PresentationPDF Available

Comparison of Gemini Ultra and GPT-4v

Authors:
  • Central Scientific Research Insitute of Armaments and Military Equipment of Armed Forces of Ukraine

Abstract

Based on the Gemini Ultra test results, can be say that GPT-4v is definitely better than Gemini Ultra!!! The hype around the absolute benefits of Gemini Ultra is just a purely business PR campaign that mainly misleads users and tries to pass off wishful thinking. The multimodal capabilities of the Gemini Ultra v 1.0 are actually very limited and do not meet the requirements. At the same time, ideally, it is necessary to use these different LLMs, supplementing the gaps of one with the advantages of the other.
14 February 2024
1
VADYM SLYUSAR
Doctor of Sciences, Professor
14 February 2024
Comparison of Gemini Ultra
and GPT-4v
2
https://gemini.google.com/app
14 February 2024
3
The advantage of Gemini Ultra is easier generation of tables and the ability to
save generated tables in Google Sheets
Please build table from this text :
Layer (type) Output Shape Param #
==========================
conv2d (Conv2D) (None, 240, 320, 256) 7168
batch_normalization (Batch (None, 240, 320, 256) 1024
Normalization)
conv2d_1 (Conv2D) (None, 240, 320, 256) 590080
max_pooling2d (MaxPooling2 (None, 80, 106, 256) 0
D)
conv2d_2 (Conv2D) (None, 80, 106, 256) 590080
batch_normalization_1 (Bat (None, 80, 106, 256) 1024
chNormalization)
dropout (Dropout) (None, 80, 106, 256) 0
conv2d_3 (Conv2D) (None, 80, 106, 256) 590080
max_pooling2d_1 (MaxPoolin (None, 26, 35, 256) 0
g2D)
dropout_1 (Dropout) (None, 26, 35, 256) 0
conv2d_4 (Conv2D) (None, 26, 35, 512) 1180160
batch_normalization_2 (Bat (None, 26, 35, 512) 2048
chNormalization)
conv2d_5 (Conv2D) (None, 26, 35, 1024) 4719616
max_pooling2d_2 (MaxPoolin (None, 8, 11, 1024) 0
g2D)
dropout_2 (Dropout) (None, 8, 11, 1024) 0
flatten (Flatten) (None, 90112) 0
dense (Dense) (None, 2048) 184551424
dense_1 (Dense) (None, 240) 491760
dense_2 (Dense) (None, 2) 482
14 February 2024
4
The advantage of GPT4v is much better image generation in DALL E-3
“Please create the high resolution and realistic picture of the virtual laboratory for students that represented as
digital avatars, are engaged in various scientific activities, such as conducting experiments, analyzing data on
floating screens, and collaborating in this virtual environment”
GPT4v
Gemini Ultra
14 February 2024
5
The advantage of GPT4v is to download many images at once for analysis
14 February 2024
6
The advantage of GPT4v is the ability to classify images according to the user's
dataset
A training sample of metal 1 images simultaneously uploaded to GPT-4v
14 February 2024
7
A training sample of metal 2 images simultaneously uploaded to GPT-4v
14 February 2024
The advantage of GPT4v is the ability to classify images according to the user's
dataset
8
Metal 2 Metal 1
14 February 2024
The advantage of GPT4v is the ability to classify images according to the user's
dataset
9
14 February 2024
SUMMARY
GPT-4v is definitely better than Gemini Ultra!!!
The hype around the absolute benefits of Gemini Ultra
is just a purely business PR campaign that mainly
misleads users and tries to pass off wishful thinking.
The multimodal capabilities of the Gemini Ultra v 1.0
are actually very limited and do not meet the
requirements.
At the same time, ideally, it is necessary to use these
different LLMs, supplementing the gaps of one with the
advantages of the other.
THANK YOU
FOR YOUR ATTENTION !
swadim@ukr.net
www.slysuar.kiev.ua
https://scholar.google.com.ua/citations?hl=ru&user=wSegaWsAAAAJ
https://orcid.org/0000-0002-2912-3149
https://www.scopus.com/authid/detail.uri?authorId=7004240035
https://www.researchgate.net/profile/Vadym-Slyusar 10
14 February 2024
ResearchGate has not been able to resolve any citations for this publication.
ResearchGate has not been able to resolve any references for this publication.