Scott Farrell
Forum Replies Created
-
sounds great
take a look at the json going to chatGPT – it’s missing from the chatGPT history
I’ll do a video of it
the system message is fine, whether you change that is up to you.
BUT: what I am saying is that you do NOT send the very first thing the user types in the chat history send to chatGPT
In my example above, the user says:msg1
msg2
msg3
but the chat history doesnt show msg1
I cut and paste my own entries.
Product:
– ID: 587
– Name: Multi Colour Cherry Shrimp pack – 10 of each Blue Yellow Green
– Category: shrimp
– Description: includes free shipping Add a splash of colour to your tank with 3 colours of shrimp. Easy to care for , keeps your tank clean.now look how it answers:
- You:do you sell red shrimp?
- Shrimply:Yes, we do sell red shrimp. We have a few options available on our website, including the Red Cherry Shrimp, Red Cherry Shrimp 12 pack, and Red Cherry Shrimp Sekura Top Grade.
Do you not think this is not needed ?
All your ‘index builder’ entries definately need meta data.
Just the description – doesn’t really say what’s going on .
plus – you just run all the descriptions together – as one big string without delimiters.
The idea is to give MORE context to allow MORE understanding from chatgpt.
yeh – the problem is after you get the embedding search result from pine cone, you add the orginal text into your current prompt, that has a lesser token limit (4096 tokens or 16000 characters ) .
So Senol’s answer is wrong. You really need to stick to around 12000 character limit. And if it chooses to bring back more than 1 big embedding result – same thing will occur.
Senol probably needs to limit the size of what is added to the prompt from the embedding search, plus a hard limit on the embeddings of around that 12000 characters.that’s me hacking a general prompt
it’s standard woocommerce meta data I am talking about:
– product name
– price
– product category
– and some basic delimiting between returned search entries – so it doesn’t look like 1 big entryyeh – that one has a big description
but try ‘do you sell red shrimp’ – it can’t see my red shrimp products, like I said the description doesn’t say red shrimp. it then talks about red shrimp that i do NOT have. it should say we have a 12 pack, and multi colour packs with red.
I’ll code the metadata stuff myself – it’s super important mate
it’s on my shrimply.shop website, I emailed you a login a few weeks back
look up his stuff on content builder, embeddings. you cut and paste the text of what you want to teach it
make sure you have both an API key for openai saved, and pinecone api and database url
mucked around, and eventually it showed some error. now I re-index and it’s working
yeppa – updated, ‘delete all’
good news- everything disappeared
bad news – nothing came back
checked cron, manually ran some cron from command line with php
you must be getting annoyed at me
https://pasteboard.co/VbB0YwTBXg34.png
upgraded to Version 1.5.42
I cant even work out where you store everything ?
I look for wpaicg in tables – dropped them
looked for %wpaicg% in wp_options – can’t see embedding/index data.
dropped and re-created pinecone.
reinstalled plugin.
Where is the index data status stored ? so i can delete it and start again ?I appreciate all your help.
I upgraded, waited, nothing different.
This is my screen
https://pasteboard.co/LrmaEBeWHASH.png
