Gary Marcus

Rank 25 of 47
|
Score 40
1/10/2025, 8:32:10 PM
X
In reply to:
Gary Marcus
@GaryMarcus
·
25d
@littmath @vedantmisra @ChrSzegedy 1. Vedant, unless you can contradict the above with data, you owe me at least a partial correction.

2. Daniel and Vedant, what does “reasonably well” mean?

Remember that the question wasn’t just about summaries, and should be performed on new novels that don’t have secondary…
Daniel Litt
@littmath
·
25d
@GaryMarcus @vedantmisra @ChrSzegedy I think it's clear AI does not yet meet the math criterion yet; as I've argued before I think the standard is a bit unclear but I wouldn't be surprised if some form of it is met by 2030. FWIW I think it's pretty clear o1, say, performs reasonably well on your "read a novel" test.
Gary Marcus
@GaryMarcus
·
25d
@vedantmisra 1. Prove it. Show me the data. I don’t know any systematic public data supporting a systematic ability to do any of what I set a criteria. cc’ing @littmath and @ChrSzegedy who know the math part, and laying out the original criteria since you oversimplified them.

Where is the…
Vedant Misra
@vedantmisra
·
25d
@GaryMarcus Gary, one easy way for me to answer this question would be to simply list everything that you said wouldn't happen before 2030, most of which has already happened. Specifically, AI can already comprehend a movie, answer questions about a novel, write 10K lines of code, and…
Gary Marcus
@GaryMarcus
·
26d
1. No reply from the DeepMind guy, @vedantmisra, who said with enormous confidence that “everything was about to change”.

2. Expect more of this from the AI industry: vague claims about how things are about to change, with no commitment whatsoever whether that means this year or…
Gary Marcus
@GaryMarcus
·
27d
@vedantmisra On what time scale? “About to change” is quite vague — a year? 5? 10? What would count as evidence that you are all in fact wrong?

The statement is part of a public discourse on the capabilities and future of AI, specifically regarding comprehension and other tasks. It references a scholarly paper to support a point in an ongoing debate about AI's progress and potential.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not appear to cause harm and is part of a constructive debate about AI capabilities. [+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 referencing a scholarly work to support its argument, contributing to a factual discussion. [+1]
  3. Principle 5:
    I will acknowledge and correct my mistakes.
    The statement acknowledges previous work and seeks to clarify or correct misunderstandings, aligning with the principle of acknowledging and correcting mistakes. [+1]