PHANTASM: The first ML framework to invert LLM hallucination, confabulation, and epistemic miscalibration into productive features. Three pillars: HGT + CMN + UC.
Companion artefact to the MODELS 2026 (NIER) paper. Does epistemic metadata change how AI consumers answer derivation questions over an MBSE model? 3 LLMs x 120 judged answers over the public Apollo 11 SysML v2 model. doi:10.1145/3822455.3838783
Context integrity for LLM chat sessions: deterministic fact capture, confabulation + sycophancy guards, and the gauntlet that scores them. Zero-dep, bring your own model.
RICE-P1: Rejection-Induced Confabulation Escalation — Pre-registered LLM confabulation study. OSF: https://osf.io/rcfby/ | Pre-reg: https://osf.io/3es5f/
Predict LLM factual recall errors via scaling laws
a pilot test of nonsense words, assisted by claude
To associate your repository with the confabulation topic, visit your repo's landing page and select "manage topics."