Choosing between a custom GPT and a standard API for my local news project

I had to decide whether to build a custom GPT model for my town's news aggregation site or just use a standard API like OpenAI's. I went with the custom model because I needed it to understand hyperlocal slang and place names, like 'the old mill by the river' meaning a specific park. It took about 3 months of fine-tuning on local forum data. Did I waste my time, or is the tailored approach always better for niche uses?

4 comments

4 Comments

jesse_price2729d ago

Nah, that's not a waste of time at all. Getting it to actually understand local talk is huge for making it useful.

bennett.harper29d ago

You're right, I used to think that was silly.

morgan_king3629d ago

Honestly, did you see that article about the team training an AI on old city council meeting tapes? They said the slang and inside jokes were the hardest part to get right. Tbh if it just speaks textbook English it feels fake and people won't trust it. Getting the local flavor right is what makes it actually helpful for real people. Ngl that's the kind of work that actually matters.

charles6407d ago

Morgan's point about trust is the key thing. If your news bot sounds like a textbook, people will just ignore it. The real win with your custom model is that it will keep working even if the big API services change their rules or go down. You built something that actually knows your town and isn't just renting general knowledge.