Gary Marcus

Rank 33 of 47
|
Score -1
4/2/2024, 5:20:12 PM
X
In reply to:
Kömpileir e/acc↗️
@Kompileir
·
229d
@GaryMarcus @rao2z Go compare Claude 3 Opus and GPT-4 and get back to me.
Gary Marcus
@GaryMarcus
·
229d
@Kompileir Base models appear to be converging on a plateau.

i have not seen data to support an exponential increase in prompting, and would advise you to read @rao2z’s systematic work.
Kömpileir e/acc↗️
@Kompileir
·
229d
@GaryMarcus No qualitative leap? No major advances in benchmarks?

The quality of the model is reliant on the users ability to prompt it. This gap was lessened between every iteration. Benchmarks are in T-data therefor widely irrelevant. Used for marketing, nothing more.
Gary Marcus
@GaryMarcus
·
229d
@Kompileir lack of major advance on benchmarks, lack of a qualitative leap a la GPT-3 to GPT-4, and more and more industry leaders saying stuff like this
Kömpileir e/acc↗️
@Kompileir
·
229d
@GaryMarcus I'll bite, show me the plateau.
Gary Marcus
@GaryMarcus
·
229d
No, we are not currently in a period of exponential growth.

GPT-4 is over a year old; nothing has exponentially improved upon it, despite an entire extremely well-capitalized industry trying.

There was a great jump; now there is perhaps a plateau.

The statement is a polite and respectful sign-off from an ongoing conversation about AI technology. It expresses gratitude and does not engage in any form of cyberbullying, harassment, or hate speech. The statement does not contribute to public discourse but is part of a civil exchange between individuals.