290913: LLMs As Search Aggregators
LLMs as a search aggregator: your prompt is your query. What's special is that you get true semantic search, you're matching vector embeddings. See also LLMs as solving certain search-based coordination (matching) problems.
Traditional keyword search demands you already speak the indexed community's idiom. LLMs can exchange in and out of diverse denominations: they translate between idiolects.
Traditional keyword search matches you to specific, particular instances. LLMs turn up generic aggregates of many instances.
Traditional keyword search ranks particulars by popularity; LLMs' equivalent of PageRank is volume in training corpus.
Retrieval-Augmented Generation attempts to minimize the weaknesses of LLM search by bolting on a traditional keyword search, resulting in a Frankenstein hybrid. (As well as keeping them up to date with the world post-training.)