It all started with a small experiment to restore an image I have using ChatGPT.

The curiosity as an engineer that profess system thinking wanted to know more on how ChatGPT enhanced the image. So I asked a follow up question “are the image rebuilt using known images that matches this image ?”

The answers feels almost perfect for folks that does not have domain knowledge about digital image. To top of that from my post graduate degree I learned an important concept about AI or Machine learning. The quote from Tom Mitchell “A computer program is said to learn from experience (E) with respect to some class of tasks (T) and performance measure (P), if its performance at tasks in (T), as measured by (P), improves with experience (E).“

Experience (E): The data or learning experience the program receives.

Class of Tasks (T): The specific task or class you want the computer to perform.

Performance Measure (P): The testing and evaluation phase used to verify if accuracy is improving.

In this case I realized and very aware that the glare on my photo has no more data, which makes it impossible to be restored without understanding of the logo and character. To dig deeper I need to use to proper keyword “data wipe out from the glare” in the prompt.

Fair explanation on the first part. Until you realized it is using lawyer tactics to misdirect and to eloquently gaslight in the second part of the explanation.

I can see that ChatGPT uses a wily constructed sentence used by a sleazy colleague or boss to project power. Gaslighting and dismissive of your counter point. ChatGPT would be able to convince folks that has no domain knowledge.

Conclusion, make sure you do a “doverjai, no proverjai” when using of AI or “generative AI“.