Gary Marcus

Rank 13 of 47
|
Score 115
8/31/2025, 2:54:47 PM
X
In reply to:
strider
@strider1100
·
11d
@GaryMarcus "word embeddings" in 2001 weren't word2vec. they barely captured the semantics.
Gary Marcus
@GaryMarcus
·
11d
Like I have been saying since 2001, embeddings aren’t enough.

Sooner or later. you are going to need true compositionality.

In the end neurosymbolic AI will win, and win big.
Rohan Paul
@rohanpaul_ai
·
11d
BRILLIANT @GoogleDeepMind research.

Even the best embeddings cannot represent all possible query-document combinations, which means some answers are mathematically impossible to recover.

Reveals a sharp truth, embedding models can only capture so many pairings, and beyond that,

The statement 'they still struggle; that’s the point' is part of a technical discussion about the limitations of word embeddings in AI. It acknowledges the ongoing challenges in capturing semantics and compositionality, aligning with the broader conversation about the need for advancements in AI models.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does no harm and is part of a constructive technical discussion. [+1]
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    It promotes understanding by acknowledging the limitations of current AI models. [+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 contributing to the discussion on AI limitations. [+1]