Gary Marcus

Rank 19 of 47
|
Score 64
9/30/2024, 4:28:10 PM
X
In reply to:
Leon Lang
@Lang__Leon
·
286d
@GaryMarcus @jam3scampbell Likewise, I don’t know of a principled solution to getting human brains to make these mistakes. These issues do in fact pop up with human brains, especially for short thinking time. Yet, we are clearly an AGI.
How do you bound AI below human capabilities in the current paradigm?
Gary Marcus
@GaryMarcus
·
286d
@Lang__Leon @jam3scampbell There is no *principled* solution to getting black boxes to deal with the issues I have raised. And same issues have popped up in hundreds of models since I first made these predictions. all of them.

That is evidence.
Leon Lang
@Lang__Leon
·
286d
@GaryMarcus @jam3scampbell Tbc., I’m just saying I wouldn’t know how to be confident that those problems will *not* be sufficiently solved by 2027. I’m not making a “hopeful” prediction of my own here.
Gary Marcus
@GaryMarcus
·
286d
@Lang__Leon @jam3scampbell O1 isn’t even more reliable across all domains, seems to work best only where data is augmented; poor performance on large arithmetic and planning problems, still hallucinates. What you say is just hope, not tethered in specific results addressing my concerns.
Leon Lang
@Lang__Leon
·
286d
@GaryMarcus @jam3scampbell What makes you so confident? There’s a chance that inference compute scaling is a “real thing” and that letting models think for minutes, hours, or days, while self-correcting, could solve all of these issues. I wouldn’t know how to rule that out by 2027.
Gary Marcus
@GaryMarcus
·
286d
@jam3scampbell Meaning, of course: AGI by 2027 is not plausible. We are too far from reliability, sound reasoning, planning, etc, and still saddled with black boxes that regularly hallucinate, etc

The statement constitutes public discourse as it engages in a substantive discussion about artificial general intelligence (AGI), its reliability, and the challenges associated with it. The tone is critical and somewhat dismissive, aiming to refute points made by others in the conversation.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not appear to cause harm directly, but the dismissive tone could be seen as unconstructive.
  2. Principle 2:
    I will respect the privacy and dignity of others and will not engage in cyberbullying, harassment, or hate speech.
    The statement does not engage in cyberbullying, harassment, or hate speech, and respects the privacy and dignity of others. [+1]
  3. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The statement does not promote understanding, empathy, or compassion; it is more focused on refutation. [-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 criticism but does not provide constructive dialogue or avoid personal attacks, as it dismisses the previous points as 'sophistry.' [-1]
  5. Principle 7:
    I will uphold the principles of free speech and use my platform responsibly and with integrity.
    The statement upholds the principle of free speech and uses the platform to engage in a public debate, though the tone could be more respectful. [+1]