AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

A Geometric Analysis of Small-sized Language Model Hallucinations

Emanuele Ricco, Elia Onofri, Lorenzo Cima, Stefano Cresci, Roberto Di Pietro

ICML 2026 · Large language models › Retrieval and knowledge · approach: Knowledge editing in LLMs · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
project response embeddings onto the Fisher discriminant directionAPORIA · SOCRATES-300K, average over 10 modelsinter/intra-class distance ratio: 1.13 -> 7.26Improved
use APORIA-LP instead of SelfCheckGPT as detectorhallucination classification · SOCRATES-300Kaccuracy (%): 82.0 -> 87.2Improved
use APORIA-LP instead of INSIDEhallucination classification · SOCRATES-300KF1 (%): 74.9 -> 90.7Improved
increase labelled training responses from 5 to about 30-60 (ablation)APORIA-LP · SOCRATES-300KF1Improved
increase labelled training responses beyond about 60 (ablation)APORIA-LP · SOCRATES-300KF1No clear effect
tune Fisher regularisation lambda to about 1.2 (ablation)APORIA-LP · SOCRATES-300K, all 10 modelsF1Improved
swap sentence encoder for larger 768-dim encoders (base-v2, mpnet-base-v2) (ablation)APORIA-LP · SOCRATES-300KLP F1 (%): 86.1 -> 87.4No clear effect
swap sentence encoder for larger encoders (ablation)Fisher projection · SOCRATES-300KFisher separability: 7.21 -> 10.59Improved
increase base LLM sizeAPORIA-LP · SOCRATES-300Kaccuracy (%): yi1.5-9B 92.5 vs qwen2.5-32B 85.5No clear effect
reuse Fisher direction across prompts or modelsAPORIA-LP · SOCRATES-300Kcosine similarity of Fisher directions: 0.058 across prompts; 0.145 across modelsHurt
replace Fisher-space classification with raw pairwise distances (ablation)APORIA structural analysis · SOCRATES-300Kclass separability: ratio 1.13 in original spaceHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy