OpenAI (OpenLie?) GPT4o: the impossible results
Mohamed Jean-Philippe Sangaré
Director of the Faculty of Science and Technology / Founder & CEO of SSVAR
From experience, I never take for granted what a developer says about their software release.
The very impressive announcements and demos by #OpenAI on May 12th, 2024, gave me the desire to test the latest version, GPT-4o.
Here is the link: https://openai.com/index/hello-gpt-4o/
And... the few examples I've tried to reproduce are not OK.
Not OK at all.
I don't know how they got their results!?
Test 1: Visual Narrative - Sally the mailwoman
What I've got:
0(zero) consistency...
Test 2: Visual Narrative - Robot writer's Block
See? the text in typewriter is consistent ...
But... Me:
Test 3 : 3D object synthesis
OpenAI :
They even have an animation... ??
Me:
Again... 0 consistency
Test 4 : Geary the robot
Same differences.
I didn't go further....
If you do it , please leave a comment.
Is the Emperor naked? ;-)
Waiting for answers...
Digital Mental Health Specialist | CEO, Mentegram | International Digital mental health expert, UNICEF | CTO, EverMind | Solution Architect, Hashcode | - Digital transformation of mental health services
6 个月It’s possible that the inconsistent images were produced because you did not attach the first image created by AI as a reference to get a variation.
Seeking New International Position: international Organisations, International Centres in Ehealth- Global Health Informatics- international Management. CMEO@SSVAR IS.
6 个月Interesting!
SUPPLY CHAIN AND LOGISTICS EXPERT Self -Employed
6 个月As I trust your analysis you save me time to test the version proposed. Thanks again Mohamed for sharing