Serious question, has anyone else tried training a small AI model on a single, weird dataset?

I was trying to get a simple text generator to write better descriptions for my online shop. The normal advice is to feed it tons of varied data, but that wasn't working for my specific niche. On a whim, I gave it only the exact product descriptions from my own best-selling items, about 150 of them, and nothing else. After two days of training, the output was suddenly spot-on for my brand voice. It stopped making generic statements and started copying the specific, detailed style I use. I'm curious if this focused approach has worked for others with very specialized tasks, or if I just got lucky.

2 comments

2 Comments

john6508d ago

My experience says more data is always better.

the_eric8d ago

Depends on the data quality though. Bad data just makes the mess bigger.