Question

How many parameters (in billions) is needed to achieve superhuman few-shot performance on SuperGLUE?

4 comments4

Total Forecasters32

Community Prediction

107

(24.8 - 627)

Make a Prediction

PDF

CDF

This question is closed for predictions, and is waiting to be resolved

	<20	lower 25%	median	upper 75%	>50k
community	19.6%	24.84	106.6	627.2	4.3%
My Prediction	—	—	—	—	—

Lower bound	community	My Prediction
<20	19.6%	—
Quartiles
lower 25%	24.84	—
median	106.6	—
upper 75%	627.2	—
Upper bound
>50k	4.3%	—

Authors:

Tamay

Opened:

Sep 15, 2020

Closed:

Dec 31, 2021

Scheduled resolution:

Dec 31, 2029

Spot Scoring Time:

Sep 17, 2020

Computing and Math

Artificial Intelligence

2020-2021 Leaderboard

Google

Machine learning

OpenAI

How many billions of parameters will the largest machine learning model trained before 2030 have?

1.44M

Will a dense machine learning model with at least 100 trillion parameters be trained before 2026?

28%

When will a Chinese entity develop a model surpassing GPT-4's few-shot performance on MMLU?

20 Feb 2025