You can basically input any data you have, to make it more accurate on the subject you choose.

For education: Learn a curriculum and help students during their home work.

For businesses: Learn about your product and answer any questions (help desk, employee training…).

For personal use: Learn it about training schedules, for a healthier workout.

The public available app is pretty much a main stream, answer as much as possible solution. The true value is the technology itself. How accurate the answers are, with the data provided.

Internal documentation knowledge base.

Many companies have documentation scattered between services - google drive, confluence, maybe something bespoke for engineering documentation. There’s also ad hoc documentation happening every day - jira tickets, github comments, slack convos, video meetings. It’s a skill in itself to proficiently search disparate services to answer questions. Knowledge is lost every time someone leaves a company. Knowledge also degrades over time - we forget the context of our decisions, later leaving us to wonder why we chose something particular in the past. Or even if we deliberately chose to do something or it’s just a consequence of other decisions. Imagine onboarding. It’s a huge pain point.

Now imagine as much of that context that’s recorded in some way is aggregated, ranked, and easily searchable in a single interface. Sounds pretty sweet in comparison to the reality many startups face.

This is the top use case I would have for chat gpt builder, but there are other LLMs not managed by a third party that are a much better choice if data governance is a concern. I imagine anything training on gpt builder is available to chat gpt itself. Perhaps anonymized to some level, but in practice with tech like this that really just means it’s impossible to attribute the creator after the data enters the model. So anonymization by consequence, not necessarily design.

My dayjob has guidelines about what we are allowed to feed into chat gpt prompts, what policies to follow if using GitHub copilot, etc. There is no way we would ever get something like handing over our IP to open ai approved. However, not every company has resources to train its own internal LLM so the trade off is probably worth it, not important enough to consider right now. It’s also likely these details are buried in legalese hiding within the terms so people agree to it without even realizing.

There are different real use cases in my opinion. One is to be able to create a specialized GPT model to be used as a virtual assistant for your company or product.

Another approach may also be to assist you in specific research. For example, you may create a model that knows everything Physics with state-of-the-art sources that may help you in doing further research

These are the main ones that come to my mind

Lots of use cases complied here. I curated many other GPTs in this directory. Check it out