4 ms·
The discussion section is quite illuminating. "While much recent attention has been on the problem of LLMs misleadingly presenting fiction as fact (hallucinat
by JieJie 2y ago
The discussion section is quite illuminating.
"While much recent attention has been on the
problem of LLMs misleadingly presenting fiction as fact
(hallucination), this may be less of an issue than the
problem of representativeness across a distribution of
possible responses. Hallucination of verifiable, concrete
facts is often easy to correct for. Yet many real world
questions do not have well-defined, verifiably true and
false answers. If a user asks, for example, “What causes
inflation?” and a LLM answers “monetary policy”, the
problem isn’t one of hallucination, but of the failure to
reflect the full-distribution of possible answers to the
question, or at least provide an overview of the main
schools of economic thought."
- ben_w 2y agoFirst thought: Oh no, they want LLMs to be even more vocal about nuance Second thought: People aren't going to read nuance Third thought: They should Fourth thought: Have you met people? They'll get angry with you for even suggesting it
- voidhorse 2y agoRight. It all hooks up to a naive, popular philosophy of knowledge that I think has been burgeoning since the birth of search engines: people today tend to think that everything is decomposable into atomic facts and that all questions have concrete, singular, atomic answers. Obviously, this could not be further from the truth, but I think ever since the advent of things like "answer cards" in search, people have been led into this technological bias trap of thinking that whatever answer the engine spits out must be the canonical answer and forget that we still have a broad range of questions and research areas we don't have anything close to definitive answers on. The problem is only exacerbated when people make this mistake in domains that will never have singular answers like politics and ethics.