Task 11: Option 2 Text-to-Image

It is surprising and disappointing that Copilot has failed to generate S&P 500 index performance for the past 30 years which are public and historical data. Based on several attempts, Copilot seems to be good at generating images based on inputted descriptions instead of picking up information from the instruction, find or process the public information and historical data to generate images accordingly. The style of images generated share similarities regarding pixels, colour scheme and styles.

Next, I tried a different topic. First attempt is general, second attempt I added 1 year old information. However, the image pops out with 3 boys!

Then I put "1 year old" information before twin boys in my instruction. The image showed up alright with toys suitable for one year old babies featuring bright colors.

This exercise made me doubt the ability for Copilot to understand human's instruction and different ways to present the same instructions could lead to different image result. This might be one of the limitations Generative AI tools experience at current stage. 

It is doubtful if AI generates image from capturing key words from instruction instead of thinking the logic behind, the purpose of the instruction but rather mechanical. It searches the digital library and match the elements with wording. Unlike human's imagination, the ability to create image from instruction is limited by human input.