Expert mode for grant writing (in Spanish)
I uploaded document A with previous grant for the same project and document B with the new call. Asked it to write the first part. First output, mediocre. But then I said, "Read it as an evaluator and find all what is wrong with it." Did very good observations. Then I said, "Now write a new version taking all this into consideration". The output was near perfect. Repeated this 3-step procedure with every part of the grant, and it performed great.
I did not use v 3.2 off expert mode to write since I concluded that it stinked at it in comparison with other models (thus I don't have a recent memory of it). Also, maybe I got better at prompting. But I believe that this kind of writing performance is new for DeepSeek. Did any of you have a similar experience with expert mode?