Contrarian Take Circulating: 'An LLM Will Write You This Algorithm in 10 Seconds — And Not One of Them Could Ever Have Invented It'

X/Twitter·low signalXBlueskyLinkedInCopy link

A post this week uses shape-intersection testing as the wedge for a capability argument: the task is trivially stated and instantly obvious to a human eye, any frontier model will emit a working implementation in seconds, and yet none of them could have originated the algorithm. It is a sharper version of the memorization-versus-invention critique because the chosen example is one where retrieval quality is indistinguishable from understanding at the output layer. Very low reach so far (4 likes, 305 views) and single-source, but it is the cleanest recent framing of the reproduce-versus-invent gap.