Gary Marcus

Rank 14 of 47
|
Score 83
5/1/2025, 11:29:42 PM
X
In reply to:
Peter Wildeford 馃嚭馃嚫馃殌
@peterwildeford
路
245d
@GaryMarcus I'd say across the board, yeah
Gary Marcus
@GaryMarcus
路
245d
@peterwildeford Across the board? Or on some tasks, many of which it got specific and training on?
Peter Wildeford 馃嚭馃嚫馃殌
@peterwildeford
路
245d
@GaryMarcus I think GPT 4 vs 4.1 isn't the correct comparison, would compare GPT 4 vs o3. And I would think o3 is 6.25 times smarter than 4.
Gary Marcus
@GaryMarcus
路
245d
GPT-4 came out two years ago; is GPT 4.1 6.25 times smarter?

FLOPs/s have scaled faster than intelligence.
Epoch AI
@EpochAIResearch
路
245d
We've documented over 500 AI supercomputing clusters from 2019 onward.

Among leading AI supercomputers, performance has been increasing at a rate of 2.5x per year.

Thread of our insights 馃У

The statement questions the claim that a newer AI model is significantly smarter than its predecessor, expressing skepticism about the metric of being '6.25x smarter.' It engages in a technical discussion about AI performance and intelligence, contributing to the discourse on AI development and evaluation.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not cause harm and is part of a technical discussion.聽[+1]
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    It promotes understanding by questioning and seeking clarification on AI performance metrics.聽[+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.
    Engages in constructive dialogue by questioning claims without personal attacks.聽[+1]