'Krapivin was not held back by the conventional wisdom for the simple reason that he was unaware of it. “I did this without knowing about Yao’s conjecture"'

https://www.quantamagazine.org/undergraduate-upends-a-40-year-old-data-science-conjecture-20250210/

Many such cases of progress being stifled by the chained elephant.

BDFL of Quilibrium, Eng @ Farcaster, ex-Coinbase, always opinionated, never hydrated

/quilibrium

I wonder if you could get an LLM to figure this out if you prodded it enough, but without giving it the answer?

Decentralized open source consciousness | Values-aligned communities | citizen of Black Rock City

Wdym by the prompting is the insight?

If it could come up with us by saying stuff like if you ignore the existing conjectures about X, could you do this thing faster or better or cheaper? What if you thought about XYZ?

Basically, this is an example of how a novel approach that’s not in the LLM’s data set yet, however, there is a logical path to get to the answer.

And yes, LLMs have been bad at this stuff, historically, but my point is trying to figure out if there’s a way we can get it to happen upon stuff that we don’t know yet

LLMs are quite bad at giving anything novel, the prompting _is_ the insight