Thanks for sharing @multitechvisions
It’s always good see others sharing their insights and knowledge while maintaining a positive and professional attitude.
Thanks for sharing @multitechvisions
Hello @nelson, and thanks for your work here. Seems like you’ve helped a great many people just in this thread alone.
My use needs are very simple. Basic Q&A for 3 categories: company service info, technical support and training data. The training I am looking to do is in the areas of dance and parkour (this data would include video links).
The problem I am running into, like many, many others here, is that fine-tuning is not doing the job. The AI still has a tendency to return it’s own answers rather than those in the datasets – even if the questions match word for word.
The solution (or one of them) to this is embeddings, that much I’ve learned. And, your Google spreadsheet to GPT3 appears to be a very good solution. My problem is that I struggle with understanding what embeddings are (no matter how many times it’s been explained to me). In your Spreadsheet, I get Step 1. But, after that, I don’t know what to put in for these:
- Go to the (Completions Sheet) and fill in the context and prompt for GPT3.
- Fill in the values in the Context Column, in the Prompt Column we use the function getCompletions to build a prompt using either a cell or input value
For example: getCompletions(A2, “What is Open AI all about?”) will get the value of cell A2 and append "What is Open AI all about as the prompt
- If you like to search using your own data and feed the most relevant information into GPT3, you can store your text in the Data Sheet and use the Search sheet to create a dynamic prompt
My mind understands how to format my data for questions and answers, prompts and completions, but I’m a bit lost in this embeddings thing. I don’t even understand the flow here: After the data is embedded, where does it actually go?
I asked in another thread if there was an “Embeddings for Dummies” book I could pick up.
Can you, or anyone, point me to a step by step tutorial which would help me to understand this process and how to utilize it to get the responses I seek?
Thank you for your kind word. The concept of embedding is kind of confusing, but think of it as a language that only computer understands. It will probably be much easier if you see how the spreadsheet work with your examples. If you like, you can send me 1-10 examples of your QA and I can fill them into a spreadsheet for you.
That would be awesome! Thank you! Messaging 10 prompt-completion lines I used to fine-tune to you now. I’m sure seeing examples with my data will help my mind begin to re-envision how to present it.
Oh I didn’t think of connecting NLP and GPT. That’s actually genius. I got some minor experience in DialogFlow, I’ll give it a go.
I also asked ChatGPT for suggestions to NLP platform to use, and it really seems to like BotPress
I used to be a huge fan of ChatFlow, made by Kitt.ai and owned by Chinese Google competitor: Baidu.
However, it seems the service has been shut down. I’ll attach a video how it worked. Maybe someone would make something similar
Thanks for sharing your projects with me the last couple weeks, I don’t think I was able to help everyone, but I hope my time benefited those who I worked with.
Moving forward, I will focus my time on maintaining and improve the Google Spreadsheet GPT3 Tool and the Fine Tune GPT Prompt Web App
I have some interesting or practical ideas, but some are too difficult, etc., share them with everyone.Please assist me in solving it, thanks！
It must be a cool idea:
On the QICQ app, become a chatbot. As long as someone calls it, it answers questions. More importantly, it remembers what everyone in the chat group has said, and can paraphrase or apply it, imitating it.
Nice to meet you
I need help making a shortcut that has the api key for the chat box I’m new to this all
How can I do this?
I am not familiar with QICQ, but have you investigate into what it takes to build on top of this app?
Are you using the OpenAI API? Can you explain a bit more on what you are trying to do?
My name is Reuben, I’m a radiology resident from Melbourne Australia.
I’ve built a “Dictation Error Corrector” web app using text-davinci-003 to address I problem I and many radiologists have.
I am having an issue that I hoped you might be able to help me out with - absolutely no problems at all if you are too busy.
As you can see in the example above, “enostoses” has been changed to “osteochondromas”. This is not correct, and unfortunately continues to recur in my experience using the app.
My prompt is very simple, and I have found that if I specify “permitted words” in the prompt the resulting output is less reliable.
I feel as though this problem could be solved with fine-tuning, however davinci is so much worse than text-davinci-003 at a base level.
Any thoughts on how I might be able to solve this dilemma
This is intersting, I’ll like to see if we can improve this use case.
Can you share the prompt you are using to make the dictation correction?
That will give me a better idea on how we can make improvements.
Thanks so much Nelson! I’ve DM’d you.
@nelson I am trying to build a text generator from an input prompt.
In GPT-4 playground, It is possible to “continue” text generation by simply providing “continue” as additional user prompt if generation stops.
But I could not figure out how to do the same with API using Python:
- I initiate the generation with both system+user prompts
- system start generating text
- It stops at a certain point for long generations
- I got the generation status with “finish_reason”
- I try to feed “continue” prompt with additional API calls ONLY using user prompt=“continue”
- But system simply does not continue to generate
I guess a new API call means starting a new chat, so system loses all its previous context.
I am sure there is a predetermined way to preserve previous conversations/context and continue from where it left off using API/Python, but I simply could not figure it out.
Any insight would be highly appreciated.
Do you mind providing the API Request you used with body so I can take a look at it?
Make sure you don’t send me your API key
I just wrapped out another project, and will have some more time for the next couple weeks.
Feel free to reach out to me if you need any assistance.
hi nelson. thanks for your help in the past. Just for your information, I took a different path at that time.
I made an office extension based on OpenAI GPT that helps creating answers or texts, even from your own and most importantly local database (without Pinecone or Weaviate, etc).
You can download it here:
Very cool, glad it worked out for you.
May I ask how you store your embedding locally using this extention?
We use FAISS with C#. We do the embedding with OpenAI ada.
The binary vector file is stored locally on the hard disk.
Hello Nelson, how are you?
I hope so!!
I’m beginner in AI and Programming too…
I have an idea to develop an interaction between gpt and an erp.
The idea is, when the user clicks on an ad in some market place he can ask questions regarding the product and ad and gpt will be able to answer.
I would like to know how I can do this in a simple and correct way. Could you help me with this?
Is it possible for gpt through its APIs to be able to answer questions to patris of a Url? For example: I pass the address of a url and gpt answers my questions like the content of this url? If so, how can I best do this?