I have searched the forum but nothing relevant for me. I also skimmed the fine-tuning page.
I recently started working in IT in an institution, but not as a software developer, but as a planner and facilitator. We offer trainings, seminars etc., man of them hybrid via Moodle.
We now have a project with Moodle in which we would like to integrate ChatGPT (preferably GPT4, or would 3.5 be enough?). If the learners are stuck on a question, for example, then they could ask ChatGPT and it would help.
Is it possible to integrate ChatGPT into Moodle fluently? What do you need to consider?
ChatGPT is pre-trained as you know, but we then need to train ChatGPT specifically in the subject area of the course. Is that possible? How would that need to be done specifically? I think we would just have to train ChatGPT specifically on hundreds/thousands of texts from that topic area. As far as I know, this is done with Tensorflow or PyTorch.
We have two service providers, one is currently working on Dialogflow from Google. But since that is not always optimal, because it is rigid and the dialog tree is predefined, we would like to integrate ChatGPT as well. ChatGPT is more flexible.
I read something from about the block_openai_chat. Is that usable? Is this only available in version 3.5?
I have read that topic, but there was not what I searched for.
I am still reading the documentation on OpenAI.
I would like to use ChatGPT (with GPT-4) directly rather than using a GPT-4 model, since ChatGPT is fine-tuned already originating to the Chat Completions API. The GPT-4 API will also be available for new developers soon.
Again, there is the Moodle plugin OpenAI Chat Block. I haven’t found another one to use that is provided within the Moodle repos.
I want to use ChatGPT within a specific course and I want to train it specifically on that course and so it gives accurate answers. If I cannot do this on my own (I am not a pro), I could deliver this work.
How to accomplish? The above mentioned plugin has the option of adapting the prompt format with the Source of Truth etc. With this, I had to train Question-Answer-pairs, that is very time consuming.
1.2) Isn’t there any other possibility to train ChatGPT afterwards, fine-tuning is not yet available for GPT-4 and GPT-3.5 Turbo.
If I integrate this im my course, is one API key is enough for, let’s say 10 students?
2.1) How would this effect the history of the chat?
Very good resources. Am still in it.
I have read about the Supervised Learning vs the Prompt-Based AI (screenshot). I don’t quite get it. Is the Prompt-Based AI now part of fine-tuning or are these different things? Is Prompt-Based AI what OpenAI Chat Block tells with Source of Truth and Prompt Format (screenshot)?
To be clear, you’re wanting your students to log into chat.openai.com ?
ChatGPT is the name for the… instant-messenger or chatroom-like experience you get on their website. It refers to a specific product, downstream and distinct from the models/completions service. Is it more like, your students are logging into your schools website, which has some moodle in the backend, which will itself be making calls to OpenAI through “OpenAI Chat Block” plugin located on your server?
My students will log into our own Moodle platform, there is a block plugin for ChatGPT. I thought about using directly ChatGPT because it is already optimized for chat (via FIne-Tuning GPT-4 I guess) except it is not trained on the course topic. That is what I want to train specifically. How to train this last step best is my question. Fine-Tuning again or usage of the OpenAI Chat Block plugin’s  functionality ( Source of Truth and Prompt Format).
Not sure if you found your answer or not. I wrote a paper on how to integrate ChatGPT into Moodle environment, and listed eight use cases. You can give it a read. Try to google “ChatGPT and Moodle Walk into a Bar: A Demonstration of AI’s Mind-blowing Impact on E-Learning”
Basically, you can use “OpenAI Chat Block” plugin and it is based on ChatGPT 3.5-turbo.
As for using ChatGPT in combination with other tools, I don’t know much because just ChatGPT for me is the most knowledge-generating thing on its own. You don’t need a complex interface, you just need to talk, because the science of creating knowledge through interactive content is the most effective way.
Since the past, it is one of the current Active Learning methods, promoting a variety of soft skills. Conversation methods have different effects, such as Morphological Analysis, which uses the separation of parts of the story to talk, or Agtumentation. It is a response that requires finding a solution to the issue that we are arguing about.
This method is quite effective with AI because AI does not have human biases. AI’s biases can be broken if the knowledge we are using is correct and conflicts exist within it. But humans, even though they are wrong, don’t accept it and find other things to argue about endlessly.
You can upload the documents you want to use as your main knowledge.
You can set behavioral responses in the instruction to reduce unnecessary words and compliments. and disagree with our ideas, giving more to the argument Can be used in many fields But be careful about the limitations of content such as law, medicine, psychology, but they can be avoided or reduced through customization. and knowledge documents. But history is worrisome because GPT doesn’t understand much about how to study it. Such as the importance of the order of content or letters in important documents. Their ordering reveals more details than just the meaning of the characters.