🚀 How I Made ChatGPT an Online Search Engine with Custom Data!🌐
ChatGPT has its limitations:
It's a frozen model with a knowledge cutoff in 2021.
Its context size is limited to 4096 tokens, which makes it challenging to parse extensive knowledge into the context for conversation.
But there's a clever workaround:
-
Prompt ChatGPT to generate search queries
-
Parse these formatted queries and run them with an external search engine (use an API or build your own vector search engine).
-
Retrieve search results and add the queries to the ChatGPT context.
Voila - you have an up-to-date, online ChatGPT!
I implemented this solution on my own e-commerce search engine, www.vexa.ai, and created a Telegram bot https://t.me/vexa_ai_bot.
As a result, I've built an AI salesperson with capabilities that go way beyond raw ChatGPT!
Example:
I initially prompted my ChatGPT with:
-
"You are a helpful salesperson in a Farfetch shop. I am a buyer. On the way to helping me, you ask questions and give answers. You create search queries for the in-site search engine. Search queries must be wrapped in like https://farfetch.vexa.ai/?q=the+thing+user+search. The buyer will not see the query, and the idea of a query is hidden from the buyer; instead, they will see the search results. Create new search queries with every message. You can create multiple queries with each message to give a choice. All recommendations must be wrapped into search queries."
-
Then, ChatGPT greets the user and asks if they want something.
-
The user says they want to buy a gift for their mother. ChatGPT generates a response that contains not only text but also a link: https://farfetch.vexa.ai/?q=women+scarves+cozy+spring
-
The link represents search results with available online items to buy.
Give it a try and let the innovation thrive! 💪
go check the bot: https://t.me/vexa_ai_bot
write me in telegram: https://t.me/dmitry_grankin
#ChatGPT #Limitations #Workaround #Innovation #AI #ECommerce #SearchIntegration #TelegramBot
