Evaluating ChatGPT 4o: Comparing Responses to Full vs. Partial Context Prompts

Today, I saw this meme on Facebook:


English Translation:
First message (at 18:31):
“If I pass away, will you find a new girlfriend?”

Second message (at 18:32):
“What exactly do you mean by sending me this picture?”

Third message (at 18:32):
“You better explain it clearly to me.”

I got curious and wanted to see if ChatGPT 4o could understand the meaning of the image.

I conducted two tests. In the first test, I assumed the role of the boyfriend. I posted the entire image to ChatGPT and said, “She doesn’t understand, I’m exhausted. Can you help explain?” ChatGPT understood it accurately.

Then, I removed the last two lines from the girlfriend’s part and, pretending to be the girlfriend, asked ChatGPT, “What exactly does he mean by sending me this picture?” This time, ChatGPT completely missed the meaning of the image.

---
I wonder if this means we need to tell ChatGPT the entire story from beginning to end for it to understand what happened. If we only give it the beginning and expect it to continue or imagine the future development, it might not perform as well.