Hi, I’m experimenting with training ChatGPT using information about different converters. You should try asking it something related to that. Since it has my conversation history, it will definitely know the answer. Hopefully this experiment will also be useful for those of you who want to train ChatGPT!
1 Like
GPT and LLM’s are just producing text which has statistically chances to be coherent from what it has absorbed from existing texts during its training (cutoff).
It is ok to produce the skeleton of common/redundant script or source code but it doesn’t understand the meaning of things and the new “knowledge” it takes is temporary, it will soon evaporate.
Better write in wiki/manual what you have learnt than spend your time in training something that will be lost in a few.
Also never trust the results of ChatGPT.
And don’t expect to become a better developer with it either.
1 Like
Yes, you’re right. AI needs knowledge from the wiki. Hopefully, there will be users who are willing to share their knowledge about Synfig and document it on the wiki so that AI can become smarter about Synfig.
In simple words of what @BobSynfig said, what you want to do requires a Artificial Intelligence that actually trains itself continuously with each new given data, and it’s called AGI, Artificial General Intelligence, which doesn’t exists yet. What we are using right now is just Artificial Intelligence, which works on data that has been already feeded before, you can’t train the AI with your chat history.
A better option is to go to more popular sources where the AI might get it’s next source of data from to learn, like Wikipedia, XDA forums and more, and edit those pages, create several blogs about it on popular websites, work on better SEOs.
There is a toggle in Chatgpt that lets the openai model train on your data, but it would still require credible sources from elsewhere to confirm your data is useful, and also if more people are interested in the same topic, the topic will get higher priority.
If you want better results right now, use GROK, it is pretty good because of the Twitter stuff.
Perplexity and ChatGPT aren’t that good to be honest.
2 Likes
Actually, ChatGPT can learn if it’s instructed to remember information we provide—but only for personal use. It can’t be used or shared with others. I taught it about the ‘add’ converter and others, then tested it with a formula, and it was able to generate the correct XML. What do you think?
1 Like
I am saying the same thing as you, by learn I refer to it adding the data you give to it and using it publically for others. That doesn’t happen.
It still has a toggle that basically enables it to learn and add it to its database but it isn’t that helpful for training.
Using competition is better in every way.
TchatGPT was simply the AI that was released to the public first. What I would tell you if you ask me is what this tchatGPT AI is: it’s the AI with the most users.
What is the smartest AI of all? I would say it’s “Claude.” It’s the AI that was initially compared to tchatGPT and others to determine the reliability rate of answers, persistence, reliability of answers, correct answer rate, etc. This was done by the computer science journal 01net.com.
So if you ask me which is the best AI, it’s “Claude.”
And finally.
Generating Images:
What is the best AI for generating “images”? It’s not TchatGPT, contrary to popular belief. Yes, it will create images for you, but you won’t be able to use them very far, only for sharing on social media, etc. (It doesn’t pay any royalties to their creators).
If you’re looking for a real AI, you should use “Adobe Stock,” firstly because it’s a paid image bank (like others), and more than 10 years ago, it already had millions of photos, images, vectors, etc., and it respects copyright and provides user licenses so you can use them anywhere without legal problems. Best of all, it pays its creators for each image sold, with a portion going to the creator via royalties. (If you want to find something better than “Adobe Stock,” type “paid image bank” into a search engine. (Besides, image banks are a virtuous circle.) paid, you pay a monthly subscription, the author of the photo, image or vector will receive on average between 20 and 30% royalties, it’s a virtuous circle.
Training AI with Synfig related data instead of improving Synfig documentation from which it theoretically can learn information from is counter-productive.
I am very sorry for putting it in such a rude way, but instead of training some goddamn unrelated to Synfig AI, it would have been nice if someone would step up to improve official documentation so it doesn’t suck ass. Speaking of converters, here is the official page for it: Converters — Synfig User Manual 1.5.1 documentation
There’s nothing. And I, along with others I am sure, would prefer a proper documentation over using AI to get Synfig related information. At least transfer information from old wiki: Convert - Synfig Animation Studio
So is anyone here wants to improve official documentation? Anyone?
2 Likes
And not even a link to the old Wiki with a “To Do” …
With the time passing, only the “Ancients” still know that Synfig is vectorial first and what is a Synfig unit …
I already wrote plenty of time that instead of create new (incomplete) features, we should focus on stabilisation and documentation of the existing first …
1 Like
Yeah, completely agree with you. I allowed myself to get angry, but it doesn’t solve anything…
I am currently working in a new place, so my free time is quite limited for now. When things get stabilized, I promise to try and transfer documentation from old wiki to readthedocs, so we can finally have everything in one place, close the old wiki and concentrate on improving existing documentation. The task is not hard, just tedious.
When the new version of the wiki arrived, I tried to understand how to create a paragraph, add an image link, etc., but I didn’t understand the new system.
It seemed too different from the old system.
What we need is for someone who understands how to do it to provide us with links somewhere, on the forum or elsewhere, to learn how to do it. What tutorial link, whether video or web page, explains how to create or edit a page in the new version of the wiki?
