Gary Marcus

Rank 19 of 47
|
Score 64
7/25/2024, 6:10:07 PM
X
In reply to:
Justin Harvey
@AI_Vidz
·
351d
@benjaminjriley @GaryMarcus When you say they are language models, full stop, what do you mean? Nearly all the frontier research now is multimodal
Benjamin Riley
@benjaminjriley
·
351d
@AI_Vidz @GaryMarcus For my part, the scientific evidence on how humans think suggests that many of our capacities are not dependent on language. Since LLMs are language models, full stop, this casts doubt on whether scaling linguistic connections will lead to LLMs having human-like cognition.
Benjamin Riley
@benjaminjriley
·
351d
@AI_Vidz @GaryMarcus ...we can also say the number of plausible theories about anything will always exceed the number that are proven to be true. What I understand Gary to be saying, and certainly what I would say, is that evidence is mounting that disfavors the "scale is all you need" hypothesis.
Benjamin Riley
@benjaminjriley
·
351d
@AI_Vidz @GaryMarcus I take your point. From a scientific perspective, we can hold open the *possibility* that these models will continue to dramatically improve in capability with continued scale. That is a testable hypothesis. But...
Justin Harvey
@AI_Vidz
·
352d
@GaryMarcus I really dont understand your thinking. We know the models get better with scale. No one has yet trained a model above GPT-4 scale, except OpenAI and that model is not yet released.

You have no evidence to support your point...
Gary Marcus
@GaryMarcus
·
352d
C’mon, you are a scientist. Or you think you know science.

When you see roughly a dozen models all about equal to one that was first trained two years ago, and none vastly better, even after NVidia sold $50B in chips, with all the financial incentive in the world to do much…

The statement 'multimodal language models are still generally regarded as language models' constitutes public discourse as it engages in a substantive discussion about the nature and classification of language models, which is a topic of public interest in the field of artificial intelligence.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not cause harm and is neutral in tone, adhering to the principle of doing no harm with words and actions. [+1]
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The statement aims to clarify a point in an ongoing discussion, which can promote understanding and empathy among participants. [+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 engages in constructive dialogue without resorting to personal attacks or ad hominem arguments. [+1]