AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

On the Limits of LLM Adaptability: Impact of Model-Internalized Priors on Annotation Task Performance

Etienne Casanova, Rafal Kocielnik, R. Michael Alvarez

ICML 2026 · Large language models › Language, safety and interpretability · approach: Statistical frameworks for LLM evaluation · 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
add the aligned task definition to the prompt9 LLMs · Toxicity annotation (5 datasets)accuracy (%): 80.3 -> 82.0Improved
add few-shot examples to the prompt9 LLMs · Toxicity annotation (5 datasets)accuracy (%): 80.3 -> 81.5Improved
add the definition to few-shot prompting9 LLMs · Toxicity annotation (5 datasets)accuracy (%): 81.5 -> 81.6No clear effect
apply DSPy automated prompt optimization9 LLMs · Toxicity annotation (5 datasets)accuracy (%): 80.3 -> 80.2No clear effect
apply DSPy optimizationLlama-3.1-8B · Toxicity annotationaccuracy (%): 76.7 -> 72.3Hurt
apply DSPy optimizationLlama-3.3-70B · Toxicity annotationaccuracy (%): 80.5 -> 77.5Hurt
use misaligned definitions instead of the aligned definition9 LLMs · Toxicity annotation (5 datasets)accuracy (%): 82.0 -> 80.0Hurt
swap aligned definition for Gaming toxicity definition9 LLMs · Toxicity annotationaccuracy (%): 82.0 -> 76.4Hurt
swap aligned definition for Fox News hate speech definition9 LLMs · Toxicity annotationaccuracy (%): 82.0 -> 82.6Improved
increase zero-shot confidence by one SD9 LLMs · Toxicity annotation, error correctionodds ratio of rescue: OR=0.84Hurt
add definition-based prompting (high-confidence zero-shot errors vs moderate)9 LLMs · Toxicity annotation, rescue by confidencerescue probability (%): 51.8% -> 20.8%Hurt
use larger model (Llama-3.1-70B instead of Mistral-Small-24B)LLM annotators · Toxicity annotationrescue rate (%): 44.2% -> 31.3%Hurt

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