MeteorGPT created

Hi all :wave:

At Vazco, we’ve been working with Meteor for a long time, and we thought to ourselves that a custom GPT assistant might not only benefit us but also the community, so we created one :slight_smile:

The current context window and knowledge isn’t that sophisticated but we have given it all of the Meteor docs including the changelog, added files with getting started tutorials, provided the link to these forums, and other material that we found helpful. We also fed it with a couple of our blog posts so that it can take advantage of those.

Feel free to give it a go:

It does require a paid ChatGPT plan though, same as all of those custom GPTs.

Our initial testing has shown a potential for it to be quite a usable asset, here are some of the prompt examples:

If you have ideas for a valuable context that we should provide to it to make it better, or you’ll find it outputting some misleading information, let us know and we’ll be happy to update it :slight_smile:

21 Likes

That is really nice! I was looking for something like this for Meteor, it can be really helpful for those who are starting and want to create something with Meteor

3 Likes

Yes helpful but you can only have access to it when you are only using the paid one. I stand for correction. Why not use it to support meteor maintance team if am to pay to have access to the documentation. This is my mind.

Stuff all the docs in there and the blog posts that are accurate, Meteor should keep that vector store up to date - FAIS or similar- that we can down load as as part of a vscode extension right in the IDE so we have all things meteor at the tip of our fingers.

Or something like that…

If we’re going against 3.5 Turbo it’s really not very expensive for this type of thing - I suppose it depends on usage tho

Btw what model are u using for the responses above ?

I wonder if we could somehow integrate this into forums so that it can provide some initial pointers to questions in the help category or at the very least say that we need some additional basic info like Meteor version and so on. That would help a lot.

1 Like

AFAIK you can’t do that from a custom GPT, but you could build an assistant with the same functionality that can be accessed via the API. Custom GPTs are free for the creators (you need ChatGPT+ to access them, so the user Pays for it). Assisstants via API are not.

1 Like

Gpt 4 is accurate than 3.5, but 3.5 is speed.
Its best we have it atched to the forum, which can be easily accesed bythe public. If not, we need to be paying 20$ every month before we can access it.

We need API access for this. We can’t use ChatGPT+ ($20/Month) for this. Biling for APIs is per Token. Hard to say in advance how much the Assistant API would cost (since we don’t have direct control over the input we feed into each LLM call). To get high quality answers, you have to give the LLM enough context - especially if it’s about questions that are pretty Meteor specific. And I think to distill useful knowledge from a forum, RAG is not enough. You will probably have to analyse entire threads to get useful information.

Bumping, as we have some good news :slight_smile:

Right after this was posted, Copilot released a feature that lets paid users create custom GPTs that are then free to use for others, yay!

I tested it for some time right after the feature release, but it wasn’t stable enough, and was hallucinating a lot. But, things got better, and now it might even be a nicer interface than with our initial ChatGPT-based MeteorGPT.

So, I’ve ported all the instructions and knowledge, and published the MeteorGPT in copilot:

Keep in mind - you need to log in, to be able to use the custom GPT, but you don’t need to have a paid account. If you go to this link not being logged in, it just redirects to the main copilot page, so it might seem like it doesn’t work, but it does after you log in.

Today, I tested it with the same prompts I gave to the ChatGPT version, and the results are even better:



What’s cool about it is that it provides sources underneath, so that the user can reference the topic further. The third reference in the last example (about password resetting) actually links to this Meteor forum question:

Also, in general, since we instructed it to reference to the Meteor forum threads, it is capable of linking to specific topics, like here:

Where the first referenced link is the forum thread about it:

Same as with the other version - feel free to use, and let us know how it works for you :slight_smile:

4 Likes

Seems like that’s solved now :slight_smile:

The custom GPTs of ChatGPT are using GPT-4 afaik, since that’s the model used when you have the paid account.

As for the Copilot’s version - I’m not sure, but I wouldn’t be surprised if they used GPT-4 too, even though they are not charging for it.

Copilot is already ‘exposing’ GPT-4 and DALL-E image generation for free to everyone who is logged in even though it is a selling point for ChatGPT. Weird product strategy since both are ‘owned’ by Microsoft, but that’s how it is.

With this simple setup - unfortunately not. It is based fully on the UI of ChatGPT / Copilot.

You could do what @janmp suggested, but yes, it needs a custom implementation and a separate bill to pay towards OpenAI.

But doable :slight_smile: