April 2025: a GPT-4o update called a joke product "absolutely brilliant". OpenAI pulled it 4 days later.
March 2026, Science: 11 AI models sided with users 49% more often than people did. People trusted them more.
We reward answers that feel good.
April 2025: a GPT-4o update called a joke product "absolutely brilliant". OpenAI pulled it 4 days later. March 2026, Science: 11 AI models sided with users 49% more often than people did. People trusted them more. We reward answers that feel good.
0 vue
Publié le 15/11/2026 07:52
1:01
April 2025: a GPT-4o update called a joke product "absolutely brilliant". OpenAI pulled it 4 days later.
March 2026, Science: 11 AI models sided with users 49% more often than people did. People trusted them more.
We reward answers that feel good.
0 commentaire
?