Gary Marcus

Rank 25 of 47
|
Score 40
4/3/2024, 6:41:55 AM
X
In reply to:
Ivan Uemlianin
@llaisdy
·
308d
@GaryMarcus Constant within a margin of error, and low enough that the margin of error goes below zero.
Gary Marcus
@GaryMarcus
·
308d
@llaisdy There is no way that the (publicly known) progress in the year since gpt-4 marches the progress in the preceding year.

Progress hasn’t been constant. It appears to have decelerated.

Exponential implies a constant. Good luck fitting the data.
Ivan Uemlianin
@llaisdy
·
308d
@GaryMarcus I did -- size of exponent the issue? Perhaps we could say that exponential *does* mean "a bit better than the last one", but hypesters use the word just to mean HUGE (and scientific).
Gary Marcus
@GaryMarcus
·
308d
@llaisdy Yeah, but read the comments.
Ivan Uemlianin
@llaisdy
·
308d
@GaryMarcus An annual 0.1% improvement is an exponential rate of improvement so, "exponential" can mean "a little better than last year".
Gary Marcus
@GaryMarcus
·
309d
Pro-tip: Exponential doesn’t mean “a little better than last year”.

The statement is part of a conversation about the interpretation of data and the meaning of 'exponential' in the context of progress in a certain field. The tone is challenging but not disrespectful, and the intent seems to be to encourage evidence-based discussion on the topic. The content is focused on the accuracy of data interpretation and does not directly address societal issues or policies.