Hacker News new | past | comments | ask | show | jobs | submit login

It's a good step up from using pure GPT4.

What do you think how they built it? How do you access content from other pages so quickly?

My guess is that they have crawled a lot of popular developer docs pages (Mozilla, Stackoverflow, Youtube, etc) and created embeddings for all paragraphs on these sites. Then for each search query they use a clever prompt + use the knowledge from the embeddings look up.




Something like that :)


It looks like you are using the bing api to search the web and then somehow integrate the result into the answer. Will do more digging ;)

Cool project and it seems like you and your team have working on it long before the hype began.


Would you need to create a new set of text embeddings ? Do you know if it could be done cheaply ?




Consider applying for YC's Fall 2025 batch! Applications are open till Aug 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: