On: Large Language Models are not a search engine

[hprotagonist]

“…the outcomes of Large Language Models are not designed to be true — they are merely designed to be statistically likely.”

yep!

[pcmaffey → hprotagonist]

A search engine does not return a true answer. It returns a set of possibilities ranked by likelihood.

An LLM is just an aggregation of search results.

[roncesvalles → pcmaffey]

No, that misses the nuance entirely. An LLM returns one answer that it thinks is likely to be true. A search engine returns many answers one of which is either certainly true or (depending on what you're searching) is the literal source of truth.

There is a world of difference between these two results.

Ultimately the problem is that a single repository of all verifiably true human knowledge just doesn't exist and both search engines and LLMs are copes to deal with this fact.

[hprotagonist → pcmaffey]

a set of possibilities ranked by likelihood… written by humans who have the capacity to be interested in truth.