Sometimes demonstrations lead people to using whatever it is you put together regardless if it requires 2-3 steps or how easy it is to follow in your head. For example, I had no idea what ms office was capable of however by chance I dedicated my 30sec -1min to watch from the video feed on Facebook. Thanks to that short video, it has opened a new way for me to animate and use what saw in conjunction with Hype.
Creating documentation, examples, and tutorials is indeed a side effort for me. However, I value these resources not just as a developer, but also from a user’s point of view. I’ve dabbled in creating a few YouTube videos, but the engagement has been minimal so far. Nonetheless, I plan to focus on expanding the written documentation and may revisit video content in the future.
I wonder how well ai would do at documenting.
I started using it to comment my completed code, it does well with a prompt to fully comment.
Not used it much to write code as I found it easier and quicker to do myself ( if you have tried it you will know what I mean)
Absolutely, Mark. The advancements from GPT-3.5 to GPT-4 are nothing short of impressive, especially when it comes to documentation recall. With GPT-4, it's like having a far more efficient assistant that has memorized the Hype documentation to a remarkable extent. This is a game-changer for the speed and accuracy at which we can generate examples, tutorials, or even troubleshoot issues.
Of course, the technology isn't perfect, and there are situations where the human intuition and experience can't be replaced. However, the tool serves as an excellent complement, reducing the time spent on looking up syntax or methodologies. It makes the creative process more fluid, allowing us to focus on our Hype project ideas rather than the nitty-gritty details. I'm eager to see how this evolves and integrates further into the development process.
The concept of fine-tuning GPT-3.5 specifically for Hype documentation is particularly compelling. Given that GPT-3.5's inference is relatively inexpensive, the upfront cost of fine-tuning could be a justified investment. Creating a specialized version—perhaps dubbed HypeGPT—could improve how we access and utilize documentation (or create functions in Hype). This could serve as an invaluable resource if integrated into Hype (although privacy and cost are an issue)…
We just need to build a fine-tune data set now for a web based version or one that hooks into the „edit with“ workflow.
Using copy and paste has the lowest friction.
I just tried to get it to document an extension I am working on.
And agree with what you say.
Ip prompted it with a long description of what I have, its expected usage and the intent of the extension.
I also gave it all the Extension code and some examples constructors for the API.
I the prompted
It's initial attempt was not bad and I could see this was going to save me a lot of time.
But it did have mistakes in its understanding. Correcting this mistakes after the initial try where the hardest part. The AI seems to forget stuff along the way and you have to remind it explicitly. It actually hard not to get frustrated and be sarcastic. Not sure it would get that any way.
Then there is you own corrections when you realise you can change something in the code. That just as frustrating. trying to get it to contain that change within the whole document context.
It struggles with context and I think its does not continually keep an eye on the big picture. It seems to at some point treat a prompt as an individual question rather than take in to account what has gone before.
It also has an habit of changing up a format. For example it would give one example in the documentation with a set of headings but later on it if I asked for it to give me another example but use a different scenario, it would do so with completely different headings. Even the wording would be miles apart.
I found I had to take some of its answers from one prompt or other and past them into my own text doc to keep track of what was right. Otherwise you would come so lost with all the correct stuff mixed withe stupid stuff in the prompt answers.
But over all , It did take half the time I would have spent to write out the Documentation by myself.
This included me then going over it all and correcting some of the ways it changed the structure and style of the sections.
Also going over the text to make sure it said what it should and changing things where needed.
What I found was it allowed me to keep my train of thought without mostly losing track of what I want.
Which is what would have happened if I had just done it all myself. So I have less of a headache..
Was it worth using it yes. Would I use it again with all I'd issues yes.
Was it easy, No. But it was easier.
Did you use ChatGPT GPT4? It is literally years more advanced than much older version 3.x…
Used 3.5 I think.
In that case you are in for a treat. Repeat your attempt with the newer model. I had tremendous better results with that model.
Hah, It appears to be only for Plus users? not sure what they are..
Probably, will look to join up to it soon. Got to also buy Typora now it seems. Which is def worth it.
You don’t necessarily need Typora. It’s nice to have though. There are plenty of other markdown editors for free.
Generous free tier:
I added some links
I have an editor with these features:
- Markdown editing with live previewing
- Versioned drafts with a side-by-side diff UI
- AI editing (though is only chatgpt-3.5 right now)
It is in early beta. Please DM me if you're interested in getting an invite .