Few shot not giving great result with flan-t5 in certain cases

I changed few shot prompt to following, 

example_indices_full = [80, 120, 200]
example_index_to_summarize = 40

few_shot_prompt = make_prompt(example_indices_full, example_index_to_summarize)

print(few_shot_prompt)

However the summary was still not great ,
---------------------------------------------------------------------------------------------------
BASELINE HUMAN SUMMARY:
#Person1# is in a hurry to catch a train. Tom tells #Person1# there is plenty of time.
---------------------------------------------------------------------------------------------------
MODEL GENERATION - FEW SHOT:
Tom is late for the train. He has to catch it at 9:30.

Do you recommend trying models other than flan-t5 in such scenarios?

I gave the dialogue input to chatgpt4 and asked it to summarize in 10 words, I got following which is pretty great.

Person1 asks time, realizes lateness, rushes for train; Person2 reassures.

1 Like

Flan T5 is a much smaller model than chatgpt and trained also in a smaller dataset, so this kind of behaviour in comparison to chatgpt is to be expected.