullying, harassment, or hate speech. [+1]
  3. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The statement promotes understanding by highlighting a recognized challenge in the field of AI. It encourages empathy by acknowledging the difficulty of the task. [+1]
  4. 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 does not resort to personal attacks or ad hominem arguments. It focuses on the issue rather than individuals. [+1]
  5. Principle 7:
    I will uphold the principles of free speech and use my platform responsibly and with integrity.
    The statement upholds the principles of free speech and uses the platform responsibly and with integrity. It contributes meaningfully to the public discourse on AI benchmarks. [+1]