Gary Marcus

Rank 19 of 47
|
Score 64
3/1/2025, 2:41:34 PM
X
In reply to:
Jonas Korreborg Bertelsen
@iamJonasB
·
132d
@GaryMarcus @kimmonismus Hey Gary, by “ignoring” the reasoning stuff, I meant you’re acting like there’s a wall when these models are moving fast. Like, o3 scored 87.5% on that ARC-AGI test and it’s much better at STEM than o1. Also, GPT-4.5 is hallucinating 40% less than 4o. That’s solid progress toward…
Gary Marcus
@GaryMarcus
·
132d
@iamJonasB @kimmonismus Not at all ignoring it, but (as i have said repeatedly) it’s pretty clear that it is somewhat limited to narrow domains that lend themselves to verification and data augmentation.

None of the reasoning models are close to AGI.
Jonas Korreborg Bertelsen
@iamJonasB
·
132d
@kimmonismus Its jaw dropping how Gary Marcus simply just ignore the reasoning paradigm
Chubby♨️
@kimmonismus
·
132d
https://x.com/kimmonismus/status…
Gary Marcus
@GaryMarcus
·
133d
A big surprise only to you and the many other OpenAI fans who refused to listen to me, when I patiently and endlessly explained that this would be exactly what would happen.

For the rest of us, this was all but inevitable.

The statement is part of a public discourse on the capabilities and limitations of AI models, specifically discussing the performance and reasoning abilities of models like GPT-4.5 and ARC-AGI. The tone is analytical and critical, focusing on the perceived shortcomings of AI models in achieving AGI and the relevance of certain training metrics.

  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, as it focuses on technical critique rather than personal attacks.
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The statement could promote understanding by discussing the limitations of AI models, but it lacks empathy or compassion in its delivery.
  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 of AI models and their evaluation metrics, without resorting to personal attacks or ad hominem arguments. [+1]