NLA Keeps Accuracy but Fabricates More
Turn_Trout · x · 2026-07-13
Experimental results show that an NLA initialized with confabulation has almost the same reconstruction accuracy as the control group, but its outputs contain a massive amount of fabricated content, reaching a high of 99.3%.
More notably, the training process does reduce its fabrication rate slightly, but not by a wide enough margin; it ultimately remains more prone to fabrication than the control group. Conversely, the control group actually learns to fabricate more after training.
More from Research
- Mathematician Daniel Litt Launches Problem Repo to Track Human vs AI Progress: 15 Problems, 1 Solved — littmath · 2026-09-11
- Open ECDSA.fail challenge uses AI agents to shrink Shor's-algorithm quantum circuits for Bitcoin keys — StefanoGogioso · 2026-09-11
- Alex Townsend posts 200 open problems in numerical linear algebra for humans and AI agents — IgorCarron · 2026-09-11
- Navier-Stokes, Riemann, P vs NP: what this week's math buzzwords mean for you — koltregaskes · 2026-09-11
- Fruit fly brain as an LLM: connectome-driven language model demo goes live — ngxson · 2026-09-11
- Harry Collins: LLMs can't do frontier science because they can't invent new language — whoamisri · 2026-09-11