3 ms·
This looks like it's just throwing your Search query directly into ChatGPT and sending you back the result. (Source code[0]) This is less intelligent than what
by freeqaz 4y ago
This looks like it's just throwing your Search query directly into ChatGPT and sending you back the result. (Source code[0])
This is less intelligent than what Bing is doing with (presumably) scraping the pages and then generating embeddings to add "context" for your query.
There isn't a "generic" project that I've seen that's doing this yet, but I could see somebody extending this plugin to scrape the first ~5 pages and then feed that into ChatGPT as the "context" for answering the question.
We've been doing something similar with another Open Source project called Buff[1] but it requires scraping the data yourself before you can ask it questions. (We're scraping vulnerability/CVE data and putting it into the Vector database via OpenAI's embeddings API. You can play around with it on our Discord server linked from my profile.)
Is anybody else aware of any other projects to make search more useful?
0: https://github.com/adamlui/userscripts/blob/master/chatgpt/duckduckgpt/duckduckgpt.user.js#L216 https://github.com/adamlui/userscripts/blob/master/chatgpt/d...
1: https://github.com/getbuff/Buff/ https://github.com/getbuff/Buff/
- fzliu 4y agoGood stuff! Just out of curiosity, what vector database are you using.
- freeqaz 4y agoPineCone is the database that Buff uses under the hood.
- tuukkah 4y agoWebChatGPT does almost what you suggest. However, it does not actually scrape the result pages but instead uses the snippets on the DDG result page as the context to ChatGPT. https://github.com/qunash/chatgpt-advanced https://github.com/qunash/chatgpt-advanced