The Illustrated Retrieval Transformer

[@raghavendra2096]

Thanks a lot for the explanation, was searching for this video as your blog had this topic long time back.

[@Skinishh]

Can the retrieval be differentiable?

[@faheemthanveer1493]

Very interesting, I want to work on this model along with a VIT similar to CLIP. What are your thoughts on this?

[@nintishia]

Thanks for the explanation. I think the knowledge source, in addition to web or database search, could just be software designed for a specific application.

[@arp_ai]

Yes exactly. That is now more broadly called "tool use", where the tool can be a calculator, a python shell, a search engine, or other tools the model may have access to.

[@nintishia]

@@arp_ai Thanks for the information. Can you please provide some references?

[@Danielle-s5q]

@@arp_ai This is really powerful as we could use graphs for derived knowledge. Even something like a symbolic system. This looks like we can finally bridge the ML symbolic computing divide.

[@RashidAlipage]

I read your various blog posts to better grasp the different AI models. Please provide a list of the tools and resources that you use to illustrate and animate AI models. Thank you very much.

[@malikrumi1206]

I don't know many people with neural databases lying around. What about Postgresql? And how would you make the connection?

[@arp_ai]

FAISS, Weaviate, are open source options. Postgres can do vector search with pgvector, but I don't believe it's widely used for this task.

[@FelipeOliveira-gt9bf]

Hey Jay! what is the tool/software are you using to create the illustrations? I would like to use to my master's :)

[@alankarshukla4385]

Great Video!

[@botfactory1510]

Thanks for the explanation