Gary Marcus

Rank 11 of 47
|
Score 114
5/6/2025, 2:20:21 PM
X
In reply to:
Mike Brooks
@DrMikeBrooks
·
87d
@GaryMarcus @ArrowFodder We will see! Gary, you know just as well as I do, that there are people as smart and as technical as you who disagree. Just keep extending the timeline. The problem is, the progress we evolved to make has resulted in us creating a world we didn’t evolve to inhabit.
Gary Marcus
@GaryMarcus
·
87d
@DrMikeBrooks @ArrowFodder Improvement is NOT inevitable. Hallucinations just got WORSE in o3, for example.
Mike Brooks
@DrMikeBrooks
·
87d
@ArrowFodder @GaryMarcus I think both things are true. There are significant problems – AND it will get better. Human beings are very good at making things better when there are profits to be made, markets to be dominated and egos to be enhanced. I use 4 models simultaneously. They are definitely better.
ArrowFodder
@ArrowFodder
·
88d
@DrMikeBrooks @GaryMarcus You are anthropomorphizing it. It’s not on its way to getting better. The edge cases are significant because they reveal how the GenAI/LLM actually functions in its current state. The frequency of hallucinations are not trending down as data sets get bigger.
Mike Brooks
@DrMikeBrooks
·
89d
@ArrowFodder @GaryMarcus I respectfully disagree. We don’t get an accurate picture when we only focus on the flaws and limitations. It’s both/and, not either/or. When AI can already do this, we need to ensure we take into account how “smart” they already are: https://arxiv.org/html/2504.1613…
ArrowFodder
@ArrowFodder
·
89d
@DrMikeBrooks @GaryMarcus I don’t think so. The religious thinking is coming from the “LLMs show AGI is imminent” camp. The multitude of examples showing how LLMs/genAi get it wrong is a fundamental problem with GenAI. Gary has pointed out these same issues for years. He’s not just getting lucky :)

The statement is part of a broader discussion on the development and challenges of AI, specifically addressing the issue of AI hallucinations and the potential for improvement. The tone is assertive, claiming a statement as fact within a technical debate.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not directly cause harm, but the assertive tone could be perceived as dismissive if not substantiated, which might indirectly affect the discourse negatively.
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The statement does not actively promote understanding, empathy, or compassion, as it simply asserts a fact without elaboration or explanation. [-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 does not engage in personal attacks or ad hominem arguments, but it lacks constructive criticism or dialogue, as it merely asserts a fact without engaging with the opposing viewpoint. [-1]