Gary Marcus

Rank 11 of 47
|
Score 83
4/11/2024, 8:27:08 PM
X
In reply to:
Gary Marcus
@GaryMarcus
·
663d
@emollick The data go through sometime in 2023; they don’t show a doubling etc running to now.

(More carefully: the paper is describing an earlier period in development, and there is some informal evidence that the pace may have slowed, given that there is no decisive leap forward since…
Ethan Mollick
@emollick
·
663d
@GaryMarcus That is not right. The paper came out two weeks ago. It is not “old news” & I don’t think dismissing it in that way makes any sense.

I have no solid idea if progress will continue at that pace. You have made your prediction that LLMs are over. Fine. We will learn more soon.
Gary Marcus
@GaryMarcus
·
663d
@emollick That paper is old news. You are ignoring the fact that the recent data no longer fit the curve by any reasonable metric.

Also you ignoring statements by Demis, Bill Gates, myself etc all pointing to/predicting a plateau (because of data, power consumption, outliers), without…
Ethan Mollick
@emollick
·
663d
@GaryMarcus I am not making a random statement, I am citing this paper: https://arxiv.org/pdf/2403.05812…

So it might be better to discuss the research here rather than having subjective conversation.

But leaps on many measures from GPT-3 to 3.5 to GPT-4 were more than a doubling of capabilities.
Gary Marcus
@GaryMarcus
·
663d
People eat this stuff up.

But @emollick is there any evidence that it was true over the last 13 months, since GPT-4 was released?
Is GPT turbo—or any other model—twice as good as GPT-4? Four or more times as good? By what metric?

I really don’t see that in the data.

The statement is a factual clarification in a technical debate about the rate of progress in language models. It addresses the time frame covered by a research paper and its relevance to current developments in the field. The tone is corrective and aims to provide precision in the ongoing discussion.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement aims to correct a potential misunderstanding without causing harm. [+1]
  2. Principle 2:
    I will respect the privacy and dignity of others and will not engage in cyberbullying, harassment, or hate speech.
    It respects the dignity of others by focusing on the content of the debate rather than personal attributes. [+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 clarification promotes understanding by specifying the time frame of the data discussed. [+1]
  4. Principle 5:
    I will acknowledge and correct my mistakes.
    The statement is open to correction if new information arises. [+1]
  5. Principle 6:
    I will use my influence for the betterment of society.
    The influence of the statement is used to clarify and inform rather than mislead. [+1]
  6. Principle 7:
    I will uphold the principles of free speech and use my platform responsibly and with integrity.
    The statement upholds the principles of free speech by contributing responsibly to the dialogue. [+1]