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
- Stanford Team Introduces Gigatoken, the World's Fastest Tokenizer — StanfordAILab · 2026-07-22
- Tabul AI launches Metal TreeSHAP to speed up Shapley values on Apple silicon — Scobleizer · 2026-07-22
- Reddit points to OpenAI’s ChatGPT Ads page — EcstaticAsparagus509 · 2026-07-22
- Open-source runtime lets each repo define its own AI code reviewer — ibabufrik · 2026-07-22
- DeepSWE: A New Benchmark for Evaluating AI Coding Agents on Real GitHub Issues — pmz · 2026-07-22
- A Rust space-economy sim runs hundreds of autonomous ships, built with Claude — kalcode · 2026-07-22