9h ago

X debate challenges view of LLMs as databases that merely rearrange text, citing frontier models' near-perfect SAT and GRE scores plus PhD-level research outputs as evidence of distinct capabilities

One response frames LLMs as neural networks sampling from learned distributions.

0
Original post

“The words already existed. All the AI did was put them together in new ways.”

6:52 PM · May 21, 2026 View on X