Trying to create my own GPT for work, any tips?

I’m working on making my own GPT for work. I’m a product manager for a SaaS company. I’ve been uploading various files, but I’m wondering if there’s a better way to organize it all. I want this GPT to understand the industry, answer product questions, help with writing internal docs, and support teams like customer service and marketing. Any advice would be great!

Sounds like you might want a RAG (Retrieval Augmented Generation) model. You don’t have to train GPT, just use a vector database to feed it info. You should check out discussions about RAG on this forum—lots of people are doing what you’re trying to do.

Noel said:
Sounds like you might want a RAG (Retrieval Augmented Generation) model. You don’t have to train GPT, just use a vector database to feed it info. You should check out discussions about RAG on this forum—lots of people are doing what you’re trying to do.

Wow, it really is a rabbit hole! Just joined the forum, and there’s a lot to read about.

Noel said:
Sounds like you might want a RAG (Retrieval Augmented Generation) model. You don’t have to train GPT, just use a vector database to feed it info. You should check out discussions about RAG on this forum—lots of people are doing what you’re trying to do.

You could also try feeding GPT a document and telling it to remember it. Then add custom instructions to use the saved data. Not sure if it’ll work, but worth trying.

Otherwise, I still think RAG is the right direction for what you’re trying to do.

Noel said:
Sounds like you might want a RAG (Retrieval Augmented Generation) model. You don’t have to train GPT, just use a vector database to feed it info. You should check out discussions about RAG on this forum—lots of people are doing what you’re trying to do.

This might be a bit much for me. Do you think it’s something I can’t do with the custom GPT option?

Noor said:

Noel said:
Sounds like you might want a RAG (Retrieval Augmented Generation) model. You don’t have to train GPT, just use a vector database to feed it info. You should check out discussions about RAG on this forum—lots of people are doing what you’re trying to do.

This might be a bit much for me. Do you think it’s something I can’t do with the custom GPT option?

Not if you’re looking for reliability and accuracy.

Yan said:

Noor said:
Noel said:
Sounds like you might want a RAG (Retrieval Augmented Generation) model. You don’t have to train GPT, just use a vector database to feed it info. You should check out discussions about RAG on this forum—lots of people are doing what you’re trying to do.

This might be a bit much for me. Do you think it’s something I can’t do with the custom GPT option?

Not if you’re looking for reliability and accuracy.

Hmm, my manager made one and says it’s really helpful for her.

Noor said:

Yan said:
Noor said:
Noel said:
Sounds like you might want a RAG (Retrieval Augmented Generation) model. You don’t have to train GPT, just use a vector database to feed it info. You should check out discussions about RAG on this forum—lots of people are doing what you’re trying to do.

This might be a bit much for me. Do you think it’s something I can’t do with the custom GPT option?

Not if you’re looking for reliability and accuracy.

Hmm, my manager made one and says it’s really helpful for her.

Maybe she just doesn’t notice or care when it makes mistakes? Some people can’t tell if the info is wrong.

Winslow said:

Noor said:
Yan said:
Noor said:
Noel said:
Sounds like you might want a RAG (Retrieval Augmented Generation) model. You don’t have to train GPT, just use a vector database to feed it info. You should check out discussions about RAG on this forum—lots of people are doing what you’re trying to do.

This might be a bit much for me. Do you think it’s something I can’t do with the custom GPT option?

Not if you’re looking for reliability and accuracy.

Hmm, my manager made one and says it’s really helpful for her.

Maybe she just doesn’t notice or care when it makes mistakes? Some people can’t tell if the info is wrong.

True!

Reese said:

Winslow said:
Noor said:
Yan said:
Noor said:
Noel said:
Sounds like you might want a RAG (Retrieval Augmented Generation) model. You don’t have to train GPT, just use a vector database to feed it info. You should check out discussions about RAG on this forum—lots of people are doing what you’re trying to do.

This might be a bit much for me. Do you think it’s something I can’t do with the custom GPT option?

Not if you’re looking for reliability and accuracy.

Hmm, my manager made one and says it’s really helpful for her.

Maybe she just doesn’t notice or care when it makes mistakes? Some people can’t tell if the info is wrong.

True!

What kind of mistakes does it make?

Briar said:

Reese said:
Winslow said:
Noor said:
Yan said:
Noor said:
Noel said:
Sounds like you might want a RAG (Retrieval Augmented Generation) model. You don’t have to train GPT, just use a vector database to feed it info. You should check out discussions about RAG on this forum—lots of people are doing what you’re trying to do.

This might be a bit much for me. Do you think it’s something I can’t do with the custom GPT option?

Not if you’re looking for reliability and accuracy.

Hmm, my manager made one and says it’s really helpful for her.

Maybe she just doesn’t notice or care when it makes mistakes? Some people can’t tell if the info is wrong.

True!

What kind of mistakes does it make?

It can mix up facts, provide wrong data, or confuse terms.

Galen said:

Briar said:
Reese said:
Winslow said:
Noor said:
Yan said:
Noor said:
Noel said:
Sounds like you might want a RAG (Retrieval Augmented Generation) model. You don’t have to train GPT, just use a vector database to feed it info. You should check out discussions about RAG on this forum—lots of people are doing what you’re trying to do.

This might be a bit much for me. Do you think it’s something I can’t do with the custom GPT option?

Not if you’re looking for reliability and accuracy.

Hmm, my manager made one and says it’s really helpful for her.

Maybe she just doesn’t notice or care when it makes mistakes? Some people can’t tell if the info is wrong.

True!

What kind of mistakes does it make?

It can mix up facts, provide wrong data, or confuse terms.

I know what AI ‘hallucinations’ are, but there are ways to minimize this, depending on how it’s set up. It’s why AI should always have human oversight. But for brainstorming, content organization, or structuring systems, a well-built GPT can be very useful.

Briar said:

Galen said:
Briar said:
Reese said:
Winslow said:
Noor said:
Yan said:
Noor said:
Noel said:
Sounds like you might want a RAG (Retrieval Augmented Generation) model. You don’t have to train GPT, just use a vector database to feed it info. You should check out discussions about RAG on this forum—lots of people are doing what you’re trying to do.

This might be a bit much for me. Do you think it’s something I can’t do with the custom GPT option?

Not if you’re looking for reliability and accuracy.

Hmm, my manager made one and says it’s really helpful for her.

Maybe she just doesn’t notice or care when it makes mistakes? Some people can’t tell if the info is wrong.

True!

What kind of mistakes does it make?

It can mix up facts, provide wrong data, or confuse terms.

I know what AI ‘hallucinations’ are, but there are ways to minimize this, depending on how it’s set up. It’s why AI should always have human oversight. But for brainstorming, content organization, or structuring systems, a well-built GPT can be very useful.

It gets a lot more accurate if you use RAG.

Ainsley said:

Briar said:
Galen said:
Briar said:
Reese said:
Winslow said:
Noor said:
Yan said:
Noor said:
Noel said:
Sounds like you might want a RAG (Retrieval Augmented Generation) model. You don’t have to train GPT, just use a vector database to feed it info. You should check out discussions about RAG on this forum—lots of people are doing what you’re trying to do.

This might be a bit much for me. Do you think it’s something I can’t do with the custom GPT option?

Not if you’re looking for reliability and accuracy.

Hmm, my manager made one and says it’s really helpful for her.

Maybe she just doesn’t notice or care when it makes mistakes? Some people can’t tell if the info is wrong.

True!

What kind of mistakes does it make?

It can mix up facts, provide wrong data, or confuse terms.

I know what AI ‘hallucinations’ are, but there are ways to minimize this, depending on how it’s set up. It’s why AI should always have human oversight. But for brainstorming, content organization, or structuring systems, a well-built GPT can be very useful.

It gets a lot more accurate if you use RAG.

RAG definitely makes the system more reliable, but there are some steps to set it up well.

Noor said:

Yan said:
Noor said:
Noel said:
Sounds like you might want a RAG (Retrieval Augmented Generation) model. You don’t have to train GPT, just use a vector database to feed it info. You should check out discussions about RAG on this forum—lots of people are doing what you’re trying to do.

This might be a bit much for me. Do you think it’s something I can’t do with the custom GPT option?

Not if you’re looking for reliability and accuracy.

Hmm, my manager made one and says it’s really helpful for her.

I’ve been playing around with this using the Custom GPT builder. There are some podcasts that helped me:

  • AI Explored with Michael Stelzner
  • Guests like Igor Pogany, Brian Piper, and Russ Somers have great tips.

Hope that helps!

Noor said:

Yan said:
Noor said:
Noel said:
Sounds like you might want a RAG (Retrieval Augmented Generation) model. You don’t have to train GPT, just use a vector database to feed it info. You should check out discussions about RAG on this forum—lots of people are doing what you’re trying to do.

This might be a bit much for me. Do you think it’s something I can’t do with the custom GPT option?

Not if you’re looking for reliability and accuracy.

Hmm, my manager made one and says it’s really helpful for her.

Custom GPTs can be super helpful, but it’s not really the same as ‘training’ a model. I don’t think you need to train a new model unless it’s for customers.

Noor said:

Noel said:
Sounds like you might want a RAG (Retrieval Augmented Generation) model. You don’t have to train GPT, just use a vector database to feed it info. You should check out discussions about RAG on this forum—lots of people are doing what you’re trying to do.

This might be a bit much for me. Do you think it’s something I can’t do with the custom GPT option?

Creating a custom GPT that works across different platforms can be expensive. You need developers for integration, user interfaces, and making sure it works everywhere. That’s why it can get pricey.

Briar said:

Noor said:
Noel said:
Sounds like you might want a RAG (Retrieval Augmented Generation) model. You don’t have to train GPT, just use a vector database to feed it info. You should check out discussions about RAG on this forum—lots of people are doing what you’re trying to do.

This might be a bit much for me. Do you think it’s something I can’t do with the custom GPT option?

Creating a custom GPT that works across different platforms can be expensive. You need developers for integration, user interfaces, and making sure it works everywhere. That’s why it can get pricey.

What?

Micah said:

Briar said:
Noor said:
Noel said:
Sounds like you might want a RAG (Retrieval Augmented Generation) model. You don’t have to train GPT, just use a vector database to feed it info. You should check out discussions about RAG on this forum—lots of people are doing what you’re trying to do.

This might be a bit much for me. Do you think it’s something I can’t do with the custom GPT option?

Creating a custom GPT that works across different platforms can be expensive. You need developers for integration, user interfaces, and making sure it works everywhere. That’s why it can get pricey.

What?

If you want your custom GPT to work across mobile and web apps, it takes some engineering work. OpenAI’s API is flexible, but you’d need developers for full integration.

Briar said:

Micah said:
Briar said:
Noor said:
Noel said:
Sounds like you might want a RAG (Retrieval Augmented Generation) model. You don’t have to train GPT, just use a vector database to feed it info. You should check out discussions about RAG on this forum—lots of people are doing what you’re trying to do.

This might be a bit much for me. Do you think it’s something I can’t do with the custom GPT option?

Creating a custom GPT that works across different platforms can be expensive. You need developers for integration, user interfaces, and making sure it works everywhere. That’s why it can get pricey.

What?

If you want your custom GPT to work across mobile and web apps, it takes some engineering work. OpenAI’s API is flexible, but you’d need developers for full integration.

I did something like this for my home system, just to play around. It worked okay, but it wasn’t optimized since I’m not a developer. It sometimes took 10 seconds to respond. Still fun, though.