The Promise of Language Models for Search: Generative Information Retrieval
Neural Search Talks — Zeta Alpha
00:00
The Future of Information Retrieval
Rome and RAG, they're basically doing retrieval from social media. With BingChat or chat GBT, what you can do is you can issue a query to Bing or some search engine and then just copy paste whatever output or whatever's returned back into the context that it can then use to answer the question. So this is a very different flavor of augmentation because there's no pre-training happening. But somehow we can't really force the language models or I was playing around with GPT-4 just an hour ago with my colleague and he said, look, you can't even make it describe the beats in a song.
Transcript
Play full episode
